Here, we're going from v56 to v108 - a huge jump. There have been so many additions and so many fixes. This version adds a lot of adult content in addition to so much more. I recommend checking the model changelog (very bottom of this document). In this version, the Showcase prompt files (and sample prompt files) contain adult material reflecting what was added to this version.

Because it has grown so much, knowing there are versions with files sorted by folder as well as the one-folder-for-everything model is important - pick the one that works for you. Both are included in the zip.

Every kind of adult, from every corner of the world, dressed in anything from a 5,000-year-old tunic to a plush hot-dog costume.

Guyanonymous' Wildcards gives your text-to-image prompts the whole world to draw from: more than 41,600 drop-in .txt files and 3.6 million lines covering adult characters of every age, body, ancestry and gender identity, plus garments and costumes from the Palaeolithic to today, accessories, footwear, makeup, tattoos, weapons, armour, surfboards and the places they stand in. *It's a big collection. I use https://civarchive.com/models/3016893/guyanonymous-wildcard-prompt-studio for designing my prompts and with this large collection, it helps! Also lets you deal with wildcard extensions that don't handle nested wildcard prompts.

Most wildcard packs repeat the same faces, clothes and colours. Guyanonymous' Wildcards is built to do the opposite. It is a folder of plain .txt files that you drop into any wildcard-aware image tool and call with a name wrapped in double underscores. Build adult characters aged 18 to 120 with weighted bodies, hair, skin and clothing, or let one prompt pick a person in proportion to the world's adult population, by region or by country. Dress them from garment families that run from ballet to biker, flapper to festival, swimwear to formal wear, corsets to onesies; uniforms and costumes; 37 eras of dress; more than 35 accessory families; a full footwear system; and costumes for fursuiters, cosplayers, LARP and drag. Control sunburn and tan against each person's natural skin tone, dress them for any season or weather, add makeup, nails, piercings and body art, and set the scene in buildings, rooms and landscapes from every region. Everything is for making adult characters only, original wording, and written to embrace global diversity.

What's new since V107

Version 108 (this release)

Level 6 same-sex panels: every adult_l6_panel_mm* and adult_l6_panel_ww* line now has a level-6 ending (semen for men, clear fluid for women) that fits the pose, so level 6 differs from level 5.
Level 1: the bent-forward-at-a-table pose is removed from adult_l1_panel_wm (it stays at level 2 and above).
Locations: level 1 and level 3 scenes at wet places (shower, steam room, sauna, plunge pool, hot tub, lake dock) now dress people in swimwear.
Panels and groups: hidden-body wording removed where the pose covers it, 69 geometry tidied, impossible floor position replaced, three-person nude sentences matched to who is in the scene, and duplicate group lines removed.
Fluid wording: 'not white' negations replaced with 'clear and transparent' in adult_l6_location, adult_l6_panel_wm and fluid_scene_squirt. *squirting and ejaculation doesn't always work, but can produce some laughter.
Activities: all 51 act_*_scene_prompt files start each sentence with a capital ('The scene includes ...'); act_*_frame_prompt uses the new act_frame_style (six on-location photo styles) so a studio backdrop no longer clashes with the place.
Body features: f_body_nipples reworded per breast, head_chin garbled and 'cleft' lines fixed, new body_navel (10 navel shapes).
Showcase files (40 prompts at 1024 and 1408 pixel counts, steps 50, cfg 3.75) and 100 sample prompts are included.

Showcase and sample files: the showcase files highlight what is new or changed in this version, while the sample files are representative of everything the model can do.

Features at a glance

People and bodies
- Adults aged 18 to 120 in nine age brackets. A person can be picked by world population share, region or country.
- Matching women’s and men’s file sets, plus cisgender, trans, non-binary, androgynous and intersex identity layers
- Faces, eyes, noses, lips, chins, hair textures and styles, facial hair and body hair, all described physically
- Bodies of every build, with realistic detail: cellulite, stretch marks, wrinkles, veins and moles
- Expressions for 68 emotions, each at three strengths
- Makeup, nails, piercings, tattoos, scars, tan and sunburn matched to skin tone, and clinical wounds with casts and bandages
Clothing and accessories
- About 560 garment families by type, era and subculture
- 37 periods of dress, from the Palaeolithic to the 2010s, and traditional dress from 30 world regions
- Uniforms, ceremonial dress, cosplay, fursuits, LARP, drag and Halloween costumes
- About 430 shoe types, more than 50 accessory families, and outerwear, hats, bags, scarves, gloves and belts built from parts
Places, scenes and framing
- Buildings, rooms, 24 kinds of public place, 28 landscape types and 8 plant biomes
- 20 weather conditions, extreme weather events, seasons, times of day and visual effects
- Framing, camera, lighting, depth and multi-panel grid helpers
Activities and roles
- 120 professions across 7 sectors
- Hobbies, music, sport, yoga, dance and martial arts
- Poses for one, two and three people
Worlds and genres
- 31 fantasy, sci-fi and historical worlds
- Cyberpunk, steampunk, western, noir, pirate, spy and horror families, each with classes, places, items and vehicles
Objects and collections
- 157 vehicle genres, with paint, damage and interiors; robots and gadgets
- Toys and die-cast models, books, comics, signage and packaging
- Money, metalware, treasure, cards, minerals, sound equipment, screens and home décor
- A condition system that describes any object, from pristine to ruined
Nature and creatures
- Animal families with behaviour and young, farm animals, pets, legendary creatures and hybrids
Styles and design
- Art media, sculpture, craft, textile and rendered art forms, plus illustration lineages and photographic eras and processes
- 104 manufacturer styles by decade, and over 400 brand-inspired design styles that describe colour, shape and material without naming the brand
- A fake name system that puts quoted invented text on logos, signs, business cards, caps, t-shirts and covers
Adult content (adults only, consensual)
- Levels L1 to L6, from suggestive to explicit
- Solo, couple and group scenes, locations, positions, kink, gear, wrestling and body-feature detail

Naming guide

Version 36 unified the file names. A name now reads family, group, variant:

f_ / m_: female and male person files. Files with no prefix are shared.
age##_##_: one layer per age bracket. bt_: body-type layers.
clothing_garment_, clothing_uniform_, clothing_onesie_, clothing_period_: all clothing, under one prefix.
acc_<family>_: accessories, with footwear as acc_shoe_. Tiers are basic, standard, fine, luxury, showpiece and practical.
costume_: Halloween and character costume. obj_: objects, weapons, armour and surfboards.
body_, skin_tan_, tattoo_: makeup, nails, piercings, tan and body art.
scene_, pose_, wx_: places, poses and weather outfits. world_ and countries_: people by region and country.
aoo_ in front of any name: the All Options Once edition.

If you used earlier versions, filename_changes.txt lists every old name and its new name. Some entry points changed: __garment_prompt__ is now __clothing_garment_prompt__, and __period_prompt__ is now __clothing_period_prompt__.

Getting started

Unzip and copy. Unzip Guyanonymous_Wildcards_v108.zip and copy the wildcards folder into your image tool's wildcard folder. Python is not needed to use it.
Call a file. Put the file name between double underscores in a prompt, for example __f_prompt_master_by_age__ for a woman or __m_prompt_master_by_age__ for a man.
Try a layer. Use __world_person__ for a random person in world proportions, __acc_prompt__ for an accessory, __clothing_garment_prompt__ for a garment, __clothing_period_prompt__ for period dress, __wx_outfit_prompt__ for a weather outfit, or __skin_tan_level__ for a tan or sunburn.
Mix and match. Combine them in one prompt, for example a person, then a garment, then an accessory. A blank line in any file means "none". __v38_entry_points__ lists a starting file for each new section.

Optional: run python3 tools/expand.py f_prompt_master_by_age to preview sample prompts, and python3 tools/build_all.py to rebuild everything (about 10 to 15 minutes). The All Options Once edition uses the same names with an aoo_ prefix.

Warning: files that point to other files

Many prompt files in this package are templates. Their lines contain references to other wildcard files, and those files can contain further references. A prompt can therefore be several layers deep, for example __world_person__ calls a region file, which calls a country file, which calls the age, body and clothing files.

This can cause problems. If your image tool does not resolve nested wildcards fully, or stops at a fixed depth, some references may be left in the final prompt as raw text such as __f_age25_34_prompt_char_face__ instead of being replaced. Check a few generated prompts before a big batch, and make sure your tool's wildcard depth or recursion setting is on.

To manage complex, layered prompts, the creator of this package uses Offline Wildcard Prompt Studio, which works with any prompting system. It is a handy way to build and check nested prompts before you run them. The package also includes python3 tools/expand.py to preview prompts and python3 tools/expand.py --check to confirm every reference points to a real file.

Short post. Guyanonymous' Wildcards v108 is out: 12,320 wildcard files for text-to-image, built for global diversity. New: plain-named character prompts such as f_create_random_woman and any_create_random_person, with clothing kept separate, full and short versions, and three ways to treat age. Also earlier: a clinical injury overlay with about 225 injury types, casts and bandages, and condition wording that takes any item from brand new to ruined. Adults only. Drop it in and go.

Feature post. One prompt, the whole world. __world_person__ picks an adult by region and country in proportion to who actually lives there. Then dress them: kilts, kaftans, kimonos, hanbok, kente, flapper dresses, rave gear, corsets and plush novelty onesies. Add a sunburn, a tan, makeup, piercings, a hearing aid or a wheelchair, and stand them in a Cappadocian cave, a Bauhaus block or a slot canyon in a blizzard. Guyanonymous' Wildcards v108.

Getting-started post. Three steps: unzip, copy the wildcards folder into your image tool, then type __world_person__ in a prompt. That's it. No Python needed. Moving from an older version? filename_changes.txt maps every old name to its new one.

Release note.

Version 38 adds a clinical wound and injury catalogue (about 225 injury types), casts, splints and bandages, condition descriptors that apply to any item, tie pins, clips and bars, bolo ties, bed-head and messy hair, fuller chastity device descriptions, fuller regional and formal garment descriptions, and quality files in place of footwear price files.

Full Changelog since V27

Version 108 (this release)

Level 6 same-sex panels: every adult_l6_panel_mm* and adult_l6_panel_ww* line now has a level-6 ending (semen for men, clear fluid for women) that fits the pose, so level 6 differs from level 5.
Level 1: the bent-forward-at-a-table pose is removed from adult_l1_panel_wm (it stays at level 2 and above).
Locations: level 1 and level 3 scenes at wet places (shower, steam room, sauna, plunge pool, hot tub, lake dock) now dress people in swimwear.
Panels and groups: hidden-body wording removed where the pose covers it, 69 geometry tidied, impossible floor position replaced, three-person nude sentences matched to who is in the scene, and duplicate group lines removed.
Fluid wording: 'not white' negations replaced with 'clear and transparent' in adult_l6_location, adult_l6_panel_wm and fluid_scene_squirt.
Activities: all 51 act_*_scene_prompt files start each sentence with a capital ('The scene includes ...'); act_*_frame_prompt uses the new act_frame_style (six on-location photo styles) so a studio backdrop no longer clashes with the place.
Body features: f_body_nipples reworded per breast, head_chin garbled and 'cleft' lines fixed, new body_navel (10 navel shapes).
Showcase files (40 prompts at 1024 and 1408 pixel counts, steps 50, cfg 3.75) and 100 sample prompts are included.

Version 107

Nipples: f_body_nipples gains 35 lines that pair five areola sizes (thumbnail-small to very wide) with seven nipple forms (flat, small, medium, long, thick, inverted, puffy dome) and a surface note, with no skin colour.
Anal area: new f_body_anus and m_body_anus describe rosette shape, radial folds, hair and the cheeks around it, with no skin colour.
Yoga: act_yoga_action grows from 3 to 24 named poses; clothes, equipment and settings are widened, with leotard, leggings and fitted-top options.
Dance classes: act_dance_classes_action grows to 20 moves (ballet, jazz, salsa, hip-hop, contemporary); clothes, equipment and settings are widened.
Martial arts: act_martial_arts_action grows to 20 stances, kicks and strikes; clothes, equipment and settings are widened.
Showcase files (40 prompts at 1024 and 1408 pixel counts, steps 50, cfg 3.75) and 100 sample prompts are included.

Version 106

Clothed men: male_bulge_tight, _brief, _jock, _medium, _loose, _clothed and _covered, written from image tests. Tight stretch fabric holds the penis flat against the belly; briefs outline it; a jockstrap has a modest close pouch; medium fabric softens the outline; loose fabric only tents over an erect penis. Length is given against the body.
Level panels adult_lN_panel_wm/_ww/_mm rewritten with full geometry: who faces whom, torsos turned toward each other, three-quarter camera, hands placed, and both people stated as men or women where needed.
Group scenes: structured three-person scenes at levels 1 to 6 (adult_lN_group), with explicit acts at levels 5 and 6.
Wrestling: new adult_l4_wrestling_mm, adult_l5_wrestling_mm, adult_l6_wrestling_mm.
Location scenes: ten venues for a woman and a man at every level (adult_lN_location), each stating position, hands, place, light and camera.
Solo scenes at level 3 use topless or shirtless in real bottoms; hand-on-stomach lines for women at levels 1 and 2 changed to hand on hip or inner thigh.
Chin: head_chin no longer has a deep cleft, which rendered as a split chin. A cleft is now a small shallow dimple low on the chin tip, with the outline unbroken.
Showcase files (40 prompts at 1024 and 1408 pixel counts, steps 50, cfg 3.75) highlight what is new or changed in this version: level panels with the clothed bulge classes, structured group, wrestling, location and solo scenes. The 100 sample prompts are representative of everything the model can do.
Body-feature showcase (kept outside the package): grids of 2x2 or 2x3 panels with one person per feature, close-ups of just the part and one title per image such as Eye Variations with Guyanonymous' Wildcards in smaller text beneath it. Generators V106_showcase_generator.py and convert_list.py are in the tools folder.

