Anima Noct

An anime merge on the 2.9B Anima, with a crisper finish and cleaner anatomy. It handles SFW and explicit prompts alike, runs in any up-to-date ComfyUI with the standard Anima text encoder and VAE, and the int8 file fits 8 GB cards.

What it does well

• Cleaner anatomy and a crisper finish than the Anima-2.9B base

• Nudity without a LoRA, and scenes that stay clothed with the safe tag

Which file

• 8 GB cards and up: int8 convrot, 3.1 GB, the default and the fastest (about 30 seconds per image at 832x1216 and 30 steps on an 8 GB RTX 4070 laptop)

• fp8, 3.1 GB, if your ComfyUI build does not load int8 convrot

• bf16, 5.8 GB: full precision, for merging and training

Settings

30 steps, CFG 4, euler with the simple or sgm_uniform scheduler, around one megapixel: 832x1216 in portrait, 1024x1024 square. Text encoder: qwen_3_06b_base.safetensors. VAE: qwen_image_vae.safetensors. Both come from circlestone-labs/Anima on Hugging Face. Negative prompt: worst quality, low quality, score_1, score_2, score_3, blurry, jpeg artifacts, sepia. No trigger words. The workflow is in the zip.

How to prompt

Start with the quality and rating tags, then who is in the image, then describe the scene in plain sentences. The rating tag sets how explicit the image gets: safe, sensitive, nsfw or explicit; add uncensored for explicit scenes. For adult characters, add mature female and write the age: "an adult woman in her twenties". Character, series and artist tags work as they do in Anima.

Template:

masterpiece, best quality, safe, 1girl, solo, mature female, An anime illustration of an adult woman in her twenties with <hair>, <outfit>, <pose and place>. <light>, <one or two props>. Waist-up, cel shading.

Prompt examples

Three prompts from the showcase, as rendered. Every showcase image carries its prompt, seed and settings.

The tram letter

masterpiece, best quality, safe, 1girl, solo, mature female, blonde hair, An anime illustration of an adult woman in her mid-twenties with long wavy blonde hair, a calm mature face with defined cheekbones and a tall womanly figure, wearing a cream cable-knit sweater under a camel wool coat, sitting by the window of an old tram at dusk and reading a handwritten letter, a faint smile on her lips. Rain streaks run down the glass, the city lights outside dissolve into amber and teal bokeh, and the warm cabin lamp lights the paper and her face. Waist-up, three-quarter view, clean line art, soft cel shading, cinematic colors.

中文：基于 Anima-2.9B 的动漫融合模型，全年龄和成人内容都能生成。30 步，CFG 4，euler，约 100 万像素；文本编码器 qwen_3_06b_base，VAE qwen_image_vae。8G 显存用 int8 convrot（3.1 GB）；fp8 同为 3.1 GB；bf16（5.8 GB）用于融合与训练。模型仅限非商业用途（CircleStone 非商业许可），生成的图片可按许可自由使用。

License

The CircleStone Model is licensed by CircleStone Labs LLC under the CircleStone Non-Commercial License. Copyright CircleStone Labs LLC. IN NO EVENT SHALL CIRCLESTONE LABS LLC BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH USE OF THIS MODEL.

Anima Noct is a modified version of the CircleStone Model and is not an official CircleStone Labs product. Non-commercial use of the model only; the images you generate are yours to use, as the license allows. The full license is in the zip.

Built on Anima-2.9B by Gazingstars123; includes One obsession by maxfeifei8.