grok edit (comfyapikey)

The "least restrictive" image model, wired into your graph

The display name is literally "grok edit (comfyapikey)" - no glamour, which fits the job. This node runs xAI's Grok Imagine image editing through Comfy's API proxy, with your ComfyUI API key. Give it one or more reference images and a prompt, and it edits them the way GPT Image or Nano Banana edit - but from xAI, the vendor the community consistently calls the least-restrictive mainstream option. Grok's generator is architecturally a different beast too: an autoregressive mixture-of-experts that predicts visual tokens rather than doing diffusion.

Grok has no open weights, so this is one of those "call it, you can't download it" cases from the API-wrapper playbook. The node gets it into a workflow next to your local nodes without you ever touching xAI's own site.

How it works

The node base64-encodes each input image into a data URL and POSTs a JSON payload to https://api.comfy.org/proxy/xai/v1/images/edits with your key in the X-API-KEY header. It's interrupt-aware - you'll see elapsed seconds on the node while it waits, and cancel actually cancels. Result images get downloaded and returned as a batched IMAGE, so you can set number_of_images above 1 and get them all out of one call.

Four models are offered, and the default grok-imagine-image-2.0 is the one you probably want:

• grok-imagine-image-2.0 - the current flagship edit model; accepts up to 3 input images, and the only one where the quality setting does anything.
• grok-imagine-image-quality and grok-imagine-image - older tiers, also up to 3 images.
• grok-imagine-image-pro - the heavy one, and it's picky: exactly 1 input image, and it ignores aspect ratio entirely.

The inputs that matter

• api_key - your ComfyUI API key, from comfy.org/api-keys. It also ends up in the saved workflow file, so keep those private.
• prompt - what to change. This is an edit, not a from-scratch generation: describe the change.
• image_1 - the first reference. image_2 and image_3 are optional references for multi-image edits.
• model - default 2.0 is fine to leave.
• aspect_ratio - here's the trap: a custom aspect ratio (anything but auto) requires multiple input images for the quality/base models. One image + a custom ratio = an error string, not an image.
• resolution (1K/2K) and number_of_images (1–10) - straightforward.

Outputs: image (the edited result, batched if you asked for several) and show_price, a string with the actual cost in USD/RMB plus credits used when the API reports it.

Install and issues

Shared pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/7BEII/comfyui-PD_comfy-api-node.git
pip install -r comfyui-PD_comfy-api-node/requirements.txt

Restart and look under PD_Tools/Image_Generation. No extra model downloads - this is a network node.

The failure modes are mostly the API's, surfaced as error strings: hitting the input-image caps (3 max, or 1 on pro), the custom-aspect-ratio trap above, and timeouts - the node waits up to 600 seconds, and a busy Grok queue can eat all of it. Costs are per call and priced per model/resolution, which is why the show_price output exists. And the usual caveat: your prompts and images leave the machine for xAI's servers, moderated by xAI's policies - for the exact "least restrictive" content people reach for Grok to generate, the moderation at the API source is still the ceiling.