CivArchive
    Preview 44495061
    Preview 44500128
    Preview 44494985
    Preview 44495186
    Preview 44498409
    Preview 44495543
    Preview 44495678
    Preview 44495558
    Preview 44500665
    Preview 44532262
    Preview 44534058

    Kofi!

    btc: bc1qp47qea47q0pku3yvgr04tvwmdr6p9qmfwf6un8

    eth: 0xf118c3ee5Dc4d86A6C61427969C912DAE648401a

    Tipping Buzz⚡ to images i like! so make sure to post some! Like, React, Follow, and all that stuff.

    Changelog

    ..BAKED-VAE v0.3.4.1: experimental version.

    i have been enjoying it, so just thought i would push and let everyone try it as well. i still think, overall, version 0.2 is better. but for strong, dramatic lighting, give this one a try!

    • better backgrounds. in general i have noticed the backgrounds having a bit more detail.

    • lighting behaves a little different. more dramatic, without the aid of extra networks.

    negative things

    • Background details can sometimes work against you if your environment is meant to be simple. Prompting specific background elements can help, but as always, avoid overdoing it. If you want a simple background, make sure to explicitly say so in your prompt—don’t leave it up to chance.

    ..BAKED-VAE v0.2: this version has a slight different style. however, the core focus is still lighting. overall, i have noticed an increase in vibrance in generations, as well as improved lighting.

    ..BAKED-VAE v0.1: i baked a VAE in to make it better for onsite generation, and local gen as well (for those who dont load a VAE separately). in general, lighting and colors are noticeably better! Because of the baked VAE, this is the ideal version for onsite use.

    illustrious-XL v0.1: First illustrious version. i tried to keep the style similar to the pony version, but ultimately wanted better lighting, and more flexible style overall. Will probably keep updating until im happy, but ive switched focus lately to Loras over checkpoints and photos in general. ill get to it eventually.

    v0.8.1: the better version. this version was supposed to be 0.8 (i have no idea how i uploaded the wrong version. sleepy brain? oops). better results without style loras, but still compliments them very well. still looking for feedback.

    v0.8: my first (public) pony merge. looking feedback to improve the newer version that’s in the works.

    Concat Positive Prompt:

    HDR, 8K, volumetric lighting, dramatic shadows, cinematic lighting,  raytraced, Mysterious Shadows, mist, bloom, realistic shading, bloom,

    Prompt:

    Prompt lol

    Neg

    worst quality, low quality,

    Settings i Use (feel free to experiment):

    • sampler: euler_ancestral_cfg_pp

    • steps: 20

    • size: just use your preferred XL size

    • batch count: 4-20 (or 1 if I'm extra impatient)

    • cfg scale: 4

    Have feedback or requests? let me know!

    Looking forward to your reviews!

    Please consider to support me so i can upgrade my struggling 3060, to a 40xx

    btc: bc1qp47qea47q0pku3yvgr04tvwmdr6p9qmfwf6un8

    eth: 0xf118c3ee5Dc4d86A6C61427969C912DAE648401a

    Description

    FAQ

    Comments (48)

    TekeshiXDec 9, 2024· 3 reactions
    CivitAI

    How exactly did you merge Pony with Illu/NoobAI?
    Here's what I tried so far without any luck:
    1. Merged a Pony checkpoint with a NoobAI/Illustrious checkpoint using all the methods available and all the possible existing mergers (Supermerger, Untitled Merger, Model Mixer, MBW, Merge Board) = fail

    2. Extracted the LoRA from a Pony checkpoint and used it with NoobAI/Illustrious model = fail

    3. Extracted the LoRA from a Pony checkpoint and injected/merged it into NoobAI/Illustrious model with different weights = fail.

    killedmyself
    Author
    Dec 9, 2024

    thank you for your comment because i was gonna try some of these eventually. nah i merged other checkpoints to get as close as i could to my pony checkpoint. IF i actually can figure out how to do so with merging between the two, that’ll probably be the next version of this.

    TekeshiXDec 9, 2024

    @killedmyself , np, I can say I tried literally everything to merge pony with illu/noobxl and nothing worked so far. The "best" result (attempt) was with DARE merge from Untitled Merger at 50/50, but still the quality is so bad, let alone all the images look washed out even with VAEs.

    killedmyself
    Author
    Dec 9, 2024

    @TekeshiX it can never be easy eh lol i do like your idea with extracting the lora tho. lots of pony lora still work with illu/noobai checkpoints, so maybe i’ll just make my other pony merge a a lora and call it at that lol

    Reep1611Dec 11, 2024

    @killedmyself So it isn´t actually a merge with pony, but an Illustrious model that is merged to be as close to pony as you could manage if I get this right? Damn fine work by the way. It does some really awesome generation and is so far one of the easier to figure out models I have tried from what little experimenting I have done so far.

    killedmyself
    Author
    Dec 13, 2024· 1 reaction

    @Reep1611 yes, i hope the title isn’t too confusing lol glad you’re enjoying the checkpoint though!

    illuminaiDec 11, 2024· 3 reactions
    CivitAI

    Very nice iteration of Illustrious, nice job!

    killedmyself
    Author
    Dec 11, 2024

    thank you !

