---
base_model:
- circlestone-labs/Anima
license: apache-2.0
pipeline_tag: text-to-image
---

[Anima](https://huggingface.co/circlestone-labs/Anima) lora for turbo generation, implemented with [DP-DMD](https://arxiv.org/abs/2602.03139) and Continuous-Time Distillation [paper](https://arxiv.org/abs/2605.06376).

This model is presented in the paper [When does training on downscaled images yield the same gradients?](https://huggingface.co/papers/2608.04448).

Try new `anima_turbo_4step_v2_longer.safetensors`


![nfe4_10k__00032_](https://cdn-uploads.huggingface.co/production/uploads/60c82ac0dfbd57a384a01127/HArQ0U-4CvHsVeRnLjgjn.png)

trained with 10k steps, 30h in single 5070ti using [anima_lora](https://github.com/sorryhyun/anima_lora)

Reproducible by CFG=1.0, steps=4, er_sde sampler