v4
Compatible with ComfyUI and Forge Neo.
For prompting guidelines, please refer to the official Anima model page.
Positive Prompt:
masterpiece, best quality, [safe / sensitive / explicit, uncensored]
Choose one rating tag: safe, sensitive, or explicit.
Negative Prompt:
worst quality, low quality, score_1, score_2, score_3, artist name
Recommended Settings:
Sampling method: Res Multistep
Schedule type: Simple
Sampling steps: 21–24 (22 recommended)
CFG Scale: 2.5–3.5 (3.0 recommended)
Resolution: Up to 1536 × 1536, depending on available VRAM,
Recommended Resolutions: 1024 × 1024, 832 × 1216, 896 × 1152, 1024 × 1344
Low-VRAM Option:
If you have limited VRAM, you may use the community-made INT8 Qwen3 0.6B text encoder. It can reduce memory usage, although it may not improve generation speed. The official text encoder is recommended if you have sufficient VRAM.
Description
v1.0
This is a partial merge of RDBT Anima and AnimaYume.
FAQ
Comments (17)
What files do I need to make "Anima" work on "Forge NEO"?
Thank you
If only their was a way to convert Pony/Illustrious to Anima
Let's hope someone can find a way to make a adapter of some kind, like this one that made pony loras > sdxl possible https://civitai.com/models/662937/pony-lora-adapter
@richerich I don't think it's possible. Pony was trained on SDXL and that's made it possible to convert. Anima, however, was trained on Cosmos-Predict, which make it impossible to do.
@Twisowk I thought Anima is a community-trained checkpoint built on Stable Diffusion XL (SDXL) 1.0, specifically optimized for generating high-quality anime-style images. It is fully compatible with the SDXL 1.0 base model, meaning it can be used with any SDXL-compatible tools, platforms, or workflows
I understand that this is a test model, but I have to say, it's probably the most REFINED model out of the other Anima models.
Keep up all the fantastic work!
Really like the model so far. Ihated the base Anima but this one seems much more stable
What kind of data is this using? Looks synthetic, but might just be base model or me imagining things. Does seem more stable sometimes though
This model merged rdbt v0.12 with another mid-scale finetuned model, let me guess, it's animayume v0.1.
@giko no offense, but i think I've made it very clear in the rdbt model page that "merging rdbt into other models is not allowed." So I kindly request you to remove your model.
@reakaakasky Why do you not allow merges? Seems kinda petty in my opinion really. You do not hold copyright over the images you trained it on most likely, and stole from artists so it's quite rich to say the least. You additionally do not control the licensing of this model to my understanding, that was decided by NVIDIA when they created Cosmos which Anima is based on.
I merged this model not knowing it was a merge of a model that didn't allow this. I will remove my merge as well if and ONLY IF
1. You can provide independently verifiable proof that your model was used
2. I have to and it isn't just "this hurts my feelings" if I don't get lost quite frankly (e.g. ToS violation, copyright violation)
OR you can give a compelling reason as to why I should give a single shit
@reakaakasky Imagine stealing artists' images and then pretending to own the model, lmao. If you want, we can do an exercise. Let's ask the artists you used in the dataset if using their copyrighted images for training is allowed. If one of them says no, then you delete your model. Shall we do it?
I took a look at the alleged base model here, and @reakaakasky what you're doing is not only petty, but also not legal according to the Anima license. According to the license, you're not allowed to limit the model further, let alone stray from the original license completely.
@GPUPoorChad This behavior and type of comment is typical for him unfortunately. The dude is a complete schizo and insanely rude to boot, best to not even engage him. He blocked me for asking what the intention of one of his loras was and how it worked, he's a highly petulant person.
Excellent model, I love it. I'm testing it to see how far it goes in the interaction between two characters, and it works perfectly. It has a lot of potential. Is there any way I can make Lora? And above all, the speed and small file size are great.
Very good work! Keep it up.
This is the best checkpoint of our time.
- Comment from Korea.