Version 105

Level panels: adult_lN_panel_wm, _ww and _mm for levels 1 to 6. Each panel introduces every person once, ties every body part to its owner, states hands and legs, the place, the light and the camera. Levels 1 to 4 are non-penetrative and step through clothed, underwear, topless in ordinary trousers or skirts, and nude.
Typed acts at levels 5 and 6: vaginal (_wm_vaginal), oral (_wm_oral, _ww_oral, _mm_oral), anal (_wm_anal, _mm_anal), scissoring (_ww_scissor), fingering, strap-on and face sitting (_ww_other) and mutual masturbation (_mm_hand). Every act is described by what the bodies are doing, not by its name.
Named positions from the Wikipedia sex position list, written as physical placement: adult_l5_position_* and adult_l6_position_* for man on top, woman on top, side by side, rear entry, standing, sitting, extra oral, extra anal and group, plus _all.
Kink panels: kink_panel_wm, _ww, _mm, _f, _m cover rope bondage, collars and leashes, impact play, chastity, strap-on, blindfolds, cuffs, latex, leather, harnesses and boots. All adults, consensual play.
Clothed men at levels 2 and 3: male_bulge_clothed and male_bulge_covered show the raised shape of a semi-erect or erect penis defined by the fabric, which stays unbroken and fully covers it, the way fabric hugs and defines a nipple.
Semen is smaller and realistic: amounts of about a teaspoon, a tablespoon at most, leaving the tip in a short low line or falling straight down; no spray or arcs. Squirting stays described physically. Level 6 masturbation for men rewritten.
Penis stages, semen, squirt and grooming files from v104 carry forward.
Showcase files (40 prompts at 1024 and 1408 pixel counts, steps 50, cfg 3.75) highlight what is new or changed in this version; the 100 sample prompts are representative of everything the model can do.

Version 104

Circumcision is a separate modifier: __penis_circumcision__ and __penis_circumcision_erect__ (with forced penis_circumcised / penis_uncircumcised forms), and a rewritten foreskin vocabulary that describes the foreskin physically, from fully covered to fully retracted. __m_body_genitals_mod__ and __m_body_genitals_erect_mod__ keep the old descriptions with the circumcision clause as a token.
Penis overhaul: independent length, girth, shape (including taper, mushroom glans, club head), curve and lean pools for soft and erect (m_body_genitals_v2, m_body_genitals_erect_v2), with left, right, up and down leans, plus shaft, pubic and scrotum pools.
New penis stages: __m_body_genitals_semi_v2__ (semi-erect, partly filled out, softer tip, lifted at a low angle) and __m_body_genitals_post_v2__ (just after ejaculation: softening, drooping, darker dusky head, wet with semen, relaxed scrotum), each with its own pools and penis_look_semi / penis_look_post clauses.
Pubic hair: stubble (pubic_hair_stubble), patchy (pubic_hair_patchy) and shaved shapes (pubic_hair_shape: arrows, hearts, stripes and more).
Ejaculation modifier: semen_look (clear and watery through translucent and thin to thick opaque white, creamy, pearly or faintly yellow, gelatinous or clumpy), semen_amount (a dribble or single stream up to a heavy load of long ropes) and semen_after. All ejaculation lines use the physical wording; the word 'ejaculate' no longer carries the meaning alone.
Squirting is described physically: thin, clear, watery fluid from the vaginal opening, with squirt_form giving a single stream, a jet, two streams, pulses, a spray, a gush or a trickle. Hand engagement is described (index and ring fingers in a come-hither motion, tips on the upper interior wall).
Structured scenes that prevent drift: fluid_scene_semen and fluid_scene_squirt introduce each person once, tie every body part to its owner, state each hand and leg position, and say who is in view. Level 1 to 6 masturbation pools for women and men (adult_lN_masturbation_f / _m).
Adult pose upgrades from v103 carry forward: person, place, light and frame layers, pair handling, device rewrites.
New Beauty and Grooming section: groom_prompt, groom_female_prompt, groom_male_prompt and per-group prompts for female hygiene (tampons, pads, cups, discs, period underwear, wipes, washes), male hygiene, male grooming (razors, brushes, beard and hair products, fragrance, skincare), female beauty (makeup, nails, hair tools, skincare, fragrance) and unisex care, about 575 object descriptions, with quoted invented names on packaged goods.
Showcase files (40 prompts at 1024 and 1408 pixel counts, steps 50, cfg 3.75) highlight what is new or changed in this version; the 100 sample prompts are representative of everything the model can do.

Version 103

Adult poses now come with people: __adult_l1_pose_person__ ... __adult_l6_pose_person__ give every pose line a distinct adult (age 20s-70s, heritage, skin, build, hair, facial hair) from __apose_person_f__ and __apose_person_m__ (1,600 each).
__adult_l1_pose_scene__ ... __adult_l6_pose_scene__ add a place, a light and a framing clause (__apose_place__, __apose_light__, __apose_frame__), so the grids no longer look like one warm lamp-lit room.
Vague device phrases ('a small remote', 'a buzzing tube', 'a ring that buzzes') were rewritten as concrete shapes and colours so the model draws them as intended.
Age pools cleaned: young skin, glow, hair colour and strenuous poses removed from the 65+ brackets, aged skin removed from the youngest brackets.
Age ladders: __age_step1_body_skin__ ... __age_step9_body_skin__ and __age_step1_hand_skin__ ... __age_step9_hand_skin__ give an evenly graded body and hand ageing for age grids.
Hand pools no longer bake in 'swollen', 'knobby' or 'arthritic' (the model drew these as round balls). Joint conditions are now an optional modifier: __hand_condition_mild__, __hand_condition_marked__ or __hand_condition__, described by concrete shapes, to add to any hand prompt.
Pubic hair: new __pubic_hair_stubble__ and __pubic_hair_patchy__ pools, and a separate __pubic_hair_shape__ modifier for hair shaved or trimmed into shapes (heart, arrow, line, triangle, diamond, star, lightning bolt and more). __f_pubichair_all__ and __m_pubichair_all__ combine them with the existing pools.
A 40-image prompt showcase at 1024 and 1408 pixel counts, steps 50, cfg 3.75, and 100 sample prompts.

Version 102

Object-first brand-style prompts now use commas instead of periods inside the style text, so the style reads as one unified description (your A/B tests favoured commas).
Brand logos, nameplates, hallmarks and monograms inside style texts are replaced by a random quiet motif (bs_motif): a random simple shape in a random kind of mark (stamped, embossed, engraved, etched, printed, woven, stitched, enamelled and more).
bs_mark now pairs a mark kind with its material (stamped leather, enamelled metal, woven cloth ...) and can show short capital-letter initials.
The 'plain and unmarked' wording was removed from style texts (text rule: do not mention what should not appear).
A 40-image prompt showcase at 1024 and 1408 pixel counts, steps 50, cfg 3.75, and 100 sample prompts.

Version 101

New bs_mark add-on: a small mark built from random shape, material, outline, size and position, carrying a fake logo and name from the name system. Add it by hand to any prompt or use bs_marked_prompt for a random one.
Brand styles are about design mentality, not one object: sentences naming shoes, cars, bags, shops and other objects were removed from 122 styles (84 styles that are only about one kind of object are unchanged).
Converse rewritten as linework, curves, materials and palette, with a random simple shape as its motif.
Brand-style prompts now start with the object tied to the style ('A wallet done in a design language of ...'), give the short style, and repeat the object once at the end, so the style and the object stay connected.
Subject lists no longer contain shops, kiosks, stations, posters or signs that invite garbled text.
Sample files now draw one prompt from every family of the package (about 90 families), so they show everything it can do; the showcase files show what is new in this version.
A 40-image prompt showcase at 1024 and 1408 pixel counts, steps 50, cfg 3.75, and 100 sample prompts.

Version 100

Brand-style prompts now pick from 24 views and lighting set-ups (late-afternoon sun, overcast, a single warm lamp, low angle, flat lay, gallery plinth and more) instead of 8, so showcases and samples no longer repeat the same studio light.
Brand descriptions that begin with a noun phrase now read 'Styled after a ...' instead of 'Styled in a design language of a ...'.
A 40-image prompt showcase at 1024 and 1408 pixel counts, steps 50, cfg 3.75, and 100 random sample prompts.

Version 99

70 new brand styles of defunct or historic brands, from Saab and Pontiac to Pan Am, Braniff, Biba, Dansette, Trabant, Sinclair, Agfa and Woolworths, each written as an object style: colours, forms, materials and patterns only.
The file name carries the brand (for example maker_bs_trabant); the prompt never names it and contains no logo or text.
They join bs_prompt, maker_brand_style and the new bs_past_prompt, with the same plain-surface wording as every other brand style.
A 40-image prompt showcase focused on the new styles at 1024 and 1408 pixel counts, steps 50, cfg 3.75, and 100 random sample prompts.

Version 98

The 746 bs_cell_*_named files no longer say 'in the style of '; they carry the same design-language text as the plain cell files (the file name carries the brand).
Brand style descriptions drop clauses mentioning badges, tags, labels, signs, banners, emblems, crests, stamps and similar, so no undefined text or marks appear.
About 6,000 unquoted phrases such as the words Steamy Car Windows in badge, medal, pin and insignia files are now quoted.
A 40-image prompt showcase at 1024 and 1408 pixel counts, steps 50, cfg 3.75, no trailing blank line, and 100 random sample prompts.

Version 97

Library-wide text audit (STANDING_RULES section 0): about 63,000 lines in 2,587 files mentioned signs, posters, banners, graffiti, labels, tags, badges, logos, lettering, inscriptions or slogans without saying what text appears, so the image model invented words. They now use plain visual nouns (lit panel, picture panel, cloth pennant, spray-painted shapes, tab, patch, motif, stitched border).
Engraved names, stamped initials and monograms in accessory files now carry quoted invented text from the name system, for example the name "name_person" or the letters "name_initial name_initial", so a wildcard expands to real quoted text you can edit.
About 150 remaining text denials ('no text', 'no words') were removed.
Not changed on purpose: badge, pin, patch and sticker files (their items are meant to carry defined text), sign_ and text_ files, and the name system.
A 40-image prompt showcase at 1024 and 1408 pixel counts, steps 50, cfg 3.75, no trailing blank line, and 100 random sample prompts.

Version 96

Removed every 'no other words or letters', 'nothing else' and 'no readable text' style negation from the brand styles and the name system (about 5,900 lines). The model notes say negations did not help and that naming words, letters and labels invites lettering; the technical report says training captions quote visible text verbatim, so the prompts now quote the text that should appear and say nothing about text that should not.
Brand styles end with a positive line: the surfaces are plain and unmarked, and the style is carried by colour, shape, material and pattern alone. No label, tag or badge is mentioned, so there is nothing to fill with invented words.
Name-system prompts keep their quoted text, one line each, and no closing denial. The blank-text prompt is now plain unmarked surfaces decorated with shapes and patterns.
A 40-image prompt showcase at 1024 and 1408 pixel counts, steps 50, cfg 3.75, no trailing blank line, and 100 random sample prompts.

Version 95

Fixed brand-style prompts printing the sharpness trigger word or the real brand name on the object. Image metadata showed the trigger word was sitting in front of the prompt and the model printed it wherever a wordmark, label or tag was described but not defined.
Every brand style ends with a defined text line: any label, tag, badge, packaging or sign is small, and the only words on it are an invented brand name from the new name system, with no other words or letters anywhere.
Brand styles describe the object, not a logo: wordmark, lettering, slogan and text clauses were removed from the style descriptions, and the prompt no longer says 'In the style of '. The file name is the brand (maker_bs_heinz), so the prompt holds only the detailed design description. The _named files now hold the same text as the plain ones.
No logo front and centre: every style says the main surfaces carry colour, shape, material and pattern, and any label, tag, badge or sign is small and discreet.
Brand-style files now draw that invented name from name_brand_invented, so each image gets a different fake name. Edit the quoted text in the expanded prompt to use your own.
A 40-image prompt showcase at 1024 and 1408 pixel counts, steps 50, cfg 3.75, no trailing blank line, and 100 random sample prompts.

