Type | |
Stats | 132 726 |
Reviews | (38) |
Published | Nov 17, 2024 |
Base Model | |
Training | Steps: 464 Epochs: 10 |
Usage Tips | Clip Skip: 1 |
Trigger Words | cinccino |
Hash | AutoV2 29252DF57D |
Cinccino LoRA trained on Pony upon request. Works extremely well, with a few consistency issues that can pop up when you stray too far from situations in the training data. Capable of producing stunning results.
Issues:
Can have trouble with tail placement or generate multiple tails when the tail gets occluded by something in the foreground
This is a common issue I see with tails in general in SD, which gets exacerbated here because Cinccino has so much fluff. I added (multiple tails) to my negative to help with this
Cinccino's hands like to default to being curled up and held against their chest
This is caused by this pose being extremely overrepresented in the source data
To address this I prompted for what I wanted the hands to be doing. I used phrases like 'hands in pocket' or 'hands behind back'. Once you prompt for a specific action, the model will follow. If the hands still go into that pose, you may need to verify the phrase you used is understood by your base model or adjust your CFG scale to enforce stricter prompt adherence.
If you add too much emphasis to a clothing prompt that covers the neck area like (sweater)1.4, it can cause Cinccino's neck fluff to disappear
Lowering the strength of the emphasis fixed this issue for me
Torso and inner thigh fur can become white or a lighter shade. It's supposed to be a uniform grey
If you run into this, adding (grey fur) to your positive and (white fur) to your negative can help
Generation Tips:
Cinccino is super fluffy, lean into the fluffy by adding tags that emphasize that like 'fluffy fur' 'fluffy' 'detailed fluffy fur' and 'large fluffy tail'!