Recommended settings (v2):
Environment - ComfyUISteps: 16+CFG scale: 1+Note - This checkpoint is capable of generating decent results with just 8-10 sampling steps
Positive Prompt
masterpiece,best quality,amazing quality,very aesthetic,absurdres,ultra detailed face,ultra detailed eyes, Negative Prompt
bad quality,worst quality,worst detail,sketch,censor, extra limbs, deformed fingers, bad anatomy, mutated body, lowres, worst quality, low quality, low score, bad score, blurry, text, ugly, hooded eyes, watermark, pale, bad hands, bad anatomy, bad proportions, poorly drawn face, poorly drawn hand, missing finger, extra limbs, blurry, pixelated, distorted, lowres, jpeg artifacts, watermark, signature, text, (deformed:1.5), (bad hand:1.3), overexposed, underexposed, censored, mutated, extra finger, cloned face, bad eyes,🌙 Unholy Desire – Lunar Rapture
This model is basically me experimenting and trying to figure out how to properly train anime models.
It was trained on a very small AI-generated dataset with only around 1000 training steps, so don’t expect too much from V1. This release is more of an experimental starting point than a polished final model.
✨ Planned Versions
Pure Trained Version (This One) → Models trained directly on AnimaBaseV1
Merge Version → Models created by merging two of my own trained AnimaBaseV1 models together
📦 About This Page
This page will act as an archive for:
Fully trained anime models based on AnimaBaseV1
Merge versions created only from my own trained models based on AnimaBaseV1
Future experimental training iterations and improvements
Description
FAQ
Comments (3)
Some of these art styles are really good and make the colors pop! What are some of the quality prompts that you use to make this happen?
Only reason why I'm asking is sometimes my generations come out like 5% foggy/light bloom.
I'm not sure what they do, but I get better results with the Turbo lora and CFG 1 or 1.5 (for neg prompt)
Hello, I have a few questions about your model training method that I would like to ask you.
Did you train it directly on the base model (full fine-tune / DreamBooth), or did you train a LoRA and then merge it back into the base model?
The reason I am asking is that I am trying to train a LoRA on AnimaBaseV1, and then merge it back into AnimaBaseV1. The merged model can generate images normally, but it cannot be used for further training.
Could you please share what training script and version you use? And do you have any advice on how to properly merge a LoRA back into an AnimaBase model without causing issues?
My English is not very good, and this message was left using AI translation. Thank you very much for your help!