Version 94

A fake name system: invented brand stems (Latin, Nordic, German, tech, Japanese), surnames, first names and trade words, built into company, shop, restaurant, pub, product, person, address, hero, villain, author, book, comic, band, movie and magazine names.
26 business types, each with its own name file and sign prompt (liquor store, bar, pharmacy, bookshop, bakery, cafe, hotel, garage, cinema and more), with a matching extra line such as OPEN or Cold Beer.
Packaging prompts where name, product type, flavour and weight agree: soft drinks, sweets, savoury food, spirits and beer, household goods.
Business cards: eight positioned layouts (corner, centred, split, minimal, corporate, vertical, front and back, trade) with company, website, email, phone and address that agree. Websites end in .example and phones use 555.
Fake logos: a described symbol, style, colour, layout and quoted wordmark, in logo, letterhead, stamp and sign prompts. Baseball caps, t-shirts, hoodies, tote bags and mugs carry defined text or logos.
Older sign, advert and label prompts (about 170 files) now draw from the system through text_brand, text_shop_name, text_pub_name, text_ship_name, text_card_name and text_map_name.
Every text-bearing prompt quotes each line of text and ends with a no-other-words statement, so the image model does not invent words. See docs/NAME_SYSTEM.txt.
A 40-image prompt showcase at 1024 and 1408 pixel counts, steps 50, cfg 3.75, no trailing blank line, and 100 random sample prompts.

Version 93

150 more brand styles rewritten after checking image grids, as object design language only (no photography, advertising, settings or people): Lotus, Peugeot, Bugatti, Maserati, Fiat, Saab, Skoda, Dodge, Chrysler, Lincoln, DeLorean, Nissan, Datsun, Lexus, Suzuki, Hyundai, Kia, Mitsubishi, Lada, Holden, Moschino, Missoni, Salvatore Ferragamo, Brooks Brothers, L.L.Bean, Mary Quant, Jean Paul Gaultier, Fjallraven, Orla Kiely, Cath Kidston, Clarks, Pendleton, Lanvin, Schiaparelli, Bottega Veneta, Mulberry, Issey Miyake, Comme des Garcons, Yohji Yamamoto, Kenzo, Onitsuka Tiger, A Bathing Ape, Havaianas, Akubra, Blundstone, R.M. Williams, Zimmermann, Arcteryx, Black Diamond, Stanley, Thermos, Opinel, Leatherman, Maglite, Burton, Quiksilver, Rip Curl, Billabong, Dickies, Alpha Industries, Coleman, Snow Peak, Shimano, Campagnolo, Brompton, Montblanc, Parker Pens, Lamy, Staedtler, Faber-Castell, Winsor and Newton, Caran d Ache, Moleskine, Pelikan, Rotring, Steinway and Sons, Martin Guitars, Gretsch Guitars, Rickenbacker, Taylor Guitars, Moog Music, Roland, Teenage Engineering, Bose, Sonos, Technics, Zenith Radio, Philips, Telefunken, Grundig, Atari, Commodore, Sega, Hasselblad, Fujifilm, Nikon, Nokia, Motorola, GoPro, Xiaomi, Lenovo ThinkPad, Nothing, Google Material Design, Windows 95 era interface, Early Mac OS interface, Kinder, Pocky (Glico), Lea & Perrins, Aesop, Lush, The Body Shop, Penhaligons, Laduree, Harrods, Selfridges, In-N-Out, Nandos, Pret a Manger, Trader Joes, Qantas, Singapore Airlines, KLM, SAS, Swissair, Air India, JAL, Aeroflot, Concorde, Boeing, Airbus, Triumph Motorcycles, Indian Motorcycle, Lambretta, Kawasaki, Honda Super Cub, Moto Guzzi, Schwinn, Raleigh, John Deere, Greyhound, Riva, Chris-Craft, Meccano, Hornby, Matchbox, Steiff, Ravensburger, Opel, Austin, Morgan.
New subject lists for brand prompts: bs_subject_building, bs_subject_accessory, bs_subject_madeup_vehicle, and a wider bs_subject_any (53 items). New prompts bs_building_prompt, bs_accessory_prompt, bs_madeup_vehicle_prompt (and _named), all included in bs_prompt.
Research notes are in docs/RESEARCH_v93_brands_batch5.txt.
A 40-image prompt showcase at 1024 and 1408 pixel counts, steps 50, cfg 3.75, no trailing blank line, and 100 random sample prompts.

Version 92

80 more brand styles rewritten after checking image grids, as object design language only (no photography, advertising, settings or people): Carhartt, Puma, Uniqlo, Vivienne Westwood, Swatch, Marimekko, Hunter, Stetson, New Balance, The North Face, Supreme, Salomon, Yeti, Victorinox, Columbia Sportswear, Canada Goose, Playmobil, Hot Wheels, Barbie house style, Mattel, Fisher-Price, Sanrio, Aardman, Hanna-Barbera, Taschen, Blue Note Records, Heineken, Cadbury, Toblerone, Haribo, Heinz, Campbell, Tabasco, Marmite, Hershey, Nescafe, Kikkoman, Milo, Bang & Olufsen, Leica, Polaroid, Canon, Casio, Seiko, Omega, Dyson, Smeg, KitchenAid, Fender, Gibson, Marshall Amplification, Samsung, Ducati, Vespa, Royal Enfield, Pan Am, British Airways, Lufthansa, Emirates, NASA, London Underground posters, Routemaster, Shinkansen, Herman Miller, Knoll, Vitra, Alessi, Kartell, Le Creuset, Wedgwood, Meissen, Lalique, Georg Jensen, Iittala, Anglepoise, Liberty, Morris and Co, Pyrex, Tupperware, Chevrolet.
Each brand has maker_bs_<brand> (style only), maker_bs_<brand>_named and the research notes are in docs/RESEARCH_v92_brands_batch4.txt.
A 40-image prompt showcase at 1024 and 1408 pixel counts, steps 50, cfg 3.75, no trailing blank line, and 100 random sample prompts.

Version 91

20 more image-checked brand styles: Bentley, Jaguar, Aston Martin, Audi, Citroen, Renault, Alfa Romeo, Lamborghini, Volvo, Ford, Cadillac, Honda, Subaru, Mazda, Dior, Prada, Versace, Fendi, Dr Martens, Fred Perry (notes in docs/RESEARCH_v91_brands_batch3.txt).
Brand prompt views no longer add settings: no surroundings, indoor, outdoor or layout views, so the style stays on the object.
Brand style files (maker_bs_<brand> and _named) now hold only object design language: shape, material, finish, palette, pattern, lettering, emblem. Photography, advertising and setting sentences moved to new maker_bs_<brand>_mood files (203 brands).
Every brand style ends with a line saying it applies to the object itself and not to its surroundings.
Brand prompts (bs_prompt, bs_prompt_named and per-group) now ask for the object to be styled with the setting kept plain and neutral, and subject lists no longer include people, rooms, streets or other scenes.
A 40-image prompt showcase at 1024 and 1408 pixel counts, steps 50, cfg 3.75, no trailing blank line, and 100 random sample prompts.

Version 90

Brand style files (maker_bs_<brand> and _named) now hold only object design language: shape, material, finish, palette, pattern, lettering, emblem. Photography, advertising and setting sentences moved to new maker_bs_<brand>_mood files (203 brands).
Every brand style ends with a line saying it applies to the object itself and not to its surroundings.
Brand prompts (bs_prompt, bs_prompt_named and per-group) now ask for the object to be styled with the setting kept plain and neutral, and subject lists no longer include people, rooms, streets or other scenes.
A 40-image prompt showcase at 1024 and 1408 pixel counts, steps 50, cfg 3.75, no trailing blank line, and 100 random sample prompts.

Version 89

Vans, Converse and Land Rover brand styles rewritten so no shoe parts or vehicle parts remain: they apply to any product.
A 40-image prompt showcase at 1024 and 1408 pixel counts, steps 50, cfg 3.75, no trailing blank line, and 100 random sample prompts.

Version 88

25 more brand styles rewritten after checking image grids: Rolls-Royce, Mini, Land Rover, Volkswagen, BMW, Mercedes-Benz, Toyota, Tesla, Louis Vuitton, Gucci, Hermes, Ralph Lauren, Lacoste, Tiffany and Co, Lego, Vans, Converse, Pepsi, Guinness, Starbucks, McDonalds, Braun, Nintendo, Rolex, Harley-Davidson.
Each style describes transferable design language (shapes, materials, finish, palette, lettering, photography) usable on any product, with no named parts or models.
Default maker_bs_<brand> gives the style only; _named adds the name. Research notes are in docs/RESEARCH_v88_brands_batch2.txt.
A 40-image prompt showcase at 1024 and 1408 pixel counts, steps 50, cfg 3.75, no trailing blank line, and 100 random sample prompts.

Version 87

Men-only staged play-fighting for levels 1 to 3 (adult_wrestling_mm, adult_lN_wrestling_mm_prompt, 90 lines): pillows, sauna, mud, oil, gym mat and ring.
Off-White, Bosch Professional, Makita and SodaStream brand styles rewritten as product-neutral essence descriptions.
Pinback and adult achievement files deduplicated: no picture or wording repeated across files.
Levels 1 to 5 cleaned of crowd, audience and spectator mentions, so no scene implies five or more adults.
Showcase lighting now matches the scene (indoor, outdoor, or neutral); place-specific lighting lines are left out.
A 40-image prompt showcase at 1024 and 1408 pixel counts, steps 50, cfg 3.75, no trailing blank line, and 100 random sample prompts.

Version 86

Pinback buttons (pinback_prompt, 676 lines): slogans, music, fandom, causes, identity, events, adult, plain designs - each a round button with size, border, print finish, picture, words and safety-pin back.
Police, fire, ambulance and military insignia (insignia_prompt, insignia_medal/pin/badge_prompt, 858 lines) with invented towns, units and names; as patches, medals, enamel pins and badges.
Adult achievement merit badges, medals, enamel pins and badges for dating, kink, sex and fetish (acc_meritbadge_adult_prompt and siblings, 1067 lines).
Adult kilts for L4-L6, pony-play gear and torn or ripped-off clothing for L1-L6 (adult_kilts_levels, adult_pony_levels, adult_torn_levels, 892 lines), and L6 ejaculate extras (65 lines).
A 40-image prompt showcase at 1024 and 1408 pixel counts, steps 50, cfg 3.75, no trailing blank line, and 100 random sample prompts.

Version 85

Adult level 6 (adult_l6_<pose|body|setting|object|group|location>, 418 lines): ejaculate, squirting described plainly, and groups of five or more adults. Levels 1 to 5 stay at four adults or fewer.
Optional modifiers for every level, L1 to L6: adult_lN_expression and adult_lN_bodychange (544 lines) - faces, flushing, sweat, trembling, calm or tolerant looks, intensity within consensual play.
Messy play and staged play-fighting: adult_lN_messy and adult_lN_wrestling for L1 to L6 (611 lines), plus per-level prompts.
Gear (604 lines): adult_gear_materials, _garments, _uniforms, _kilts, _tailoring, _pony, _furniture, _roleplay, _fetish_focus, all in adult_gear.
Existing L1-L5 adult lines audited: jargon replaced with plain visual descriptions, ejaculate and large-group lines moved to L6, urine, school, sleeping or drunk partner lines removed.
20 top brand styles rewritten after checking product images (Jeep, Nike, Adidas, Coca-Cola, Apple, Sony, Levi Strauss, Chanel, Burberry, Porsche, Ferrari, Studio Ghibli, Disney classic, Pixar, Patagonia, Barbour, Zippo, Muji, Olivetti, IKEA).
A 40-image prompt showcase at 1024 and 1408 pixel counts, steps 50, cfg 3.75, no trailing blank line, and 100 random sample prompts.
A list of open questions for review: docs/ADULT_QUESTIONS_v85.txt.

Version 84

Adult content (adults only, consensual): adult_l1 to adult_l5 x pose, body, setting, object, group, location (1847 lines in 30 files). L1 suggestive, L2 sensual, L3 partial nudity, L4 explicit nudity, L5 explicit acts in clinical wording.
Mixed sets adult_<genre> (all five levels), and prompts adult_l1_prompt to adult_l5_prompt, adult_solo_prompt, adult_group_prompt, adult_prompt.
Objects split by visibility, for example worn under clothing with the shape visible through fabric versus uncovered.
286 existing sex, BDSM and fetish object files moved under the adult_ prefix (normal clothing and underwear stay put).
Brand styles (746) rewritten as transferable design essence: mood, shape language, materials, colour and lettering, applicable to any subject (no exact products or parts).
Brand naming: maker_bs_<brand> is the style only, with the brand name left out (use it in a prompt like 'A maker_bs_jeep typewriter' or after the subject). maker_bs_<brand>_named starts 'In the style of ,'. The same pairing exists for maker_brand_style* (all, group, region, era, industry), bs_cell_<brand>[_named] and bs_prompt[_named].
A 40-image prompt showcase (8 per level) at 1024 and 1408 pixel counts, steps 50, cfg 3.75, plus a list of open questions for review.
100 random sample prompts, one per line with no blank lines.

