Superside Grok Imagine Image Quality Edit

xAI's Grok Imagine, in the "quality" flavor, wrapped as an edit node - up to three reference images, an aspect ratio dial, and a genuinely nice feature at the end: it hands back the revised prompt the model actually generated with. You don't usually get that from an edit model, and it's the best debugging tool in the pack.

Inputs: prompt, image_1, and api_key required; image_2 and image_3 for extra references; aspect_ratio (14 options including auto, the default, plus everything from 2:1 down to 9:19.5); resolution (1k or 2k, default 1k); output_format (jpeg/png/webp); num_images (1–4); and sync_mode. Outputs are images (IMAGE) and revised_prompt (STRING).

That second output is worth a paragraph of its own. Many edit models silently rewrite or expand your prompt before generating - "make this a sunset" becomes a paragraph about golden-hour backlighting - and you'd never know unless you compared frames. This node surfaces it: wire revised_prompt into a text preview or a save node, and you can see exactly what the model decided you meant. If the output keeps drifting from your intent, the revised prompt usually tells you why, and it's also a great source of effective prompting style for your next run.

The resolution knob is simple because this is the quality tier: 1k or 2k, that's it. If you need bigger outputs, that's what an upscaler is for - the pack even has Crystal Upscaler for portrait work and the local resize tools for the rest. aspect_ratio at auto follows the input image's shape, which is the right default for editing; pick a fixed ratio when you want a deliberate reframe.

Honest positioning: Grok Imagine is a solid, characterful edit model and it's open about its rewriting (hence the revised-prompt output), but it's also one closed model among several in this pack doing the same job - GPT Image 2 has the mask modes and the 4K ceiling handling, Nano Banana's a different personality entirely. If you already have one of those wired and working, this node isn't a strict upgrade; it's the pick when you specifically want xAI's take on an edit, or when you want the revised-prompt visibility for prompt iteration.

It's a fal call, so the pack's standard rules apply: paste your key into api_key (blank falls back to FAL_KEY, both blank = immediate error), calls burn metered credits, and your images leave your machine. xAI's moderation rides along with the model - if a prompt gets refused, that's the closed model's filter, not something the node can bypass.

Install - ComfyUI Manager (search "comfyui-superside-nodes") or:

cd ComfyUI/custom_nodes
git clone https://github.com/Superside/comfyui-superside-nodes
pip install -r requirements.txt

Restart ComfyUI, find it under Superside. No model downloads - requirements.txt is fal-client, pillow, numpy, torch, requests.