How is this different from a standard img2img or inpainting workflow?

ChronoEdit uses a video-model backbone (Wan 2.1) adapted for editing a single frame. Those temporal priors translate into stronger structural and style consistency during edits. You prompt for changes much like img2img, but results tend to preserve lighting, geometry, and identity more reliably. Masks aren’t required for global edits; for local changes, keep edit strength modest and use precise prompt phrasing.

What hardware do I need?

This workflow targets NVIDIA GPUs in ComfyUI. A modern CUDA-capable GPU with 12–16 GB of VRAM is a practical starting point for 512–768 px edits. If you hit out-of-memory errors, reduce resolution, steps, or batch size. Larger images or heavier settings may require more VRAM.

Where do I put the text encoder and model files?

Place the UMT5 XXL text encoder (e.g., umt5_xxl_fp8_e4m3fn_scaled.safetensors) into ComfyUI/models/text_encoders or the directory specified by the model pack you’re using. Install Wan 2.1/ChronoEdit weights per their documentation. The Subgraph will reference these paths when running.

How do I keep identity and style consistent across edits?

Use clear, repeatable descriptors in your positive prompt, keep the seed fixed while iterating, and start with a lower edit strength. Increase strength only as needed. Negative prompts can help suppress unwanted style drift. Make small, incremental changes to steps and CFG to fine-tune fidelity versus creativity.

ChronoEdit 14B: Image Edit - ComfyUI Workflow

ChronoEdit 14B: Image Edit is a ComfyUI workflow that repurposes a video model’s dynamic understanding to make single-image edits that look physically plausible while preserving character and style consistency. The core path is simple—LoadImage feeds your source into a ChronoEdit subgraph (node id 2b61e18f-9327-49e6-98af-da8e557c2336), and the result is written out with SaveImage. The heavy lifting happens inside that subgraph: it loads the Wan 2.1/ChronoEdit weights, encodes your prompt with a large text encoder (UMT5 XXL), and runs an edit-oriented denoising pass designed to keep structure, lighting, and identity intact.

Technically, leveraging a video model for a single frame means the sampler benefits from priors that normally enforce temporal stability—here, they help maintain geometry and style during edits. Prompts are handled by a UMT5 XXL text encoder (for example, umt5_xxl_fp8_e4m3fn_scaled.safetensors), which gives strong language guidance for nuanced changes. A MarkdownNote in the graph provides inline guidance, while the Subgraph encapsulates typical controls like positive/negative prompt text, edit strength, steps, CFG guidance, seed, and resolution. The result is a practical, prompt-driven image editor that excels at gentle to moderate changes without breaking identity or composition.

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.

image_chrono_edit_14B.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 "ChronoEdit 14B: Image Edit" (TypeScript)
import { Comfy } from "@comfyorg/sdk";

const client = new Comfy({ apiKey: "comfyui-..." });

// This workflow, exported in API format (see note below)
const wf = await client.workflows.fromFile("image_chrono_edit_14B_api.json");

const asset = client.assets.fromFile("input.png");
wf.setInput("52", "image", asset);  // LoadImage

const job = await client.run(wf);
await job.getOutputs("60")[0].toFile("output.png");  // SaveImage

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