Version 83

Brand styles (746 brands in 1669 files, maker_bs_<brand>, maker_brand_style). Each is one line, 'In the style of , ...', a distilled description of what that maker's products look like (shapes, materials, finishes, colours, lettering, patterns, mood), under 250 words. Companies, marques, houses and workshops from every era (ancient and 18th century houses to today) and every region: Europe 293, North America 157, East Asia 112, Russia and Eastern Europe 41, South Asia 31, Latin America 31, Oceania 27, Africa 19, Southeast Asia 18, Middle East 17.
Browse by group, region, era or industry: maker_brand_style_<group> (cars, transport, fashion, tech, home, media, packaging, heritage, modern and world), maker_brand_style_region_*, maker_brand_style_era_* (before 1900, 1900-1945, 1946-1979, 1980-present) and maker_brand_style_industry_*. Ready prompts: bs_prompt and bs_<group>_prompt, with subject lists bs_subject_* and views bs_view.
The existing vehicle maker files (maker_<make>_style) now also include the full style description for that make.
Illustration and photography style lists enriched: 59 files, 2336 lines rewritten with full visual descriptions (lineages and movements, media, textures, signature devices, photo eras, processes, cameras, lenses, filters, tonality, finishes, editing, artifacts, grain, lighting, atmosphere and signature photographic styles), same order and same slots.
A 40-image prompt showcase for this release, in two sets (1024 and 1408 pixel counts) with varied image ratios, steps 50 and cfg 3.75.
100 random sample prompts, one per line with no blank lines.

Version 82

Sculpture and carving (about 234 files, af_sculpt_). 19 forms: marble, bronze, granite, soapstone, alabaster, wood, chainsaw carving, totem, terracotta, plaster, welded steel, wire, assemblage, mobiles, ice, sand, relief, topiary and stone cairns.
Ceramics, glass and decorative craft (about 306 files, af_craft_). 25 forms: stoneware, porcelain, raku, celadon, majolica, kintsugi, blown and fused glass, lampwork, mosaic, cloisonne, lacquer, marquetry, pyrography, scrimshaw-style engraving, leather, tiles, stucco and terrazzo.
Fibre, paper and miniatures (about 270 files, af_fibre_). 22 forms: tapestry, quilt, embroidery, felt, crochet, macrame, lace, batik, quilling, origami, paper cutting, papier-mache, lanterns, dioramas, dollhouse rooms, model railways, puppets and ships in bottles.
3D renders, toys and installations (about 367 files, af_render_). 30 forms: clay, product, low-poly, voxel, wireframe, toon and glass renders, claymation, felt puppet scenes, brick builds, vinyl toys, neon, LED and mirror-room installations, sand mandalas, chalk art, ice, sugar, fruit and floral art, fireworks and land art.
Each form has its own subjects, materials, techniques, surface finishes, colours, scales, wear and display settings; plaques and labels carry their text in quotation marks.
A 40-image prompt showcase for this release, in two sets (1024 and 1408 pixel counts) with varied image ratios, steps 50 and cfg 3.75.
100 random sample prompts, one per line with no blank lines.

Version 81

Robots (about 316 files, rb_). 25 robot types, each with its own body, head, eyes and sensors, arms, hands, movement and feature lists: tin toy robots, humanoids, androids, helper robots, industrial and laboratory arms, four-legged and many-legged walkers, rovers, tracked robots, drones, giant mechs, snake robots, orbs, cleaners, pets, clockwork automatons, combat robots, underwater and space robots, tiny insect robots, exoskeletons, social robots, warehouse robots and alien machines.
Shared robot pools. Colour schemes, wear, markings with quoted names and warnings, places, lighting and camera shots, grouped by style (toy, modern, rugged, antique, strange).
Gadgets and tech devices (about 485 files, gd_). 42 types covering phones from the early 1980s brick to foldables, smartwatches, bands, rings, glasses, headsets, earbuds, headphones, e-book readers, tablets, organisers, handheld consoles, music players, cameras, pagers, radios, calculators, navigation units, speakers, chargers, mainframes, home computers, towers, all-in-ones, laptops, gaming computers, terminals, server racks, keyboards, monitors, printers, routers and mice.
Every device is described by its look (shape, size, screen, controls, material, colour, wear); any text on a screen is written out line by line in quotes.
Four reference files maker_robot_real_models, maker_robot_fiction_models, maker_gadget_models and maker_computer_models hold the real and fictional names with full descriptions, kept out of the prompts.
A 40-image prompt showcase for this release, in two sets (1024 and 1408 pixel counts) with varied image ratios, steps 50 and cfg 3.75.
100 random sample prompts, one per line with no blank lines.

Version 80

Treasure (about 80 files, tr_). Crowns and tiaras, jewellery, goblets and vessels, idols and statuettes, chests and hoards, relics, gemstones, and treasure maps, each fully described.
Cards (about 80 files, cd_). Playing cards, tarot cards, trading cards, postcards, business cards and labels, board game cards, flash cards and collector packs. Printed captions appear in quotation marks.
Minerals and finds (about 80 files, mn_). Crystal clusters, polished stones, geodes, ores, fossils, meteorites, cut gemstones and amber, with where each was found and a collector label in quotes.
Sound equipment (about 80 files, sd_). Record players, radios, tape players, speakers, microphones, headphones, mixing desks and jukeboxes.
Screens (about 85 files, sc_). Televisions, monitors, laptops, handheld games, arcade cabinets, digital signs, instrument screens and projectors. Each screen has its exact text, one quoted line at a time.
Art (about 80 files, ar_) and nature finds (about 80 files, nt_). Brushes, paints, easels, palettes, sketchbooks, sculpting and printing tools, framed art; shells, feathers, leaves, nests, cones, driftwood, pinned insects and plants.
A 40-image prompt showcase for this release, in two sets (1024 and 1408 pixel counts) with varied image ratios, steps 50 and cfg 3.75.
100 random sample prompts, one per line with no blank lines.

Version 79

Film noir (about 250 files, nr_). 8 visual styles, 16 character types, 22 places, 17 objects and 8 vehicles, plus pair and group scenes. Signs and papers carry their words in quotation marks.
Pirates (about 260 files, pt_). 8 visual styles, 16 character types, 20 places, 18 objects and 8 ships and boats, plus pair and group scenes.
Spy thriller (about 270 files, sy_). 8 visual styles, 16 character types, 22 places, 17 objects and 8 vehicles, plus pair and group scenes. Text on screens and boards is spelled out, one quoted line at a time (sy_screen_text).
Money (about 40 files, mo_). Coins, banknotes, purses and wallets, money boxes, cheques and bonds, gaming chips and tokens, account books, counting equipment and bullion, each with type, material, feature, era, invented maker marks, condition and display.
Metalware (about 40 files, mt_). Brass, copper, pewter, silver, cast iron, tin, locks and keys, hand tools and bronze, with the same pieces of detail and a finish pool.
Fixes: the 14 named e-reader models and 13 named phone models now have full physical descriptions instead of a few words, and every screen pool (ereader_screen, phone_screen, phone_screen_old, phone_screen_future, and the cyberpunk cy_screen_text) gives each line of text in quotation marks.
A 40-image prompt showcase for this release, in two sets (1024 and 1408 pixel counts) with varied image ratios, steps 50 and cfg 3.75.
100 random sample prompts, one per line with no blank lines.

Version 78

Cyberpunk (about 270 files, cy_). 8 visual styles, 18 character types, 24 places, 18 objects and 8 vehicles, plus pair and group scenes.
Steampunk (about 290 files, sp_). 8 visual styles, 18 character types, 24 places, 18 objects and 10 vehicles, plus pair and group scenes.
Western (about 300 files, we_). 8 visual styles, 18 character types, 24 places, 18 objects and 8 vehicles, plus pair and group scenes. Signs and posters carry their words in quotation marks.
Steampunk phones (about 40 files, spp_). 12 designs from a crank wall phone and candlestick phone to a pocket handset, a wrist communicator and a steam slate, plus a timeline row.
Home and decorative collections (about 70 files, hd_). Vases and ceramics, lamps, mirrors, picture frames, glassware, rugs, candlesticks, cushions and textiles, wall art and ornaments, each with type, material, feature, era, invented maker marks in quotation marks, condition and display.
A 40-image prompt showcase for this release, in two sets (1024 and 1408 pixel counts) with varied image ratios, steps 50 and cfg 3.75.
100 random sample prompts, one per line with no blank lines.

Version 77

Fantasy (about 500 files, fa_). 12 visual styles, 24 folk from elves to beast-kin, 20 character types, 30 places and 20 objects, plus party and duel scenes.
Horror (about 400 files, ho_). 11 styles from gothic to folk horror, 24 entities, 16 survivor types, 30 places and 20 objects, described by looks and mood only, with no gore.
People interacting (about 200 files, pi_). 46 adult pair actions (handing over, dancing, swordfights, chases, heists, arguments, embraces and more) and 14 group scenes of 4 to 6 people, with positions given left to right and people drawn from the existing person pools.
Toys and models (about 300 files, toy_, dc_). Action figures, fashion, porcelain and rag dolls, wooden toys, plush animals and creatures, resin statues, 16 playsets, eight kinds of boxed games and small toys, 13 kinds of toy packaging by era and condition, and die-cast models of cars, trucks, aircraft, ships, trains, buildings, figures and miniatures.
A 40-image prompt showcase for this release, in two sets (1024 and 1408 pixel counts) with varied image ratios, steps 50 and cfg 3.75.
100 random sample prompts, one per line with no blank lines.

Version 76

Science fiction (about 615 files, sf_). 15 visual styles (space opera, retro-futurism, biopunk, solarpunk, space western and more), 52 alien species, 48 locations from ring stations to alien jungles, 32 vehicles, 16 robots and 10 spacesuits, each described by looks only.
E-reader and phone timelines (about 120 files). 18 e-reader generations from the late 1990s to a 2026 foldable e-ink book, and 31 phone generations from the 1983 brick to rollable and future designs. Real device names are in maker_ereader_ and maker_phone_ files only.
Timekeeping and instrument collections (about 110 files, col_). Watches, clocks, scientific instruments, cameras, typewriters, radios and early electronics, telephones and musical instruments, each with type, material, feature, era, invented maker names in quotation marks, condition and display.
A 40-image prompt showcase for this release, in two sets (1024 and 1408 pixel counts) with varied image ratios, steps 50 and cfg 3.75.
100 random sample prompts, one per line with no blank lines.

Version 75

Legendary creatures (about 1,000 files, leg_). 121 creatures from folklore on every continent, described by looks only: dragons, griffins, trolls, kitsune, djinn, thunderbirds and many more. Each has a look, habitat, prompt, scene with time and weather, encounter with a small witness, and an art version (painting, relief, tapestry). Regional mix files and a two-beings face-off file.
Archetypal sacred and divine figures (18, leg_sacred_). Sun-lord, sea-mother, storm-father, trickster, underworld queen and others, with no named deities, shown with poses, offerings and surroundings.
Locations (about 3,800 files, loc_). The Location Inventory pilot: 325 places (home interiors and exteriors, workplaces, circulation spaces, bathrooms, food and hospitality, unusual places), each with item, furniture, installed feature, signature detail, material, palette, light, human trace and condition pools, a clean scene, a full scene and a scene that adds items from the shared anywhere pools. Tag indexes list scenes by era, light, scale and more.
Object builders (about 560 files, ob_). 66 everyday objects found in the locations that had no builder: tickets, lockers, napkins, radiators, fridges, jars, trolleys, microphones, projectors and more, each built from type, colour, material, era, wear and detail.
A 40-image prompt showcase for this release, in two sets (1024 and 1408 pixel counts) with varied image ratios, steps 50 and cfg 3.75. Every description in it comes from the wildcard files.
100 random sample prompts, one per line with no blank lines.

Version 74

Books, magazines, comics, newspapers and zines (about 200 files). Eight book types (novel, cookbook, field guide, textbook, picture book, poetry, detective story, technical manual) with size, binding, cover, spine, page edges, jacket, condition, state and resting surface, plus about 2,000 invented titles and 400 invented authors. 12 magazine kinds, comics and graphic novels across six printing eras, newspapers, zines and pamphlets, bookmarks, book holders and reading lights. Every one has a version with invented text and a fill-in version (_prompt_fill) where the title, author and headline are placeholders in quotation marks for you to edit.
Signage and the shared lettering engine (about 340 files). 16 sign structures (shop fascias, pub signs, neon, billboards, chalkboards, enamel plates, street plates, licence plates, ship name boards and more), each paired with fitting lettering materials, plus lettering styles, colours, layouts and wear. Invented shop names, slogans, notices, menus and brand names, all in quotation marks. The same engine drives advertising signs (18 kinds from enamel plates to ghost signs) and labelled tins, jars, bottles, cartons and cans (28 container kinds with era, label elements, print method and condition).
Vehicle interiors (about 200 files, vint2_). 18 interior classes (saloon car, sports car, truck cab, bus, train carriage, tram or subway, light aircraft, airliner flight deck and cabin, helicopter, boat helm, ship bridge, submarine control room, tractor cab, motorcycle cockpit, steam locomotive cab, airship gondola, horse-drawn carriage), each with layout, seats, dashboard, controls, instruments, materials and extras, plus era, wear, light and the view outside. Group entries for road, rail, air, sea and old vehicles.
Architecture rebuilt in layers (about 700 files, bld_). 33 architectural styles (Tudor to brutalist and contemporary glass) and 32 regional and vernacular types (adobe, longhouse, riad, machiya, hanok, stilt house and more), 20 building types, then parts, ornament, foundations, entrances, boundaries, gardens, wealth, age and the existing condition pieces. Entries come in prompt, short, full and scene forms, plus a blend of two traditions. The older arch_ family is unchanged.
Creature and hybrid builders with powers and effects (about 250 files, cb_ and pw_). 81 animals described part by part (head, body, limbs, tail, covering), single-animal creatures, correlated hybrids with two or three source animals, human-and-animal hybrids (adults only), winged creatures, body plans, sizes, eyes, features, poses and habitats. 20 powers (fire, ice, lightning, light, shadow, nature, water, wind, earth, arcane, energy beam, ray gun fire, force field, aura, transformation, telekinesis, portals, healing, illusion, time and gravity, summoning, curse), each with form, colour, motion and surroundings, written as what you would see.
Paper and print collections (about 75 files). Stamps with invented countries, trading cards with eras and conditions, ephemera, maps and atlases, postcards, rare books with provenance and condition, and graded comics, each with a fill-in version and a spread or page entry.
A 40-image prompt showcase for this release, in two sets (1024 and 1408 pixel counts) with varied image ratios, steps 50 and cfg 3.75. Every description in it comes from the wildcard files, and each prompt was reviewed before it was kept.
100 random sample prompts, one per line with no blank lines.

