KIE Grok Imagine (I2I)

Grok Imagine's image-to-image mode is the edit sibling of the pack's Grok T2I node: source image in, reworked image out, with a task_id you can hand off to Grok's video model. If you want the Grok family's style and its less-restrictive content stance applied to your image rather than a fresh generation, this is the node.

Like its T2I sibling, the interesting output isn't the image - it's the task_id. The pack is explicitly built around the chain: Grok I2I produces the edited image and its server-side task reference, and KIE Grok Imagine (I2V) can consume that id directly with an index to pick which result animates. Image generation and video generation stay on the same side of the API, no download/re-upload in between.

How it works

Standard KIE image job flow: the node uploads your source image, POSTs a createTask with the grok-imagine/image-to-image model id plus your (optional) prompt, polls recordInfo every 10 seconds until it resolves (300-second timeout, auto-retry on transient errors), then downloads the result into a ComfyUI IMAGE batch and returns the task id from the server.

The inputs that matter

• images - required. Your source. Note the pack uses the first image in the batch - the I2I side of Grok isn't the multi-reference machine that GPT Image 2 I2I is; it's a one-source edit.
• prompt - optional, and genuinely optional - the model will do a style/format pass on the source even with an empty one. Give it a real instruction to steer the edit: "make it a neon-noir poster" beats nothing.

Two outputs:

• image - the edited result as an IMAGE batch, for preview/save/upscale.
• task_id - a STRING, the Grok task reference. Feed it into KIE Grok Imagine (I2V) with an index (0–5) to animate a specific result.

Installing

Installs with the whole pack. ComfyUI Manager: search "ComfyUI Kie API". Or:

cd ComfyUI/custom_nodes
git clone https://github.com/gateway/ComfyUI-Kie-API
# restart ComfyUI

No model files. API key at config/kie_key.txt in the pack folder (copy from kie_key.example.txt).

Where people get burned

The single-source limitation catches people who expect to wire a whole batch of references in - only the first image is used, so build your input deliberately. The task_id is a live handle, not a saved asset: chain it into Grok I2V while it's fresh. And the usual credit discipline applies - editing at image-model prices adds up, and a dead balance produces failures that look like node bugs. Check Get Remaining Credits first, and if a job does die, kie.ai/logs has the server-side reason while the console shows each state change.