Grok Imagine Image

Grok Imagine Image is xAI's text-to-image model - the one behind the "imagine" feature in the Grok app - exposed as a ComfyUI node (xai:grok-imagine@image) that runs on Runware's cloud. It's a closed model with no weights to download, so for most people this node is the way to use it inside a ComfyUI workflow. No GPU, no checkpoint folders, no VRAM concerns; just a prompt, an API key, and a pay-per-image run.

The pack itself is the enabler: install once (ComfyUI Manager → search Runware, or git clone https://github.com/Runware/ComfyUI-Runware into custom_nodes/), add a Runware API key in ComfyUI Settings, and every model the API hosts shows up in your node menu. This node is one of the more interesting members of that catalog because it's a mainstream closed model most people only ever touched through a chat app - now it's a graph node like any other.

What it's good at

xAI's image model carries the Grok personality: it follows natural-language prompts with less hedging than most, and it has solid reference-image conditioning. If you've seen Grok's image output in the wild - the reliably good prompt-following, the character consistency in feeds - this is that same backend without the chat UI. The referenceImages socket is the power move: wire in a subject photo and keep that identity across prompts.

Inputs that matter

• positivePrompt (required) - write it like you're talking to a very literal assistant: subject, style, framing, lighting. This model rewards descriptive prose.
• referenceImages - IMAGE socket for conditioning on existing images (style or subject reference).
• size - 15 presets, default 1K (16:9). Pick aspect ratio and resolution in one dropdown.
• numberResults (1–20) - batch variations; each uses a different seed.
• watermark - a builder socket for the Runware watermark feature.
• safety gate (checkContent) and the standard outputFormat/outputQuality.

The image output is a native IMAGE tensor, straight into Preview or Save - nothing exotic, which is exactly the point.

The honest assessment

• It's a closed model on a metered API. Per-image cost shows on the title bar, and the economics are pay-as-you-go. Fine for real work, less fine if you were hoping for unlimited tinkering.
• No seed control beyond numberResults. This node doesn't expose a seed widget, so "same prompt, same image" reproducibility isn't the workflow - you generate, pick, and move on.
• Cloud round-trip. Fast, but it's a network call; don't expect local-render latency.
• Reference images matter. Feed it a strong subject reference and the results are noticeably better than prompt-only. It's the difference between "a woman" and "the woman you've been building consistency for."

Grok Imagine Image is a "nice to have" node for most ComfyUI users - you already have good open models locally. But for Grok-specific prompt-following, for consistency work via reference images, or for machines that can't run a local model at all, it's the easiest possible way in.