Version 73

Expressions have intensity now. 68 emotions come in mild, strong and extreme tiers, each described by what the mouth, eyes, brows and head do, with the emotion name kept as a label. Both expression files were rebuilt from them, and expr_mild, expr_strong and expr_extreme let you pick the intensity yourself.
A consistency pass over the scene entry points. City time and weather are drawn as matching pairs, penguins and flightless birds never fly, clothing condition prompts no longer mix a spotless item with broken fastenings, and a check script flags such pairs (182 flagged before, 0 to 1 after).
A depth and occlusion family (depth_*) for choosing who or what is in front of whom, with a scale rule (far figures about one third the height of near ones) taken from the 16-person scene tests.
Anchoring phase 2. 6,568 sentences in 448 long-form person files now start with 'The woman's', 'The man's' or 'The person's' instead of Her, His or Their.
Cleanup. 272 file names with a double underscore were renamed and every reference fixed, blank padding in mark_tattoo and dem_unique became single-space none options, curly quotes, dashes and a colon were removed, school-backpack wording was reworded, and a general chastity_belt group was added.
New tools. check_build.py is a standing check before every release, and the All Options Once builder now builds safely into a temporary folder, finds its modules by itself and refuses to replace the real folder after a failed run.

Version 72

Facial hair is now described, not named. Image tests showed that style names such as soul patch, balbo beard or Van Dyke are unreliable, and a shaved face often came out with a white beard. All 303 distinct lines in m_facial_hair, the nine m_age<lo>_<hi>_facial_hair files and the nine any_age<lo>_<hi>_facial_hair files now say where the hair grows, where the skin is bare, and how long and how shaped it is. The soul patch is a tiny tuft of hair directly under the lower lip about the size of a thumbprint, and shaved faces use positive wording (smooth, shiny, hairless skin on the chin, cheeks and upper lip, like freshly shaved skin). Line counts and age weights are unchanged.
The person templates now say His lower face has ... or Their lower face has ... instead of facial hair is, so hair lines and bare-skin lines both read correctly.
Scar lines describe how the scar looks. The 398 generated lines in mark_scars and skin_scars (a shape word plus a body place) now give colour, relief and length, for example a thin, pale, slightly raised scar about four centimetres long on the forearm. The same short wording had drawn a pale raised scar in one image and a deep shadow-filled one in another. Hand-written scar lines are unchanged.
The anchor pass from the build called v71 is included. Furniture descriptions that started a sentence with It has are merged into one The <object> has A, B and C sentence, room openers use the room noun, the person templates (m_, f_ and any_ create_random files and world_c_ files) are one chained sentence per person with no pronoun-start sentences, the ageing summary pieces start with The man, The woman or This person, and six bare none furniture options are reworded. The reason is that in a scene with several people the model mixes up attributes when a sentence starts with He or She.
The scripts that make these changes (_tools/v71_build.py, _tools/v72_facial_hair.py, _tools/v72_scars.py) are in the _tools folder. Run them in that order after any rebuild. Each can be undone with --restore.
100 random sample prompts, one per line with no blank lines.

Version 70

104 real manufacturers and makers across 10 categories (43 car makers, 15 motorcycle makers, 11 truck, bus and tractor makers, 9 aircraft makers, 8 bicycle makers, 5 skateboard and skate styles, 4 boat builders, 4 rail and bus styles, 3 space programmes, 2 scooter makers), with 213 decade descriptions from the 1930s to the 2020s.
This is the one part of the library that uses real brand names on purpose, as you asked. Each style is also described in plain visual terms (grille shape, roofline, lamps, fins, stripes, tank shape and so on), so it works even when a model does not know the name.
Per maker: maker_<maker>_<decade> (one decade), maker_<maker>_style (any decade), maker_<maker>_genre (the vehicle genres it makes, taken from the v69 genres and described without a random paint job so the maker's own look decides the colours and details), and the entries maker_<maker>_prompt, maker_<maker>_scene_prompt (with a pose and camera angle) and maker_<maker>_person_prompt (with a person). maker_<maker>_<decade>_prompt pins a decade.
Mix and entry files. maker_prompt, maker_scene_prompt and maker_person_prompt pick any maker. maker_<category>_prompt picks within cars, trucks, motorcycles, scooters, bicycles, boats, aircraft, spacecraft, rail or skates. maker_decade_<decade>_prompt and maker_decade_prompt pick by decade. maker_blend_prompt blends two makers' styles on a random real vehicle, and maker_style_any can be dropped into any vehicle prompt.
The index docs/MAKERS_v70.txt lists every maker, the genres it applies to and the decades covered.
100 random sample prompts, one per line with no blank lines.

Version 69

157 vehicle genres in 9 domains. Road (24: city car to supercar, muscle car, hot rod, rally car, race car, limousine, microcar, lowrider, art car, taxi, police car), trucks (12), two-wheelers and skates (30: cruisers to dirt bikes, road to cargo bicycles, kick and electric scooters, unicycle, skateboard, longboard, roller skates, inline skates), rail (10), working and leisure vehicles (16: tractors, harvesters, forklifts, carriages, wagons, sleighs, campers, caravans, buses), watercraft (26), aircraft (16), spacecraft (8) and imagined vehicles (15: flying cars, hoverbikes, steampunk airships, mechs, sky galleons, starships, flying carpets, clockwork carriages and more).
Every genre is built in parts: vx_<genre>_form, _feature and _setting, assembled by vx_<genre>_prompt with the paintwork from the paint system. Each prompt starts with a plain explanation of what the vehicle is.
Poses and states. Eleven vehicle classes (car, truck, motorbike, bicycle, skate, rail, boat, ship, air, space, imagined) each have vs_<class>_state (wheels turned, lights on or off, doors, bonnet or boot open, drifting, parked, covered, on a jack, heeled over, taking off and more), vs_<class>_camera (angles) and vs_special_<genre> for 49 specialised vehicles (tipping dump truck, extended fire ladder, winching tow truck, hovering helicopter, inflating balloon, lifting-off rocket and more).
People with vehicles. vs_<class>_person has poses for one person using or touching the vehicle (hands at the wheel, kneeling at a tyre, looking under the bonnet, changing a wheel, leaning on the door, leaning into a bend, coiling a rope, climbing into a cockpit) and vs_<class>_group has poses for groups (a pit crew, a family loading the boot, riders side by side, a ground crew). vx_<genre>_person_prompt, _group_prompt and _person_frame_prompt put them together, with the v57 full-body framing for people.
Paint. vpaint_colour (40 named shades), vpaint_finish (17 finishes from gloss and metallic flake to patina, chrome wrap and vinyl) and vpaint_effect (20 markings from racing stripes and flames to camouflage and sign-writing), combined in vpaint_prompt and vpaint_scene_prompt.
Damage. vdmg_<class>_light, _moderate and _heavy give wear, scratches, dents, cracked glass and wrecks for six classes. vxd_<class>_prompt, vxd_prompt and vxd_ruined_prompt apply them, and vx_condition_prompt links the v62 condition system.
Interiors and details. vint_<class>_prompt and vint_prompt give cabins, dashboards, seats, an open boot (empty or packed), engine bays, wheels and brakes, exhausts, chains and sprockets, underside of a deck, driver cabs, carriages, cockpits, galleys and flight decks, each with matching trim and materials.
Next: v70 adds real-world manufacturer styles by decade. The existing v54 vehicle files (veh_*) are unchanged. 100 random sample prompts, one per line with no blank lines.

Version 68

Condition parts fixed. The part lists for buildings, vehicles, furniture, clothing and statues in the v62 condition system were too specific (a bridge could get a damaged front door, a glove could get damaged sleeves). They now name parts that exist on every subject of that kind, such as entrance, stonework, wheels or running gear, legs or base, main fabric and plinth.
Sea animals fixed. Jellyfish, corals, anemones, starfish and urchins now have their own behaviour family (drifting with the current, waving tentacles, clinging to a rock) instead of sharing one with crabs and shellfish, so a jellyfish is never asked to burrow into sand.
Object condition prompts fixed. They now use tools only, because machines such as tractors belong to the vehicle kind and were being given handles.
World place prompts fixed. world_<name>_place_prompt now reads "Visible nearby are ..." so the sentence always starts correctly.
Also included: the full v57 to v67 gallery of 200 sample prompts, organised in groups, as a separate download.

Version 67

Nine effect domains, each with 8 to 10 effects written twice. The real tier is a physically plausible approximation (what could actually be photographed) and the spectacle tier is the film-style version. Both stay photographic, so an illustrative style from v62 translates cleanly. Domains: magic and spellwork, transformation (partial changes with swirling marks), belief-neutral miraculous and mythic events (parted waters, pillar of fire, staff to serpent and others), sacred light and atmosphere, science (arcs, plasma, lasers, bioluminescence, ferrofluid), cultural and secular (fireworks, lanterns, powder, drone shows), stage tricks and illusions, optical illusions, and atmosphere and light physics.
Effects are built from parts. fx_<domain>_real or _spectacle is what, fx_where_<person|place|sky|object> is where it sits, fx_strength is subtle, moderate or dramatic, fx_colour_real and fx_colour_spectacle are the light, and fx_interact_<target> says how it affects the scene.
Entry points. fx_<domain>_real_prompt and fx_<domain>_spectacle_prompt attach an effect to a ready scene, a person, a place or a stage. scene_with_effect_prompt and fx_prompt pick any. fx_real_prompt and fx_spectacle_prompt pick one tier.
Photographic effects that apply to a whole image (long exposure, haze, lens flare, motion blur, bokeh, light leaks, double exposure and more) are in fx_photo_effect, with fx_photo_prompt to use them on a landscape, city or product.
Extreme weather in 15 events (tornado, hurricane, blizzard, thunderstorm, hailstorm, flood, wildfire, drought, dust storm, ice storm, avalanche, tsunami, volcanic ash, monsoon, fog). Each event has matching sky, land, people, things, animals, plants and light files, so everything in one prompt belongs to the same storm.
Three layers. wx_<event>_sky_prompt is the sky and air, wx_<event>_place_prompt adds the effect on the land, and wx_<event>_living_prompt shows people, things, animals and plants. wx_<event>_prompt combines all three, and wx_severity gives mild, strong, severe and extreme.
Gallery sets. wx_<event>_gallery_prompt makes one image of four panels divided by thin white lines, showing the same place at the four severities. wx_gallery_prompt picks any event.
100 random sample prompts, one per line with no blank lines.

Version 66

31 worlds in three groups. 16 historical settings (ancient Egypt, Greece, Rome, early medieval northern Europe, medieval Europe, Heian Japan, Tang China, Mughal India, pre-1500s Mesoamerica, Renaissance, age of sail, Victorian Britain, the American frontier, the 1920s, 1950s and 1970s), 8 fantasy settings (kingdom, woodland realm, mountain halls, gothic realm, pirate sea, desert kingdom, fairy tale cottage, wizard's tower) and 7 science fiction settings (starship crew, space station, alien world, neon megacity, ruined wasteland, retro-futurist city, robot workshop).
Each world has dress, props, setting and action files, such as world_renaissance_props, with every garment and object described in plain words. All are original archetypes with no named characters, franchises or brands.
Entries. world_<name>_prompt puts a random person in a world. world_prompt picks any, and world_history_prompt, world_fantasy_prompt and world_scifi_prompt pick within a group. world_place_prompt shows a place with no person, and world_frame_prompt adds the v57 framing for a 1:1 image.
Consolidation. A full index (docs/INDEX_v66.txt) lists every family and every version's entry points, and a model guide (docs/MODEL_GUIDE_v66.txt) says which file to start from for each kind of image.
Fixes. Region names in v63 now read correctly in the middle of a sentence ("of the Arctic"), and v64 activity prompts say "Equipment nearby includes" so fixtures like a barre no longer read as something carried.
100 random sample prompts, one per line with no blank lines.

