grok / v1.0 / editImage

The editing side of Grok's image model, in node form. xAI's generator - the autoregressive, token-predicting model known as the least restrictive closed image service around - takes your source image plus an instruction and hands back an edited IMAGE. It runs on Civitai's orchestration layer, so you don't need xAI's API key, just a Civitai account with Buzz.

This is the node for editing jobs where the stricter closed models (Google, fal's line) would refuse or water down the instruction, and you'd rather not fight a local pipeline for it. Grok's whole market position is "the closed model that flinches less," and the edit variant inherits that. It's also the simplest edit node in the pack: image in, prompt in, nothing else to steer.

How it works

The civitai-comfy-nodes recipe flow: upload your source image and instruction to the Civitai Orchestration API as a grok / v1.0 / editImage job, long-poll until it succeeds, and convert the result back to a native IMAGE tensor. Because Grok isn't a diffusion model, there are no strength/sampler knobs to expose - the API doesn't have them, and neither does the node.

The inputs that matter

• prompt and images - both required. Instruction in the prompt, source in images.
• quantity - 1 to 4 edited outputs per job.
• output_format / image_metadata - format choice and optional metadata.

That's the entire surface. No seed, no strength, no negative_prompt. If you want a subtler edit, you say "subtle" in the prompt - there's no slider for it.

Outputs

images → Preview/Save Image. errors, workflow_id, raw_json are diagnostics.

Installing and authenticating

Pack-wide install:

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

or ComfyUI Manager → "Civitai Comfy Nodes" → Install → restart. Dependencies: requests + python-socketio[client].

Auth: Civitai Auth node → CIVITAI_API_TOKEN → stored key/OAuth → browser login. Headless:

export CIVITAI_API_TOKEN=your_token

Where people get burned

• No strength control. Local editors give you a denoise knob; Grok v1.0 edit doesn't. Prompt wording ("keep everything else identical") is the only lever.
• The moderation is looser, not absent. Grok's edge is that it flinches less, but xAI still filters at the source, and you can't edit a model you can't download.
• Your source image leaves the machine, subject to xAI's policies via Civitai's orchestration. Keep that in mind before editing anything sensitive.
• Billed per job in Buzz - and each quantity step is a separate billed output.
• No reproducibility. Save good results immediately; there's no seed to replay.
• If you want the resolution/quality options, the Grok v2.0 edit node adds resolution and quality knobs - v1.0 is the simpler path.