⚠️ GPU grant pending

Anima — Anime Text to Image

A 2B text-to-image model for anime characters, illustration and art styles — trained on millions of anime images, promptable with Danbooru-style tags, natural language, or both.

这个 Space 目前是 Static 占位页

完整 demo 代码已在仓库里（app.py + requirements.txt）， 但 Hugging Face 对未满 30 天的免费账号不开放 ZeroGPU / Gradio Space（返回 402）。 已提交 community GPU grant 申请，通过后这个 Space 会切换成 Gradio + ZeroGPU 正式运行。

Parameters2B
Checkpointanima-aesthetic-v1.1
Base modelCosmos-Predict2-2B-Text2Image
InferenceDiffSynth-Studio · bf16 · ZeroGPU

样图（本地 RTX 5070 实跑，30 步 / CFG 4.5）

silver-haired girl on a rooftop at sunset
1girl, silver hair, white sundress, rooftop at sunset, cinematic lighting · 896×1152
witch with staff in a moonlit forest
1girl, witch hat, staff, glowing runes, moonlit forest, dramatic lighting · 832×1152
small dragon sleeping on books
a small dragon sleeping on a stack of books in a candlelit library, painterly · 1152×896

Demo 会提供什么

示例提示词

1girl, solo, silver hair, blue eyes, white sundress, standing on a rooftop at sunset, wind, looking at viewer, detailed background, cinematic lighting
1girl, witch hat, black dress, glowing runes, moonlit forest, dynamic pose, dramatic lighting, highres
a small dragon sleeping on a stack of books in a cozy candlelit library, warm light, highly detailed background, painterly
1boy, samurai, black haori, rain, narrow alley, lanterns, low angle, cinematic composition, monochrome

自己跑一份

pip install diffsynth==2.1.9

from diffsynth.pipelines.anima_image import AnimaImagePipeline, ModelConfig

pipe = AnimaImagePipeline.from_pretrained(
    torch_dtype=torch.bfloat16,
    model_configs=[
        ModelConfig(model_id="circlestone-labs/Anima",
                    origin_file_pattern="split_files/diffusion_models/anima-aesthetic-v1.1.safetensors"),
        ModelConfig(model_id="circlestone-labs/Anima",
                    origin_file_pattern="split_files/text_encoders/qwen_3_06b_base.safetensors"),
        ModelConfig(model_id="circlestone-labs/Anima",
                    origin_file_pattern="split_files/vae/qwen_image_vae.safetensors"),
    ],
)
image = pipe(prompt="1girl, silver hair, sunset rooftop, looking at viewer",
             num_inference_steps=28, cfg_scale=4.5, width=1024, height=1024)

提示词要点