Version 65

24 city and public place types in parts: street, market, railway station, airport, harbour, public square, park, bridge, metro platform, shopping arcade, food street, city at night, old town alley, canal, skyline, university campus, library, museum, stadium, festival ground, rooftop terrace, bus stop, building site and residential neighbourhood.
Each type has layout, ground surface, detail, activity and signage files, such as city_market_detail, assembled by city_<type>_prompt. Signage is always generic (painted names, number plates, route numbers) with no brands.
Shared files. city_style (12 building styles from stone and tile to glass towers), city_time, city_weather and city_crowd suit any place, and times and weather are written so they never contradict each other.
Entries. city_prompt picks any place, city_<type>_empty_prompt leaves people out, city_frame_prompt adds the v57 framing for a 1:1 image, and city_person_prompt and city_person_frame_prompt add a random person to the scene.
100 random sample prompts, one per line with no blank lines.

Version 64

51 activities in three groups. 18 hobbies (gardening, pottery, knitting, sewing, painting, woodworking, baking, birdwatching, photography, fishing, camping, chess, reading, model building, calligraphy, beekeeping, dance classes, stargazing), 21 sports (football, basketball, running, swimming, cycling, tennis, climbing, skiing, surfing, boxing, gymnastics, rowing, archery, cricket, yoga, skateboarding, martial arts, sailing and more) and 12 kinds of music making (violin, guitar, piano, drums, woodwinds, brass, choir, folk instruments, DJ, orchestra, busking, bagpipes).
Each activity has clothes, equipment, setting and action files, such as act_rock_climbing_equip, with every piece of equipment described in plain words.
Entries. act_<activity>_prompt places a random person in an activity. act_prompt picks any, and act_hobby_prompt, act_sport_prompt and act_music_prompt pick within a group. act_frame_prompt adds the v57 framing for a full-body 1:1 shot.
Without a person. act_gear_prompt lays equipment out on a plain surface and act_scene_prompt shows the place for the activity.
100 random sample prompts, one per line with no blank lines.

Version 63

Regional dress for 30 regions across every inhabited continent (Japan, Korea, China, India, South Asia, South-East Asia, Central Asia, the Middle East, the Arabian Peninsula, Turkey and the Caucasus, North Africa, West, East and Southern Africa, Ethiopia and the Horn, Amazigh regions, Northern, Central, Mediterranean, Eastern and British Isles Europe, the Balkans, Northern Eurasia, the Andes, Mexico and Central America, North America, the Arctic, the Caribbean, the Pacific Islands and Australia).
Each region has its own garment, headwear, adornment, hairstyle, setting and occasion files, such as culture_west_africa_garment. Every garment is described in full (what it is made of, how it is worn), and every item is neutral about who wears it.
Entries. culture_<region>_prompt builds a person in the dress of one region, with matching hair, place and occasion. culture_prompt picks any region. culture_frame_prompt adds the v57 framing so the whole person is in a 1:1 frame.
Without a person. culture_garment_prompt shows a set of clothing laid out on a plain surface, and culture_scene_prompt shows a place and occasion. Mixed lists such as culture_garment, culture_headwear, culture_adornment, culture_hairstyle, culture_setting and culture_event can be mixed with any other file.
Everything is described as everyday or festive dress, with no sacred regalia and no stereotyped costume. 100 random sample prompts, one per line with no blank lines.

Version 62

Art and media styles. medium_name has 38 fully described media (oil, watercolour, woodblock, mosaic, stained glass, embroidery, paper cut, clay, marble, pixel art, comic, technical drawing and more) with medium_palette, medium_finish, medium_subject and the entry medium_prompt.
Product and packaging shots with no brands. prod_item (30 described products), pack_type (20 described packages), pack_label (plain, hand-written or stamped labels), plus set, light and angle files, in prod_prompt and pack_prompt.
Signage helpers. sign_type, sign_lettering, sign_message (30 neutral words) and sign_prompt give readable signs with chosen wording.
The condition system. Eight subject kinds (building, vehicle, furniture, clothing, object, sign, book, statue) each have ten condition families (damage, missing pieces, wear, dirt, decay, graffiti, repair, fine quality, cheap quality, faded, burnt) at three strengths (light, moderate, heavy). Each file names real parts of that kind of thing, such as cond_vehicle_decay_heavy (doors, windscreen, mudguards).
Using it. modify_<kind> picks any condition, cond_<kind>_<family> picks a strength, cond_<kind>_<strength> picks a family. condition_<kind>_prompt adds it to a ready subject, and condition_prompt and condition_ruined_prompt pick any kind.
100 random sample prompts, one per line with no blank lines.

Version 61

Behaviour by animal type. 19 behaviour families (predators, grazers, primates, small mammals, marine mammals, bats, marsupials, raptors, waterbirds, songbirds, parrots, game birds, snakes, reptiles, amphibians, fish, invertebrates, bugs, prehistoric animals) each have their own behaviour, group, young and close-up detail files, such as ani_raptor_behaviour, so a bird is never asked to graze and a fish never walks.
Animal scenes. ani2_<class>_prompt pairs an animal from the v54 lists with a group, a behaviour and a habitat scene from ani_<family>_scene. ani2_prompt picks any class.
Young animals and close-ups. ani2_young_prompt shows cubs, chicks, calves and larvae. ani2_closeup_prompt shows the eye, paws, feathers, scales or other detail of a named animal.
Farm and domestic animals. 14 farm animals each with a full description (farm_<animal>), farm_activity, and farm_prompt for farm scenes.
People with animals. farm_interaction has ten ways a person can work or bond with an animal. farm_person_prompt, farm_person_frame_prompt (framed for a full-body 1:1 shot) and pet_person_prompt (dogs and cats) combine a random person with an animal.
100 random sample prompts, one per line with no blank lines.

Version 60

Landscapes in parts. 28 landscape types (mountain range, alpine lake, desert dunes, rocky desert, rocky and tropical coast, temperate forest, rainforest, savanna, steppe, tundra, glacier, volcano, canyon, waterfall, river valley and delta, wetland, farmland, rolling hills, moorland, fjord, island archipelago, salt flat, geyser field, cave, bamboo grove, vineyard valley, dune coast). Each has its own feature, vegetation, water, foreground and scale files, such as land_tundra_feature, assembled by land_<type>_prompt.
Shared parts. land_season, land_time, land_sky, land_clouds and land_night_sky work with any landscape. Seasons, times and skies are chosen to suit the scene, and a cave is lit by a lamp instead of the sky.
Flora by biome. Eight biomes (alpine, desert, tropical, temperate, grassland, wetland, coastal, arctic) each have about eight fully described plants and trees, in flora_<biome>, with flora_prompt choosing a group of plants in a season.
Rock formations and water features. land_rock_formation (arches, columns, hoodoos, mesas, slot canyons and more) and land_water_feature (waterfalls, rapids, hot springs, tarns and more) can be added to any scene. land_close_prompt and land_night_prompt use them on their own.
Entries. land_prompt picks any landscape, land_<type>_prompt makes one type, and land_frame_prompt frames a landscape for a 1:1 image with the v57 framing files.
100 random sample prompts, one per line with no blank lines, show every landscape type, flora, rock formations, water features and night skies.

Version 59

Professions and roles. 120 jobs in 7 sectors: health and care, trades, farming and food, public service and transport, office and professional, creative and performing, service and hospitality, and spiritual and ritual roles. Each has its own parts, job_<name>_outfit, _tool, _workplace and _action, and an assembled job_<name>_prompt.
Spiritual and ritual roles from many traditions, described neutrally and respectfully as jobs, with their clothing, tools, places and actions. They include a rabbi, cantor, Anglican, Catholic, Orthodox and Protestant clergy, Methodist preacher, imam, muezzin, Buddhist monk, Zen priest, Tibetan lama, Hindu priest, Sikh granthi, Shinto and Taoist priests, druid, shaman, nun, pagan priestess, vodou priest, Sufi mystic, yogi, pilgrim and church organist.
Pick by job or by sector. job_prompt picks any profession, job_sector_<sector>_prompt picks one sector (for example job_sector_spiritual_prompt), and job_<name>_prompt makes one job. job_names and job_sector_<sector>_names list the jobs.
Put a person in the role. job_person_prompt adds a short person description to a random job, and job_frame_prompt also frames it as a whole-body 1:1 image with the v57 framing files.
100 random sample prompts, one per line with no blank lines, show professions from every sector, including the spiritual roles, with and without people.

Version 58

Outerwear, hats, bags, scarves, gloves and belts are now built from parts. 16 outerwear types (trench coat, parka, peacoat, puffer, bomber, denim and leather jackets, overcoat, rain poncho, cloak, anorak, windbreaker, duffle coat, quilted vest, blazer, kimono-style jacket), 12 hats, 11 bags, 7 scarves and wraps, 6 gloves and 6 belts.
Each item is fully described. A fixed description of the shape, then a material, colour, pattern, lining, closure, pockets, collar, cuffs, hardware and a condition are chosen separately, so every result is different. Shared pools sit in wear_<category>_<part> and special cases (a poncho has no cuffs, a suitcase has wheels, a turban has no band) have their own files.
Use them whole or by part. wear_prompt picks anything, wear_outerwear_prompt, wear_hats_prompt, wear_bags_prompt, wear_scarves_prompt, wear_gloves_prompt and wear_belts_prompt pick one category, and wear_<category>_<type>_prompt makes one type (for example wear_outerwear_trench_coat_prompt). wear_colour, wear_pattern and wear_condition can be used on their own.
Fix. frame_animal_quad_prompt now reads naturally (each panel is introduced by its corner and then the full animal prompt).
100 random sample prompts, one per line with no blank lines, show every category of wearable item, products and framing.

Version 57

Framing and clarity pack. New frame_* files wrap any subject for a square 1:1 image. frame_clarity is a short opening line that asks for a sharp, clean result (replace its four lines with your own favourite), frame_ratio sets the square frame, and composition files describe how to show each kind of subject completely.
People are shown whole. frame_person_full and frame_group_full say head to toe with space above the head and below the feet and nothing cut off at any edge. Ready-made entries are frame_person_prompt, frame_person_prompt_short, frame_person_age<band>_prompt (nine age bands), frame_group_prompt, frame_outfit_prompt and frame_face_prompt for close portraits.
Subjects get a composition that suits them. Products and tableware are centred on a plain backdrop, rooms and buildings are shown whole with straight lines, vehicles in three-quarter view, landscapes wide, animals and pets with margin around them, insects as macro shots, food at a 45-degree angle. Entries are frame_accessory_prompt, frame_furniture_prompt, frame_tool_prompt, frame_room_prompt, frame_building_prompt, frame_vehicle_prompt, frame_landscape_prompt, frame_animal_prompt, frame_pet_prompt, frame_insect_prompt, frame_food_prompt, frame_table_prompt and frame_plant_prompt, with frame_prompt choosing any of them.
Multi-panel images. frame_animal_quad_prompt builds an image of four different animals in four panels divided by white lines. frame_feature_grid_prompt builds a 2 by 2 grid of close-ups to compare nails or makeup looks.
Model profiles. docs/MODEL_PROFILES_v57.txt explains which files suit natural-language models, short-prompt models and token-limited models.
100 random sample prompts, one per line with no blank lines, use the new frame files across people, groups, outfits, products, rooms, buildings, vehicles, animals, food and plants.

Version 56

Grammar pass. Every library line was checked for a/an (for example a Oxford shoe is now an Oxford shoe, A alien jumpsuit is now An alien jumpsuit). 4,093 lines in 712 files were corrected. References such as __file__ are never touched.
Folder edition. A new, optional download, Guyanonymous_Wildcards_Browse_v56.zip, holds the same 16,501 files grouped into 12 family folders (people, clothing, accessories, body, pose_camera_style, places, groups, objects, animals, nature_buildings_vehicles, food_tableware, other). References are rewritten as __family/name__ and all of them resolve. Use it with tools that support folder paths. The flat folder stays the main edition and works everywhere. A folder edition will be included in every new version.
Quality check. The people prompts were rechecked and fewer than 1% show any problem. The remaining cases are rare mixes such as an older age with a youthful word.
100 random sample prompts, one per line with no blank lines, show people in random clothing, pose, setting and photographic style, period clothing, costumes, groups, outfits, animals, plants, buildings, vehicles, food, tableware, rooms and tools.

Version 55

Food in parts. 8 food categories (cheese, fruit, vegetables, bread, desserts, whole dishes from many cuisines, drinks, snacks) with 100+ foods, each fully described (for example Swiss cheese, which is a pale yellow cheese with large round holes). food_<category> lists them, food_<category>_name gives just the names, and food_<category>_prompt adds a state, a garnish, a surface and light that suit the food. food_prompt picks any category.
Tableware in parts. 8 types (dinner plate, bowl, cup, glass, cutlery set, teapot, serving platter, jug), each with shape, material, pattern and finish files (ware_plate... style names such as ware_bowl_material), assembled by ware_<type>_prompt and ware_prompt.
Table settings and meals. table_setting_prompt builds a whole table (layout, cloth, plate and glass materials, a dish and light). meal_prompt picks a dish, a table setting or any food.
Consolidation. New docs/INDEX_v55.txt (every family of files with counts and ready-made prompt files), docs/QUICK_START_v55.txt (how to use the library in five minutes) and docs/MODEL_NOTES_v55.txt (which files suit natural-language models, short-prompt models and token-limited models).
Regression run. All 166 entry points were expanded 15 times each with no errors, no unresolved references, no empty output and no { } | characters. The v51 quality checker found no problems in the new food and tableware prompts.
100 random sample prompts, one per line with no blank lines, show every food category, tableware, table settings, animals, vehicles, buildings, and random humans with random clothing, pose, setting and photographic style.