    Reep1611Dec 11, 2024· 4 reactions
    CivitAI

    I really like this Checkpoint. It does some awesome generation on your provided basic settings and so far, from what little experimentation I have done, looks to be relatively easy to promt. I especially like how well it manages lighting. It does some awesome stuff there.

    MegasherruDec 24, 2024· 4 reactions
    CivitAI

    This is my ALL time Favorite Model for illustrious! The results are amazing!

    killedmyself
    Author
    Dec 24, 2024

    i’m really glad you enjoy it (:

    hoxv00186Dec 25, 2024· 3 reactions
    CivitAI

    Its reappearance of lora is amazing, a great creation

    guzzzlyDec 30, 2024· 4 reactions
    CivitAI

    i get a grey filter as the gem finishes... help..

    how do i fix?

    killedmyself
    Author
    Dec 30, 2024· 1 reaction

    sounds like you need a vae. you can find plenty here on civit, i use SDXL vae. just place it in your vae folder and then restart your gui. if you use a1111, you have to go to settings and select it. if comfy, just reload then select from your vae node. idk about other ones. hope this helps!

    mommymiaJan 3, 2025· 4 reactions
    CivitAI

    Easily my favorite checkpoint so far. Always produces incredible results.

    killedmyself
    Author
    Jan 3, 2025· 2 reactions

    thank you (: glad you’re enjoying

    LinklaineJan 7, 2025· 5 reactions
    CivitAI

    Invoke have troubles with this models or vae s, even with oversaturated vaes it outputs grey and blurry images, but on fp16 vaes (there is fp16 sdxl vae yeah) it works just fine. edit: i could avoid troubles if i just read properly yeah

    zeroapprentice375Jan 10, 2025· 5 reactions
    CivitAI

    Its a great model, compatible so far with my usual pony based loras, results are great, and is easy to prompt

    killedmyself
    Author
    Jan 10, 2025

    glad you're enjoying

    mystifyingJan 17, 2025· 6 reactions
    CivitAI

    ty ty ty... for the love

    love the results i get with your model

    tried about ten so far and this one is my fav

    i know i am not doing illus prompts but your model is forgiving

    killedmyself
    Author
    Jan 17, 2025· 1 reaction

    glad you enjoy. great images so far! hope to see more (:

    OttoDAIJan 22, 2025· 9 reactions
    CivitAI

    Thanks to this checkpoint + NTR.. I no longer wish to play with Pony - 10/10

    AndroidXLFeb 2, 2025· 3 reactions
    CivitAI

    is therre a reason its not available in the civit online generator?

    killedmyself
    Author
    Feb 2, 2025· 1 reaction

    according to support, sadly its not up to me. maybe one day, but as of right now, they said they cant accept every checkpoint for storage reasons

    7181678Feb 18, 2025· 1 reaction

    @killedmyself  so sad this isnt on civit when it looks so good

    jbsbiomeFeb 18, 2025· 3 reactions
    CivitAI

    how to have the face not be so flat, the body has beautiful lighting and rendering but the face looks like a drawing,i'd still like to maintain anime shape though, and hopefully more detail in the face, thanks in advance. lovely model btw.

    killedmyself
    Author
    Feb 18, 2025· 1 reaction

    flat in terms of lighting, color, or physically?

    i could suggest using lighting tags if you mean lighting. i always use tags like "partially illuminated, (backlighting), Intricately designed, Mysterious Shadows," as well as other tags to get stronger shadows and create contrast. i could suggest adding "flat color" and other similar tags to negative, that could help. you can have a look at any of my prompts to see what all tags i use for lighting.

    i hope this helps a bit

    jbsbiomeFeb 18, 2025

    @killedmyself flat as in not 3D/2.5 like the body so physically? i used your lighting prompts earlier, using ForgeUI, its just the body will look semi-real like your images but the face doesn't have that kind of depth and even a bit of lineart under the chin, still new to this, thanks.

    killedmyself
    Author
    Feb 18, 2025· 1 reaction

    @jbsbiome this isnt something ive noticed before, but my best suggestion would be to add things you are trying to negate to negative, like "lineart," and focus on lighting, as lighting and shading is the main difference between flat 2d and 2.5d/3d you know? if you need help with recognized tags to help with this, i always use https://danbooru.donmai.us/wiki_pages/tag_groups

    if you checked out my prompts i can also suggest adding detail tags to you prompt like "photorealistic, beautiful detailed eyes, detailed skin,"

    im not sure what else it could be beyond these things ive suggested

    hope this helps, sorry i cant help more

    jbsbiomeFeb 18, 2025· 1 reaction

    @killedmyself thank you i will try all of your suggestions,my machine is slow, i really appreciate it, love the images i've gotten so far and the prompt adherence is wonderful!

