V3!!

Use to produce neotenous adults (adults with youthful features), OR to counterbalance other lora's that make your outputs ugly (I also do this a lot).

Am finding SOME prompts (artistic shot for eg) produce a little blurrier images with this one despite improved general realism. I found a really small touch of around 0.2 of Hyperrealism 3D (linked in resources) tends to solve this when it arises (it doesn't always)

It's strongly/heavilty trained so you can use lower weights and still get a decent effect. Still has an influence as low as 0.4. It appears I was lucky with the mixed style dataset that this was not wildly overtrained. Which is good, because it means the tags have heavy impact. In any case, yeah, either use a little extra detail juice, or lower weights, all good. Some tags you'll need lower weights. I tend to use CFG in the 5.5 ish region.

Read through to the end if you want extra prompting options.

Updated dataset with 118 hand curated and hand labelled pictures trained for 45 epochs. 12 Removed from the last dataset and many added. Improved labelling with lighting, pose and some particularly facial expressions added (mostly eyes and types of smiles, more subtle than expression lora's like amused, shy smile, warm smile etc, see below for a list), as well as a few more sci-fi/fantasy pictures. Still at least 60% real pictures + synthetic from good PXL tunes and Flux.

I found the few fantasy adds really did help the lora play well with other loras, so more stuff added there, including some cyberpunk. Only intended to help it play well with other loras, not intended to be stand alone for fantasy/sci-fi. I may add more of these in a future version, as you have to prompt around the limited examples a little.

Standard prompt is:

Neoteny, with any of (soft jaw, small nose, plump lips, small mouth, wideset eyes, large eyes) optional.

Adding very on most of those, ie lips, mouth, nose, are also tags.

Instead of just barely for extra cute 18-20 yo's, use (barely, cute) instead. Added 'awkward face' for those neotenous faces that are a little odd (big forehead, chubby cheeks, awkward teeth etc), so something to vary the round face/narrow face divide. This will produce wonky teeth sometimes unless you neg prompt.

Should also respond better to lighting prompts, such as outdoors, indoors, dark lighting, bright lighting, shadow, warm lighting, dappled light, light on hair. You should be able to trigger warmer lighting with prompting but an added lora will bring it out more. For darker pictures, use a vante black. Added enchanting eyes, artistic shot, orc, halfling, cyberpunk and a variety of styles of facial expression (see below). Photorealistic added, and added semirealistic which of course could also be used as a negative prompt. Enchanting eyes does seem to pop.

I've noticed the longer training has generalized the results more, so you can try weighting lighting effects and artistic shot to reproduce that more romantic style v2 sometimes hit. Basically it bleeds a little semirealistic/artistic/romantic and a little realistic into everything unless you direct it more strongly.

Full list of old and new tags for prompting

Neoteny, (soft jaw, small nose, plump lips, small mouth, wideset eyes, large eyes)

Barely, cute

Mature, milf, middle aged

latina, latino

asian, (japanese, taiwanese)

african, black, (dark black skin)

european, (pale skin)

(enchanting eyes, artistic shot, cute smile, shy smile, eyes crinkled, warm smile, amused, playful, innocent look, dreamy expression, playful expression, bedroom eyes, crinkled eyes, pose)

(round face, awkward face, narrow face, chubby cheeks)

(photorealistic, realistic, semirealistic)

elf, small elf ears, large elf ears, (fantasy, fantasy clothing, modern clothing, cyberpunk, cyberpunk clothing)

barbarian, primitive clothing, furs, fur lined, fantasy

tiefling, sheep horns, demon girl, pointed ears, (fantasy, fantasy clothing, red eyes, yellow eyes, modern clothing, cyberpunk clothing)

halfling, halfling ears, pointed ears, fantasy, fantasy clothing

orc, orc ears, goblin ears, green skin, (fantasy, fantasy clothing, cyberpunk, cyberpunk clothing, primitive clothing, barbarian)

victorian clothing

cyberpunk, (cyberpunk city, cybernetic arms, neon cyberpunk city, shiny clothes, transparent clothes, tattoo, peircing, dyed hair, multicolored hair, assymetric hair)

Neoteny v2

First image is basically the same prompt as the last version, so you can see the difference. Neoteny is a specific set of youthful facial features. This lora is designed to reproduce those features in adults. The results from this version will be more realistic.

Big revision. This is built on entirely new larger 85 image dataset of ~70% real pictures, and 30% synthetic from flux and the best of XL, trained on real dream for 40 epochs.

Main prompt "Neoteny, soft jaw, small nose, plump lips, small mouth, wideset eyes, large eyes", with any of those being optional apart from neoteny. Read below for additional prompting options!

New inclusions: (Mature, milf, middle aged) for hopefully older neotenous faces, (barely) for youthful looking 18 year olds, a few samples of hispanic (latina, latino), asian (asian, japanese, taiwanese) and black (african, black) optionally (dark black skin) if you find it's not pushing hard enough, with european for the majority of hand labelled images. I also added 4-5 elf images for good measure, and one with horns, and two barbarians so there's some fantasy clothing and themes in the mix, which should make it easier to use in non-standard ways (elf, small elf ears, large elf ears, barbarian, fantasy clothing, primitive clothing, modern clothing, horns, goat horns)

Pale skin is also included, and soft jaw, small nose, plump lips, small mouth, wideset eyes, large eyes has been labelled seperately. These as well as neoteny are the main keys.

And as last time, I labelled round face and narrow face, for variation there.

This is all hand curated and should provide better granular control, a wider variety of potential outputs, and because of a bigger dataset with higher resolution trained on real dream, one would hope a much better quality of output. Not being trained directly on pony, however it may give less exaggerated results.

Training like this costs credits so feel free to send me a tip, lol :P

...

V1:

So this lora was basically trained on about 50 pictures of real and synthetic adults (18-20 year olds) with neoteny - wideset large eyes, small nose, small mouth with plump lips. And that's what it's intended to produce - adults with neoteny.

Should be triggered with something like: large eyes, small nose, small mouth, plump lips, neoteny

One of my first lora's. Trained on Pony this time (although much better results from pony realistic, as anime/cartoon features are already exaggerated).

Recommend lowish weights (.4-.45). Recommend adetailer application for anything but a facial close up. Could use adetailer there even if not, as it's a small dataset and will have biases on poses, clothing etc.

Did use some NSFW in the dataset too (although doesn't seem too biased there). Did also use mostly neutral faces, so it will 'war' with any expression loras - try lower weights like .45. Seems to go really ham with asians (which I guess can have naturally more neoteny), so lower weights there too. In general probably .4-.45 is a better strength as Pony takes to this much more than 1.5 models did, presumably because of the anime basis. Even 0.3 is fine. A little goes a long way.

Has also been trained on glasses, and round face/narrow face, so you can use those as key words.