CIVITAI / Workflows

Anima Image-to-Image Workflow

This ComfyUI workflow is designed for Anima image-to-image generation, anime-style image reconstruction, prompt-assisted redraw, and controlled visual transformation from an existing image. The main purpose of this workflow is to let creators upload a source image, automatically analyze it with Qwen3-VL, convert the image content into a useful text description, and then use Anima Preview to redraw or transform the image through an image-to-image pipeline. Unlike a pure text-to-image workflow, this graph starts from an existing image. The source image provides the original structure, composition, subject placement, and visual direction. The model then uses the generated prompt and the encoded image latent to create a new result based on that input. This makes the workflow useful when you already have an image idea but want to restyle it, improve it, anime-fy it, rebuild it with Anima, or create a controlled variation without starting from zero. The workflow is built around Anima Preview, using anima-preview.safetensors as the main diffusion model. It also uses qwen_3_06b_base.safetensors as the CLIP/text encoder and qwen_image_vae.safetensors as the VAE. This gives the workflow a lightweight but practical image-to-image generation structure for anime-style and illustration-style reconstruction. One of the most useful parts of the workflow is the Qwen3VLProcessor node. The source image is passed into Qwen3-VL with the instruction “Describe this anime image.” Qwen3-VL then generates a text description of the image content. This response is connected directly into the positive prompt CLIPTextEncode node. In other words, the workflow can automatically turn the uploaded image into a prompt, then use that prompt to guide the Anima redraw process. This automatic prompt reconstruction is useful for users who do not want to manually describe every detail in the source image. If the image contains a character, clothing, pose, background, color palette, or scene style, Qwen3-VL can produce a descriptive prompt that gives Anima a clearer semantic direction. This helps the workflow preserve the image concept while still allowing the model to rebuild the final result. The source image is first loaded through LoadImage, then resized through image_scale_pixel_v2. The resize node controls the total pixel count and aligns the image to a model-friendly grid. In the uploade