    killedmyself
    Author
    Feb 18, 2025· 1 reaction

    @jbsbiome thank you, glad you enjoy it! good luck soldier

    jbsbiomeFeb 20, 2025

    @killedmyself just wanted to say been getting better results by switching to close-ups, the new problem i'm having it it seems impossible to have front lighting, is there a prompt that works for that?

    killedmyself
    Author
    Feb 20, 2025

    @jbsbiome glad you're getting better results!

    for lighting, you can try:

    minimize any tags having to do with shadows, especially if youre doing close ups now. for shadows to fall on the face, indirect lighting must be used, so you wont get front facing lighting.

    try having a look at lighting tags here, you may find some usable tags: https://danbooru.donmai.us/wiki_pages/tag_group%3Alighting

    my suggestion would be: if youre still using a lot of tags from my prompts, get rid of "partially illuminated, (backlighting)," and "dramatic shadows," and (most) other lighting tags. just remove a couple at a time, and see if you get closer to the desired result!

    jbsbiomeFeb 20, 2025· 1 reaction

    @killedmyself  thank you again, i will try that!

    Myeong_Feb 22, 2025

    @jbsbiome I’ve been using her AI for awhile, I loved the previous version! When I got this I was faced with the same problem, all I did was add detail qualities into the face detailing prompt.

    I looked at her prompts and decided to add ultra-detailed and photorealistic and that instantly fixed the issue I was facing with the face.

    jbsbiomeFeb 23, 2025

    @Myeong_  thank you i'll try that, strangely the new issue i'm having is everything is looking realistic except for the hair, lol. love this model though.

    Myeong_Feb 23, 2025

    @jbsbiome can you show me your prompt?

    jbsbiomeFeb 23, 2025

    @Myeong_  it's a bit messy:
    HDR, 8K, ultra-detailed, photorealistic, depth of field, outdoors, high above in the clouds, walking on clouds, leaning against roman column,BREAK, 1girl, blue skin, short hair, rainbow hair, wearing ancient greek clothes, white toga, thin rope around waist, cute face, smile, beautiful detailed eyes, detailed skin, (detailed hair), detailed face, lush full long eyelashes, looking at viewer, sexy pose, blue eyes,(finely detailed face),

    <lora:Semi-realistic_IL_v2:1>

    neg: outline, lineart, glowing eyes, long body, lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, worst quality, bad quality, bad anatomy, jpeg artifacts,((signature, watermark, username)), lipstick, ((selfie:1.5)), shiny skin,
    ----------
    i get some better hair results if i use <lora:Dreaming_Death_Horror_Scifi_Illu:0.7> dre4mingde4th
    but it changes other things too much so i've stopped using it, i'm sure there's a lora out there that can make the hair more realistic, i've just not found it yet.

    Myeong_Feb 23, 2025

    @jbsbiome try this instead; I use this from the creator:

    HDR, 8K, high contrast, masterpiece, best quality, amazing quality, very aesthetic, high resolution, ultra-detailed, absurdres, newest, scenery, photorealistic, depth of field, 1girl, detailed, solo, detailed hands, 1girl, skinny, solo, blue skin, short rainbow hair, short hair, detailed face, (finely detailed face), slim face, blue eyes, detailed eyes, detailed skin, (detailed hair), long eyelashes, wearing wearing ancient greek clothes, white toga, thin rope around waist, smiling, sexy pose, outdoors, blue sky, clouds, walking on clouds

    ________

    Its not the best, but doing things this way gives me way better results.

    jbsbiomeFeb 23, 2025

    @Myeong_  very little difference for me, in some ways worse and now the background is wrong which is why i put it at the start, hair is the same. thanks though.

    Myeong_Feb 23, 2025

    @jbsbiome hmmm, I'm sorry it didn't work. Maybe you could try with some vaes and such? might help, even finding loras to help strengthen a few of the aspects you have.

    Myeong_Feb 23, 2025

    @jbsbiome I also HIGHLY suggest looking at getting a face detailer, I can send the workflow I use if you would like? 

    jbsbiomeFeb 24, 2025

    @Myeong_ i use a face detailer, thanks, i've tried the exact prompt on another checkpoint and the hair turned out more real so it's just an aspect of this one, i will try to adjust what i'm doing cause i really love this one

    Myeong_Feb 22, 2025· 5 reactions
    CivitAI

    I’ve been a long time user of your stuff, this is top tier 🩷 keep up the good work! The only issue I’ve faced is any water prompts, was trying but either way I’ve been having a lot of fun.

    Thank you 🌸🩷

    killedmyself
    Author
    Feb 22, 2025· 1 reaction

    glad you enjoy! thanks for using (:

    killedmyself
    Author
    Mar 12, 2025· 6 reactions
    CivitAI

    on site generation available at last!

    DunkingOnSoulMar 14, 2025· 1 reaction

    <3

    killedmyself
    Author
    Mar 14, 2025

    fixed with newest version