Apppublic

hugging-apps/anima-turbo-4step-demo

sourceHugging Faceupdated 2mo agoView on Hugging Face

4likes

App README

Anima Turbo 4-Step Demo

This Space demonstrates the sorryhyun/anima-turbo-4step LoRA, a turbo distillation applied on top of circlestone-labs/Anima, a 2B-parameter text-to-image model for anime-style illustration.

The LoRA is distilled using DP-DMD and Continuous-Time Distillation, enabling high-quality generation in just 4 steps at CFG=1.0.

—Base model: CalamitousFelicitousness/Anima-sdnext-diffusers (diffusers conversion of Anima, using Cosmos-Predict2 architecture with Qwen3 text encoder)
—Turbo LoRA: sorryhyun/anima-turbo-4step (anima_turbo_4step_v2_longer.safetensors)
—Settings: CFG=1.0, steps=4, shift=3.0 scheduler

The model is trained on Danbooru-style tags. Recommended positive prefix: masterpiece, best quality, score_7, safe,

Tag order: [quality/meta/year/safety tags] [1girl/1boy] [character] [series] [artist] [general tags]