Version 54

Animals, organised by kingdom and class using common English names. About 2,700 animals from your list sit in 36 classes (dogs, cats, big cats and wild cats, bears, hoofed animals, primates and apes, rodents and rabbits, bats, marsupials and egg-laying mammals, birds of prey and owls, waterbirds and seabirds, parrots, game birds, songbirds, snakes, lizards, turtles, crocodiles, frogs, salamanders, sharks and rays, fish, insects, spiders, centipedes, crustaceans, molluscs, jellyfish and corals, starfish and urchins, worms, prehistoric animals and more). animal_<class> lists the names, animal_<class>_prompt makes a full animal prompt, and animal_kingdom_<name> lists the classes.
Insects have their own list (animal_insects, animal_insects_prompt). Aquatic animals are split into salt water and fresh water (animal_aquatic_salt, animal_aquatic_fresh and their _prompt files, with animal_salt_<class> and animal_fresh_<class> lists), and animals that live in both appear in both.
Every animal prompt is assembled from parts. A view, a pose, a habitat and a detail are chosen from pools that match the kind of animal, so a shark is never placed in a kitchen and a crab never waves tentacles. The pools are in animal_kind_<kind>_view, _pose, _habitat and _detail.
Nature in parts. 40 plants, trees, flowers and crops (plant_names), each with a described look, a season or light (plant_state) and a habitat that suits it, assembled by plant_prompt. nature_prompt picks a plant or an animal.
Architecture in parts. 9 building types (house, apartment block, temple, castle, bridge, skyscraper, barn, lighthouse, market hall), each with its own part files such as arch_house_roof, arch_house_walls, arch_house_door and arch_house_windows, assembled by arch_<type>_prompt and arch_prompt.
Vehicles and machinery in parts. 12 types (bicycle, car, bus, train, boat, aircraft, tractor, excavator, crane, motorbike, cart, sled), each with part files such as veh_car_body, veh_car_finish, veh_car_wheels and veh_car_extra, assembled by veh_<type>_prompt, veh_prompt and machinery_prompt.
100 random sample prompts, one per line with no blank lines, show animals from many classes, salt and fresh water life, insects, plants, buildings and vehicles. Descriptions use no colons, and the library contains no { } | characters.

Version 53

Wardrobe by occasion. 20 occasions (professional office, job interview, clinical workplace, busy kitchen, workshop or building site, wedding guest, formal evening, funeral, graduation, place of worship, street festival, team sport, mountain hike, freezing weather, heavy rain, warm beach, long journey, relaxed day at home, stage performance, night out) each have parts: occasion_<name>_fabric, _top, _bottom, _onepiece, _layer and _extra, plus _top_dressed, _bottom_dressed and so on (colour and fabric already added), and assembled outfits in occasion_<name>_outfit.
Use them whole or by part. occasion_<name>_outfit_prompt gives a full outfit, occasion_<name>_top_prompt, _bottom_prompt, _layer_prompt and _extra_prompt give one part, occasion_prompt picks any occasion, and occasion_person_prompt puts a random person in one. Garment names carry no fabric or colour words, so the parts never repeat each other, and trousers, shorts and overalls say a pair of.
Bad taste, in four kinds. badtaste_clothing, badtaste_tattoo and badtaste_jewellery each have _ugly (clashing, outdated, ill-fitting, badly made), _tacky (kitsch, gaudy, novelty), _crude (rude sayings, rude logos, jokes worn in the wrong place) and _gross (stained, mouldy, filthy, scabbed or tarnished). Each is built from parts, such as badtaste_clothing_crude_saying and badtaste_clothing_crude_logo.
The crude section covers rude sayings and logos on clothing (badtaste_clothing_rude_saying_prompt, badtaste_clothing_rude_logo_prompt), rude tattoos and rude jewellery. Swearing is allowed. It has no hate speech or hate symbols, no slurs or racist content, nothing illegal, and no brands or real people. badtaste_prompt, badtaste_person_prompt and badtaste_all_three combine them.
No colons in item text. Descriptions that used Name: description now read as natural phrases, for example Swiss cheese, which is a white cheese with random holes, At the front, ... instead of Front: ..., and . before a new sentence. This changed 139,755 lines in 1,789 files and left no : in the library.
100 random sample prompts, one per line with no blank lines, show every occasion, every bad-taste kind, and random people in random clothing, pose, setting, photographic style and composition.

Version 52

Groups and relationships. Ten relationship types (couple, parent and adult child, three generations, siblings, friends, colleagues, team with lead, mentor and mentee, classmates, neighbours) each have a group_<relationship> file that builds a group of 2 to 8 adults, described from left to right, with a relationship line, an interaction and an optional clothing theme. group_prompt picks any relationship.
Age rules by relationship. By default the relationship decides the age spread: parents are 19 to 45 years older than their adult children, three generations are spaced 20 to 40 years apart, siblings are within about 15 years, couples are usually close in age with some wider gaps, mentors are 12 to 38 years older, teams have a slightly older lead. Everyone is 18 to 120.
Age modes. Every relationship has three overrides: group_<relationship>_any_ages, group_<relationship>_same_decade and group_<relationship>_multigenerational, plus matching group_prompt_any_ages, group_prompt_same_decade and group_prompt_multigenerational.
People restricted to one age bracket. any_person_age<18_24 ... 95_120>_w_clothes and _short are the random person, limited to a single bracket (the groups use the short version), built by cloning and rewiring the 15 files in the person chain.
Settings that agree with themselves. setting_exterior now uses only outdoor places (116 indoor places such as foyers, garages, salons and laboratories moved to setting_interior_place), and replaces independent light and weather files with one setting_conditions line where time of day, season and weather agree: no snow in summer, no stars by day, no rainbow at night, hot places stay out of freezing weather and cold places out of heat.
Fixes. 291 clothing entries that read Name: "no clothing,description" in the age and body clothing files are now plain descriptions, and a mid-sentence label in the three-generations intro is gone.
100 random sample prompts, one per line with no blank lines, show every relationship and age mode, single-bracket people, outdoor settings and random people in random clothing, pose, setting, photographic style and composition.

Version 51

Prompt-quality tools. _tools/v51_check.py samples any entry point (people, poses, settings, rooms, furniture, appliances, liquids and more) and reports contradictions, grammar slips and item mismatches. Across 17 entry points the rate of prompts with a problem fell from 6% to under 1%.
Repeated words are gone from the wordlists: 7,881 lines in 539 files had slips such as flannel flannel shirt, straight straight hair, Micronesian Micronesian and They have have. v51_build.py now fixes these on every build.
Grammar fixes in the v48 to v50 files. Rooms say in the industrial style instead of in a industrial style, grouped furniture says an armchair not a armchair, labels such as Extras: became sentences (It also has ...), and size wording such as Its size is a standard tall size is fixed.
Hair conflicts removed. Hairline and hair-volume entries that described baldness (clean shaved bald hairline, sparse balding patches hair, mostly bald crown) no longer appear next to a full hairstyle.
Tools are now random assemblies. 20 tool families (agriculture, carpentry, cooking, plumbing, welding, textiles, painting, pottery, mining, surveying and more) gained tool_<trade>_random and tool_<trade>_phrase: a tool with its own material, finish, feature and condition. The existing Holding ... prompts are unchanged.
100 random sample prompts, one per line with no blank lines, show the tool assemblies, the repaired rooms and furniture, and random people in random clothing, pose, setting, photographic style and composition.

Version 50

Furniture, fixtures and appliances are now broken into parts. A bed has a headboard (or none) and a footboard (or none), a frame, feet, a mattress, bedding and a size, and each part has its own file (furn_bed_headboard, furn_bed_footboard, furn_bed_frame and so on). furn_bed_built assembles a whole bed from them.
Matching or independent. furn_bed_boards gives a headboard and footboard that match (same material, design and finish, with the footboard usually lower), including no-board and headboard-only cases. Use furn_bed_headboard_prompt or furn_bed_footboard_prompt to apply either part on its own with its own style.
Not just beds: about 56 items are split into parts, including chairs, armchairs, sofas, tables, desks, wardrobes, bookcases, chests of drawers, bunk beds, toilets, bathtubs, washbasins, showers, sinks, cookers, refrigerators, dishwashers, washing machines, ovens, microwaves, kettles, televisions, lamps, chandeliers, mirrors, curtains, rugs, vases and more. Each has furn_<item>_<part>, furn_<item>_<part>_prompt, furn_<item>_built, _built_short, _built_phrase and _prompt.
Shared part pools: part_wood, part_fabric and part_metal, plus comp_colour, so a part can take any material or colour.
Group files such as furn_beds_mix_phrase, furn_bathroom_mix_phrase, furn_kitchen_mix_phrase and seat_mix_phrase now draw from the part-built items, so rooms (room_prompt_full) and seated poses get parts-built furniture. furn_built_prompt and v50_entry_points list every entry point.
100 random sample prompts show part-built furniture and appliances, matched and individual bed parts, rooms, and random people in random clothing, pose, setting, photographic style and composition.

Version 49

Every new object is now a random assembly. The v44 to v48 object families (furniture, household and office objects, seats, bottles, perfume, eyewear, shawls, pins and patches, balls, sport gear, adult-only objects, chastity devices and room gear: 89 families) no longer give you only a list of finished sentences. Each <family>_prompt builds a description from components: the type, a material file that belongs to that type (with a colour where the material takes colour), a finish pool chosen by material class (wood, metal, plastic, glass, soft), a feature pool for the category, and a condition pool.
Component files. comp_colour, comp_fin_wood, comp_fin_metal, comp_fin_plastic, comp_fin_glass, comp_fin_soft, comp_fin_hard, 24 comp_feat_<category> files and comp_condition are shared, so you can also use them directly. About 1,100 per-type <family>_<type>_material files hold the materials each type really comes in.
Rooms and poses get randomised furniture too. The _phrase files that rooms, poses and liquids call now draw from the same components, so every room and every seated pose gets a different version of each object.
The finished-line files remain. <family> (premade lines) and <family>_short are unchanged for anyone who wants them; <family>_random is the new assembled version and v49_families lists every family covered.
93 random sample prompts show the assembled objects on their own, rooms built from them, and random people in random clothing, pose, setting, photographic style and composition.

Version 48

Random rooms. 35 room types, each with a brief and a full version: room_<type>_prompt_brief and room_<type>_prompt_full, plus room_prompt_brief and room_prompt_full to pick any room. Rooms describe the space only (no people, no poses) and call the furniture, object and gear files by reference.
House rooms. Living room, bedroom, kitchen, bathroom, dining room, hallway, home office, laundry room, garage, attic, basement, balcony and conservatory.
Work and public rooms. Open-plan office, meeting room, reception, staff break room, library, café, restaurant, bar, hotel room, classroom, workshop, art studio, music practice room, shop, train carriage and spa treatment room.
Fitness and water rooms. Gym, weight-training gym, dance studio, indoor swimming pool hall, yoga studio and sauna, each with its own described gear (gear_gym_cardio, gear_gym_weights, gear_gym_floor, gear_pool, gear_yoga, gear_sauna and more), usable on their own as objects.
Room modifiers. room_size, room_floor, room_walls, room_light, room_condition and room_style (16 regional and period styles) vary every room; room_types lists the room names.
Wording fix. Names that end with a plural word but describe one item (for example "barbell rack with plates") no longer get "a pair of".
100 random sample prompts include rooms on their own, gear on its own, and several random people in random clothing, pose, setting, photographic style and composition.

Version 47

Furniture as objects. furn_tables, furn_storage (credenzas, sideboards, chests), furn_beds, furn_bathroom (toilets, baths, basins), furn_kitchen, furn_office, furn_lighting and furn_decor, plus the v46 seats, each described by shape, parts and material. furn_prompt picks from all groups and furn_all is one combined pool.
Household and office objects. home_cookware, home_tableware, home_textiles (blankets, quilts, throws, towels), home_cleaning, home_electronics, office_supplies, with home_prompt and home_all to mix them. Every item stands alone: no room, person or pose is implied.
Bottles and closures. bottle_* (22 shapes from hip flasks and amphorae to novelty bottles) and closure_* (12 caps, corks, stoppers and seals), each with a _phrase file that can be dropped into a sentence.
Perfume bottles and samplers. perfume_* covers atomisers, flacons, rollerballs, solid compacts, decanters, antique and art-deco bottles, vials, sample cards, spray pens and discovery sets, without brands or labels.
Liquids. liquid_look (colour, transparency, viscosity and extras), liquid_liquor, liquid_medical and liquid_strange describe a liquid by how it looks. liquid_in_bottle and liquid_in_perfume_bottle place a liquid in a fully described container.
100 random sample prompts show the objects on their own, with a person only where a seat is used by a pose.

