Grok Comfy Workflows

The starting point for Grok comfyui workflows is that Grok is a cloud model from xAI, so these workflows reach it through ComfyUI's API nodes rather than loading weights. There is no local download, which means you trade self-hosting for access to xAI's image and video stack without managing a GPU. The workflows on this page wire those calls into editable graphs, so the prompt, the reference images, and the output flow through nodes you can adjust and chain.

The image side is built on Aurora, xAI's autoregressive generator, which leans toward photorealistic results and renders text and logos in-frame more accurately than many models. You can generate from a prompt, request several variations in one run, or edit an existing photo by uploading it with a text instruction. The Grok Image Edit node accepts up to three source images, and xAI quotes outputs in roughly four seconds per image, so iterating on framing stays fast. On the video side, Grok Imagine turns a prompt into a clip and can animate a still image you provide, with native audio generated alongside the picture and clip length set in the workflow.

Because these are open ComfyUI graphs around a closed model, you keep control of the inputs that shape the result: the prompt wording, the reference images for editing, the number of outputs, and the aspect ratio. Partner nodes also add reference-to-video, which takes up to seven reference images for a consistent clip, and a video-extend node that lengthens an existing clip while keeping motion and sound continuous. The model itself is fixed and hosted by xAI, but everything around it in the graph is yours to tune, and the output drops straight into the rest of your ComfyUI pipeline.

To run one, open a Grok workflow on Comfy Cloud with nothing to install, since there is no local option. Write your prompt or upload the photo you want to edit, set the options the workflow exposes, and generate. Re-run with a revised prompt or different references until the framing and detail are right. xAI notes that Grok Imagine is still a beta under active development, so it pays to layer your own controls and combine its output with other nodes rather than relying on a single pass.