Instructions to use circlestone-labs/Anima with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusion Single File
How to use circlestone-labs/Anima with Diffusion Single File:
# No code snippets available yet for this library. # To use this model, check the repository files and the library's documentation. # Want to help? PRs adding snippets are welcome at: # https://github.com/huggingface/huggingface.js
- Notebooks
- Google Colab
- Kaggle
Thank you for everything ❤️
This is a message CircleStone Labs, Comfy Org and tdrussell.
First of all, I would like to say a massive thank you for open sourcing this incredible anime model, especially in a world where companies are closed sourcing models. I've been using anima since preview one and it's been absolutely incredible and seeing the journey of this model has been great, especially regarding updates, community feedback the overall community Checkpoints and LORAs... It really has got similar SDXL and illustrious vibes From back in the day.
The final bass version has been amazing as it has got better prompt adherence, quality and overall stability, especially with the new official turbo LORAs as well as as the great RDBT - Anima and Cosmos Predict 2.5 DMD2 LoRA: https://civitai.com/models/2466415/cosmos-predict25-2b-base-distilled-extracted-dmd2-lora
I think it's fair to say that for a long time anime fans like myself have been waiting for a great model like this as there are some amazing models but they are based on older architecture this anima model with the very capable but tiny Qwen3 0.6b has been really good. It would be interesting in the future a new Anima model based on the newly released Nivida Cosmos 3.
Is exciting to see what the future holds for this great model and I definitely can't wait for the Anima turbo model. I've used some great distilled turbo models like Flux Klein, Ernie Image and the legendary Z Image turbo. I know there are some newer techniques being used have evolved from standard DMD2 (steps and CFG distilled) like Continuous-Time Distribution Matching (CDM) which looks really promising: https://huggingface.co/papers/2605.06376
Thank you also to the great community that has been keeping this model alive on Huggingface and CivitAI.
Thank you very much tdrussell for everything you've done as I can't imagine how much time, work, money and dedication It has taken to get this far and a huge congratulations on reaching the final base version as each preview version has been a big positive jump to one another. Take care.