grok / editImage

The edit half of the Grok pair, and it's about as simple as a Comfy node gets: prompt and images are required, that's the whole required list, and the optional set is just quantity, output_format, image_metadata, and api_config. You feed it an image, tell it what to change, and xAI's autoregressive image model rewrites it - hosted on Civitai's cloud, costing Buzz, using none of your GPU.

Why you'd reach for it. Grok's image model has a genuinely different personality from the diffusion engines in this pack. It's not a diffusion model at all - it's an autoregressive mixture-of-experts that predicts visual tokens - and that shows up as a more "conversational" response to edit instructions: you can describe a change in natural language and get something that tracks your intent instead of just your strongest keywords. It's also, per the standing community read of xAI's product, the least-restrictive mainstream generator on the market. Both of those make it a reasonable first stop when a strict engine like Gemini or Flux has refused or fumbled an edit. Civitai's own moderation still applies on the request path, so the "least restrictive" has limits - the launch thread was explicit that nodes inherit the same restrictions as the web generator.

How it works. Standard pack machinery under the hood. The node serializes your input image into a workflow payload, submits to orchestration.civitai.com, long-polls until the job is terminal, and returns four outputs: images (the edited result), errors (JSON of step errors), workflow_id, and raw_json for cost and debugging inspection. Wire the result into Preview or Save Image and you're done - no model loading, no VRAM spike, nothing local to configure.

Getting good edits. Since there's no strength slider, prompt quality is the whole game. Grok rewards conversational instruction: "change the lighting to golden hour and add a rain-soaked street reflection" beats terse tags. If an edit comes back too aggressive, rephrase with more constraints - there's no dial to rescue it. quantity (1-4) is the cheap insurance: generate a couple of takes when the edit is fiddly.

Install. It ships in the pack with everything else:

cd ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git
pip install -r civitai-comfy-nodes/requirements.txt   # requests

Or ComfyUI Manager → Custom Nodes Manager → search Civitai Comfy Nodes → Install, then restart. Nothing to download for the model - it's cloud-side.

Auth and gotchas. Credentials resolve from a wired Civitai Auth node, then CIVITAI_API_TOKEN, then the stored key or OAuth login from the sidebar. Know the error hints: 402 insufficient Buzz, 401 bad/expired token, 429 rate limited; jobs cap at 30 minutes and the pack is early preview, so node shapes can shift between releases.

The data caveat applies fully - your source image and prompt leave the machine for xAI's model via Civitai's API. For a cloud edit that takes instruction well and refuses less than its peers, though, this is a quietly underrated node, and worth keeping wired next to the Grok create node in any workflow that wants a less-filtered option.