Which models are compatible with this workflow?

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.

Why do my images look blurry or washed out?

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”).

How do I change aspect ratio without harming quality?

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.

How can I get consistent or varied results?

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.

Anime Generator - ComfyUI Workflow

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. Finally, VAEDecode turns the denoised latent into a full-resolution image and SaveImage writes the output to disk.

Technically, the core of the generation happens in KSampler, which accepts the UNet weights (UNETLoader), text embeddings (CLIPTextEncode), and an initial noise latent (EmptyLatentImage). The ResolutionSelector ensures the latent size stays near 1 MP for speed and quality balance—ideal for anime linework and clean shading. The VAE you load matters: it defines how colors and fine details are encoded/decoded. CLIPTextEncode supports positive and negative prompts, so you can include quality tags and style hints (e.g., "masterpiece, anime, dramatic lighting") while excluding defects (e.g., "worst quality, blurry, jpeg artifacts"). This combination delivers crisp, stylized anime art with consistent control over composition, detail, and style.

API

Use this workflow from code

Every Comfy workflow is a JSON graph. The payload below is this workflow, exactly as ComfyUI runs it — fetch it from the URL, keep it in version control, or load it in ComfyUI and run it node by node.

9f0b568bf8a1.json

Fetching workflow JSON…

Run it from TypeScript or Python with the Comfy SDK. The same code targets Comfy Cloud or a ComfyUI you host yourself — only the base URL changes.

// Install (beta)
npm i @comfyorg/sdk

// Run this workflow (TypeScript)
import { Comfy } from "@comfyorg/sdk";

const client = new Comfy({ apiKey: "comfyui-..." });
const wf = await client.workflows.fromFile("workflow_api.json");
const job = await client.run(wf);
await job.getOutputs("<output-node-id>")[0].toFile("output.png");

The SDK takes a workflow in API format: open this workflow in ComfyUI and use File → Export Workflow (API).

Comfy Cloud API access requires a plan with an API key.

SDK docs Get an API key

FAQ

Frequently Asked Questions