Anima: simple workflows for generation & inpainting/refining + tips & info — anima_simple

This workflow set is for creating and refining images with Anima: it can generate a new image from a plain-language prompt or edit an existing image through inpainting, refining, or image-to-image, producing an image.

Use the generation path for a new image, or the editing path for an existing image.

The editing path can switch between image-to-image and inpainting modes and blend two images with a semi-transparent mask.

Generation accepts plain-language descriptions, with suggested style terms such as “anime,” “digital anime,” “digital painting,” and specific artist styles.

An image generated in the first path can then be further edited in the inpainting path.

The listed components are Anima v1.0, Qwen 3 0.6b, and Qwen Image VAE; the main model is described as about 4.5GB and usable with as little as 6GB of VRAM.

Check for these model files: anima-preview2.safetensors, qwen_3_06b_base.safetensors, and qwen_image_vae.safetensors.

Sugerencia · sin verificar

RES4LYF is used for the Clownshark Ksampler; rgthree-comfy is optional and is called RGTHREE in the notes.

For the inpainting path, the notes list ComyUI Essentials and ComfyUI LayerStyle.

Try ETA values from 0.2 to 0.5 and 22 to 30 steps; around 24 steps is suggested, with more steps producing deeper shadows.

Sugerencia · sin verificar

For higher resolutions, start smaller, resize the image, then feed it into inpainting for image-to-image at around 40 denoise.

Sugerencia · sin verificar

For a more anime-leaning style, add “anime” or “digital anime” to the end of the basic positive prompt; “digital painting” and specific artist styles are also suggested.

Sugerencia · sin verificar

Qwen Image VAE can sometimes produce a noticeable grid-like halftone pattern, especially at higher resolutions; SeedVR2 may reduce it, but it requires at least 16GB of VRAM and is not always the best upscaler.

The notes describe Preview 3 as unstable, with art-style inconsistency and drift, and point users toward Preview 2 unless there is a specific reason to use Preview 3.