Nodes/ComfyUI/Grok Image
ComfyUI Node Runs on cloud

Grok Image

XAI's generator, called straight from your ComfyUI graph

By Comfy-Org·Created 4 years ago·Updated about 19 hours ago· 130,663
Grok Image
    • IMAGE
    model
    prompt
    aspect_ratio
    number_of_images1
    seed0
    resolution
    quality

    This node doesn't touch your GPU. Not one tensor of it. Grok Image is a text-to-image node that forwards your prompt to xAI's Grok Imagine models through Comfy's hosted API proxy and drops the finished image back into your graph as a normal IMAGE. It's part of ComfyUI's "partner node" family - the cloud-backed sibling to all the local samplers you already use - which landed in core starting in late 2025 and grew through 2026.

    Why reach for it? If you want to compare an xAI generation against your local model without leaving the canvas, or you're on a laptop with no VRAM to speak of, this gets you Grok's output as a first-class node result. You can wire it straight into Preview Image or Save Image, or chain it into anything else that eats an IMAGE. The catch is the same one for every partner node: it needs a Comfy account with credits, and every generation costs money. There's a price badge right on the node, so you see the estimate before you hit Queue.

    How it works

    Under the hood it's a thin client. The node POSTs your prompt to xai/v1/images/generations (via Comfy's /proxy), waits for the result, downloads the image, and returns it as a tensor. No model files, no VRAM, nothing installed into models/. The heavy lifting happens on xAI's servers; Comfy just brokers the call and bills your account.

    The inputs that matter

    • prompt - the actual text prompt. Same thing you'd type anywhere else, but with Grok's conventions, not SD's.
    • model - pick from the Grok Imagine family: the current grok-imagine-image-2.0, plus -quality, -pro, and the base grok-imagine-image. Pro is the fancy one; 2.0 is the general default.
    • aspect_ratio - a long list from square up to tall phone ratios like 9:20 and 19.5:9, so you can generate social-media-ready crops directly.
    • number_of_images - 1 to 10. Remember this multiplies the bill, not just the queue time.
    • resolution - optional 1K or 2K if you want the bigger canvas.
    • seed - and here's the classic partner-node gotcha: the tooltip says it plainly. Seed only decides whether the node re-runs; the actual result is nondeterministic no matter what seed you set. Don't treat it as reproducibility.

    The single output is IMAGE, ready to feed a Save Image node or anything downstream.

    Getting it running

    It ships with ComfyUI core - no Manager, no git clone, nothing to install. First run will ask you to link a Comfy account and add credits (bought in Comfy's settings, priced per call). If the node shows up but errors out, that's almost always the account/credits step, not your install.

    Where people get burned

    • The bill. Each run costs real money, and number_of_images = 8 means eight charges. Watch the price badge; it exists for a reason.
    • Seed won't reproduce. If you expected to lock a result, you'll be disappointed - that's by design here.
    • The prompt style. Grok is not Stable Diffusion. If you write SD-style tags, you'll get mid results; write a real sentence.
    Categorypartner/image/Grok

    Inputs (7)

    NameTypeDefaultDescription
    modelCOMBO4 options: grok-imagine-image-2.0, grok-imagine-image-quality, grok-imagine-image-pro, grok-imagine-image
    promptSTRINGThe text prompt used to generate the image
    aspect_ratioCOMBO13 options: 1:1, 2:3, 3:2, 3:4, 4:3, 9:16, +7
    number_of_imagesINT11–10Number of images to generate
    seedINT00–2147483647Seed to determine if node should re-run; actual results are nondeterministic regardless of seed.
    resolutionoptCOMBO2 options: 1K, 2K
    qualityoptCOMBOQuality level, supported only by the grok-imagine-image-2.0 model.

    Outputs (1)

    NameTypeDescription
    IMAGEIMAGE