Initial release, does an okay-ish job on standing Yinglets, but tends to render extra arms, legs, tails or a freak amalgamation of the three (tail feet seem to be a particular favourite), negative prompts/embeds seem to have no affect on rectifying this (that or I need to git gud).
Dataset consisted of 112 images that were cropped, manipulated if they contained more than one character and had all text/watermarks scrubbed manually, than an images under 512x512 were upscaled. Captioned was generated by BLIP and not edited in the slightest, so they were extremely inaccurate, as a result the shelltooth seems to be invoked at complete random (a CFG of around 9 or lower seems to increase chances), and the tail tuft isn't sex-specific yet.
Trained on bb95FurryMix_v110, but seems to play nice with most other furry checkpoints, my recommendations are kavkaMixVersion16_v16, yiffymix_V32 and snoutmix_cgiV1. Some models like Fenris have a tendency to render characters without ears unless specified.
Planned improvements for v2.0 onward:
Hand write the captions for accuracy
Add triggers for Yinglet anatomy (shelltooth, tail tuft, eyebrow whiskers and eyes with slit pupils + no sclera) and specific characters (Kass, Lopin, Vizlet etc.)
Find more images of Yinglets looking at the viewer
General training stuff (I'm currently restricted to 8GB VRAM, so we'll see)
Initial release, does an okay-ish job on standing Yinglets, but tends to render extra arms, legs, tails or a freak amalgamation of the three (tail feet seem to be a particular favourite), negative prompts/embeds seem to have no affect on rectifying this (that or I need to git gud).
Dataset consisted of 112 images that were cropped, manipulated if they contained more than one character and had all text/watermarks scrubbed manually, than an images under 512x512 were upscaled. Captioned was generated by BLIP and not edited in the slightest, so they were extremely inaccurate, as a result the shelltooth seems to be invoked at complete random (a CFG of around 9 or lower seems to increase chances), and the tail tuft isn't sex-specific yet.
Trained on bb95FurryMix_v110, but seems to play nice with most other furry checkpoints, my recommendations are kavkaMixVersion16_v16, yiffymix_V32 and snoutmix_cgiV1. Some models like Fenris have a tendency to render characters without ears unless specified.
Planned improvements for v2.0 onward:
Hand write the captions for accuracy
Add triggers for Yinglet anatomy (shelltooth, tail tuft, eyebrow whiskers and eyes with slit pupils + no sclera) and specific characters (Kass, Lopin, Vizlet etc.)
Find more images of Yinglets looking at the viewer
General training stuff (I'm currently restricted to 8GB VRAM, so we'll see)