Anime Generator
Creating your own generated anime characters and anime-style art with this app.
About
This ComfyUI workflow generates anime characters and anime-style scenes from text prompts using a classic Stable Diffusion pipeline. In Step 1 (Load Models), UNETLoader, CLIPLoader, and VAELoader load your chosen anime-tuned checkpoint, text encoder, and VAE. Step 2 (Image Size (1MP)) uses ResolutionSelector to set an efficient, 1‑megapixel canvas, which then feeds EmptyLatentImage to create a latent noise tensor at the selected dimensions. Step 3 (Prompt) processes your positive and negative prompts through CLIPTextEncode, and KSampler iteratively denoises the latent with the UNet, guided by your prompts and sampler settings.
What you provide
- Reading the workflow
Runs in ComfyUI. This workflow can't run on this page. Open it in Comfy Cloud with everything loaded, or download it for your own ComfyUI.
Try in CloudQuestions & answers
This setup is designed for SD 1.5-style checkpoints that use a single CLIP and VAE (e.g., anime-tuned models). Load a matching trio: UNet checkpoint in UNETLoader, its corresponding CLIP in CLIPLoader, and a compatible VAE in VAELoader. Mixing families can produce washed colors or artifacts.
Common causes are a mismatched or low-quality VAE, too few sampling steps, or overly low CFG. Try a VAE known to match your checkpoint, raise steps to 25–35, set CFG around 6–8, and add a strong negative prompt (e.g., “blurry, jpeg artifacts, low contrast”).
Use ResolutionSelector to pick dimensions near 1 MP that fit your aspect ratio (e.g., 1216×704 for wide, 896×1152 for tall). Staying near 1 MP balances detail and speed. Then adjust prompts with composition hints like “portrait, headshot” or “full body” to guide framing.
Set a fixed seed in KSampler for reproducible outputs with small, controlled prompt edits. For variety, randomize the seed, slightly alter wording, or try a different sampler (e.g., DPM++ vs. Euler a). Keep CFG and steps stable when comparing results.