Version 46

Poses sorted by what they need. pose_free (standing and walking, safe in any setting), pose_sitting (on the ground or floor), pose_lying, pose_leaning (self-supported) and pose_floating never imply furniture. pose_sitting_w_seat, pose_lying_w_surface and pose_leaning_w_support name and fully describe their own seat, surface or support, so nobody sits in empty air. pose_prompt mixes them, and pose_prompt_no_support uses only the safe ones.
First seating and support files. seat_* (24 seats from stools and benches to sofas, swings and wheelchairs), surface_lie_* (13 beds, mats, loungers and lawns) and support_lean_* (14 walls, rails, trunks and pillars), each described by shape, parts and material, and usable on their own as objects. seat_short variants are provided for group prompts.
Interaction files. interact_two, interact_three and interact_object are written by position (the person on the left, the person in the middle, the person on the right) to match group blocks, with _w_seat versions that describe the seat.
Settings kept separate. setting_prompt, setting_interior and setting_exterior describe only the place, the light and the weather. prompt_setting_exterior and prompt_setting_interior now point to them. Person, outfit and object files describe only themselves.
Audit report. _tools/audit/AUDIT_setting_pose_v46.txt lists every file that still contains setting, lighting, pose or interaction words, so older templates can be cleaned up.
100 random sample prompts show seats, settings and poses on their own, and people only where the pose or interaction is the point.

Version 45

Chastity devices split by anatomy. The single acc_chastity family is replaced by chastity_penis_*, chastity_vagina_*, chastity_clitoris_*, chastity_nipple_* and chastity_anus_*, each described as an object (shape, parts, material, lock). chastity_prompt picks from all five.
Merkins and pasties. acc_merkin_* (20 hairpieces from theatrical to couture) and acc_pasties_* (21 decorative covers from tassel to jewelled), described as items only.
Adult toys, objects only. acc_buttplug_*, acc_dildo_*, acc_vibrator_*, acc_adulttoy_* (rings, sleeves, beads, harnesses, cases, kits) and acc_sensation_* (pinwheels, feather ticklers, violet wands, paddles, floggers, crops, canes and similar), each with shape, size, parts, material and finish. No people, no body detail, no acts, no brands.
Full descriptions in every prompt. The _prompt files for the v44 and v45 families now always use the full object description, so models never get a bare name such as "tawse" or "kegel balls"; the _short files remain for when you want them.
Sample prompts show objects on their own. 100 random prompts are included, plus a labelled version, and a revised v44 set that shows the new v44 items on their own.

Version 44

Eyewear sections. New acc_monocle_* (15 forms, from chained to loupe), acc_noveltyglasses_* (22 joke and celebration glasses), acc_sunglasses_* (26 frame and lens styles by era) and acc_glassesadd_* (20 loupes, clip-ons, cords, pads and straps), each fully described.
Shawls and muffs. acc_shawl_* has 25 wraps from lace and pashmina to regional and evening styles; acc_muff_* has 14 hand muffs from court to novelty.
Wearable tech by era. acc_wearabletech_* runs from cassette players and pagers through smart rings and VR headsets to invented far-future devices, with _retro, _current, _near_future and _far_future splits.
Pins, patches and badges. acc_lapelpin_* (24 pins), acc_patch_* (26 patch types), acc_meritbadge_skills_* (22 wholesome skill badges and displays) and acc_meritbadge_adulting (humorous adults-only badges, kept in a separate file).
Earrings rebalanced. Small and delicate earrings (acc_earring_small_*: studs, huggies, small hoops, climbers) and plastic and novelty earrings (acc_earring_plastic_*) are now mixed into acc_earring_prompt, with small ones the most common.
Range audit. Ten more families (bracelet, necklace, ring, brooch, watch, belt, wallet, umbrella, glove, fan) gained _plastic and _novelty ranges, wired into each family's prompt.
Costume teeth, balls and sport. acc_teeth_* (fangs, grills, monster teeth), obj_ball_* (36 balls from ball bearings to bowling balls) and sport_* gear by sport (tennis, baseball, football, ice hockey, cycling, climbing, gym).
Garment fit modifiers. Manual-only fit_* files describe clothing that is too tight, too loose, too short or too long in specific areas (chest, bust, hips, thighs, waistband, sleeves, collar, seams), with fit_random_prompt and severity levels; clinical, item-focused wording.
100 random sample prompts and a labelled version are included.

Version 43

LoRA text removed. Six files carried model-specific LoRA tags and trigger words; all are gone and a LoRA-only file was deleted.
Anatomy as a random system. f_body_vulva is built from outer-labia, inner-labia, clitoris and skin-tone pieces, with the skin matched to the person. m_body_genitals has near-average sizes weighted by realistic odds. Both default to the relaxed state; f_body_clitoris_aroused and m_body_genitals_erect are manual-only files that nothing else calls.
Teeth that follow age and care. Every age bracket now has weighted teeth from perfect to missing, dentures and gold caps, plus teeth_access_high, _mid and _low for dental-care access.
Three new tattoo kinds. tattoo_temporary, tattoo_henna and tattoo_bad (including misspellings described in plain words), with tattoo_novelty_random to pick one.
96 random sample prompts and a labelled version are included.

Version 42

Plain-named character creation. New prompts with names that say what they do: f_create_random_woman, m_create_random_man, any_create_random_person, and fixed identities such as f_create_random_trans_woman, m_create_random_trans_man, any_create_random_intersex_person and any_create_random_nonbinary_person. The any_ files use they/their wording and cover male, female, intersex, trans, non-binary and in-transition adults.
Clothing kept separate from the person. Every prompt comes as person only, _w_clothes, _nude, _topless or _bottomless, and as a full or _short version for token limits. Optional _traits versions add scars and hormone-related traits. About 810 files in all.
Three ways to treat age. Realistic (age drives the look), _mismatched (looks decades older or younger than the stated age) and _unlinked (a fit, smooth-skinned 100-year-old or a wrinkled 18-year-old).
Better old age. Models under-age people over 50, so each age is now stated as a number, an age word and visible features, with a short closing reminder for older ages.
Fewer contradictions. Shaved and bald styles no longer get curl patterns, hair and eye colours come only from the colour files, build words no longer fight with stature, young brackets lose aged-skin lines and old brackets lose youthful-skin lines. A contradiction test script (_tools/v42_contra.py) is included.
Clean-up. Emphasis weights such as (word:1.5) became plain wording, health conditions were removed from the random-person templates (the files stay), blank-line odds were made explicit so tools that strip blank lines keep the intended proportions, and openings such as "The subject is a woman" became "A woman, 34 years old, ...".
Sample prompts for this version are included: about 9,000 completed prompts in random order, covering every category.

Version 41

Text-trigger cleanup. Some prompts began a phrase with a label and a colon, such as "Printed or worked on it:", and image tools printed the words after the colon as text at the bottom of the picture. The artwork and logo lines on keychains, lighters and every other accessory are now plain sentences ("Its surface is decorated with…"), and about 2,500 files across the library had their label-and-colon openers turned into natural clauses.
One banned descriptor removed. A phrase that must not appear in lighter descriptions was replaced everywhere with "sealed fuel reservoir", and the build now checks for it.
Eyewear in the person files. The "She wears…" and "He wears…" eyeglass lines in the woman's, man's and age-bracket files now use the fully described frames, so each frame is explained by its shape, rims, bridge, arms and lenses.
Plastic and novelty lighters and keychains. Both families were missing whole ranges. Lighters now include 20 plastic types (disposable, child-proof, clear, glitter, glow-in-the-dark, marbled, rubberised, utility, electric and more) and 30 novelty types (animal, food, skull, sneaker, camera, telephone, car, rocket, lipstick, robot, mushroom and more). Keychains now include 23 plastic types (acrylic, epoxy-dome, PVC rubber, silicone, foam, bottle-opener, torch, whistle and more) and 30 novelty types (miniature books, shoes, foods, instruments, globes, bells, dice, puzzles, compasses and more). Each is described by shape, parts and material. Brand-style words in the lighter files were replaced with plain descriptions. Files: acc_lighter_plastic*, acc_lighter_novelty*, acc_keychain_plastic*, acc_keychain_novelty*.
Library map. The map page now covers couture, outerwear, extreme-weather, diving and trades tools.

Version 40

Haute couture and silly fashion. About 60 of the strangest, hugest, most surreal and most impractical creations (cage-skirt gowns, a six-metre floating train, a skirt too wide for a door, a dress of living moss, a mirror-tile gown, a balloon dress, a swan-neck gown, a tower headdress, a shoe-shaped hat) and about 30 comic pieces (trousers pulled up to the chest, several pairs of jeans at once, a cowboy hat with udder-like bulges, boots with flip-flop soles, gloves with very long extra fingers). Files: clothing_couture_haute*, clothing_couture_silly*.
Outerwear. About 50 coats, jackets, cloaks, capes and wraps, old and modern, each with cut, collar, closure and fabric. Files: clothing_outerwear*.
Extreme-weather gear. About 45 pieces for polar cold, mountains, desert, monsoon, storm, heat and hazards. Files: clothing_weather_extreme*.
Scuba and diving. About 37 pieces: wetsuits, drysuits, masks, fins, regulators, tanks, rebreathers, the copper hard-hat helmet and an atmospheric suit. Files: clothing_diving*.
Trades tools and equipment. About 235 tools in 17 trades, each described by shape, parts, materials and use, plus hand, power and gear groupings. Files: tool_<trade>*, tool_all*, tool_hand*, tool_power*, tool_gear*.
Purses, and only purses. The purse family no longer holds clutches, totes, satchels, crossbody bags or other handbags; it has 25 true purses (coin, change, frame, drawstring, beaded, a sporran and more). Files: acc_purse*.

Version 39

Fully described eyeglasses, about 50 frame forms with rim shape, bridge, arms, hinges, nose pads, materials and lens tints in plain words. Files: acc_eyewear*.
Fully described collars, about 40 forms from leather and locking collars to chokers, medical collars, ruffs, torcs and ceremonial bead collars. Files: acc_collar*.
An accessory audit. About 2,000 thin lines for unusual forms in 24 families gained a second sentence on build and use, and body-part words that had leaked in from the shoe files (such as "across the ankle" on scarves, ties, gloves, fans and handkerchiefs) were replaced with a part the item really has.

Version 38

A clinical, detached injury overlay of about 225 injury types, from paper cuts and stings to fractures, burns, infections and amputations (body_wound_*), and casts, splints and bandages (acc_medwrap_*).
Condition descriptors for any item, from pristine to ruined (condition_*).
Tie pins, clips, bars and bolo ties (acc_tiehw_*, acc_bolo*); bed-head and messy hair (hair_state_*).
Fuller chastity-device descriptions; fuller descriptions of about 100 regional and formal garments.
Quality files instead of price files for footwear, from cheap to absurdly expensive (acc_shoe_quality_*).

Version 37

Assistive and adaptive items (acc_assist_*); makeup, nails, piercings, body art and skin features (body_*); eyewear; character costume (costume_char_*); subject objects (obj_<subject>_*); buildings, rooms and landscapes (scene_*); activity and interaction poses.
Body-change, cold-weather, formal and regional garments.
Ethnicity tied to skin colour, so a skin colour is always stated; manual tan and sunburn by natural skin tone (skin_tan_*); season and weather outfits (wx_*); the wardrobe role index (wardrobe_*); optional detail files for garment families.

Version 36

All file names unified under one structure (clothing_*, costume_*, obj_*; filename_changes.txt maps every old name to its new one).
Halloween masks and costumes in five quality tiers across 1920s to 2020s; corsets and neck corsets; collars and mouth gags; and a gap assessment (docs/ASSESSMENT.txt).

Versions 33 to 35

Nightwear, robes, undergarments and lingerie in many families, plus shop-style sections of about 1,500 detailed garment descriptions (front, back, coverage, cut, fabric and trim).
Rave, festival and counterculture dress across the 1920s to 2020s in 21 families, from flapper and mod to punk, goth, rave and drag.
Swimwear in many cuts, including odd and extreme styles and fabric that turns translucent when wet.
Non-gunpowder weapons, armour and costume pieces, surfboards of every era, and chastity devices as accessories.
Printed and sewn T-shirt costumes gained extra costume pieces (insignia, props, belts, headwear emblems); onesies gained a novelty-embellishment layer.

Versions 28 to 32

Version 28: one-piece garments (coveralls, zip suits, bodysuits, overalls) and dance, performance and wrestling wear (leotards, singlets, ballet, ballroom, folk dance and more), each with 300 named options.
Version 29: printed-tee prompts rewritten in positive wording (what the shirt is, rather than lists of things to avoid), and adult bodysuit-tee and romper-tee onesies.
Version 30: printed rompers and one-piece zipper leisure suits.
Version 31: sewn onesies (real cut-and-sew garments), 184 creature onesies (every one hooded and carrying the features that identify the animal), and hoods that always match a uniform's headwear.
Version 32: sewn T-shirts and ugly-sweater takes (sweater, cardigan, vest, dress and onesie versions of every uniform).

Version 27

Tee and onesie fixes: tees no longer print the lower garment, no scenery or backdrop on sleeves and shoulders, the outfit is enlarged to the real collar and armholes, and printed onesies are cloth wrapped over every sleeve and leg.

Since version 36 the files follow one naming structure. See the naming guide above, and filename_changes.txt for every old and new name.

Files with SHA256: 74830bb5bbc0e69a12c382191033016d3d0f5e6a5e33fb2dd0156ad2ec7344df