Modelverse GPT Image 1 Edit

GPT Image 1 is OpenAI's image generation model, and the Edit variant is the one with a rare luxury in this pack: an actual mask input. Most of the editing nodes here trust the model to figure out what to change from your prompt alone; this one lets you point. Through UCloud's Modelverse API, that means a genuinely surgical editing node - change one object in a busy photo and leave the rest untouched. If you've been through the local inpainting wars, you know how much of a win that is.

What it is

An image editor with optional mask-guided inpainting. Feed it an image and an instruction, optionally a mask marking the region to change, and it returns the edited image. The mask is what separates it from the "describe the whole change" nodes: wire a mask and the model limits its work to that area, which is how you get "add a beach ball in the center" (the shipped default prompt) without the model redesigning your entire photo.

The inputs that matter

• client - required, from the Modelverse Client / Secret Client node.
• prompt - the edit instruction.
• image - required. The input to edit.
• mask - optional, and the star feature. Any standard mask tensor works - from a segmentation node, an inpaint mask, or a hand-painted one.
• size - 1024x1024, 1024x1536, or 1536x1024.
• quality - low/medium/high. The bill climbs with quality; medium is a fair default.
• output_format - png or jpeg.
• output_compression - 0–100, default 100 (least compressed).
• num_requests / num_images - batch knobs and bill multipliers.

Output is a single image (IMAGE tensor), ready to preview or save.

Installing and keying up

Standard pack install. ComfyUI Manager → search "ComfyUl-UCloud" → Install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/ucloud/comfyui-plugin

restart ComfyUI. Light deps (requests, pillow, numpy, openai, GitPython), no model downloads - GPT Image 1 runs on UCloud's side. Get a ModelVerse API key from the UCloud console and enter it via the Modelverse Secret Client node (name in workflow, key in secrets.json) or a config.ini MODELVERSE_API_KEY=... entry.

Common issues

• Mask ignored / whole image changed - no mask wired means prompt-guided editing of everything. If you want surgical, the mask is mandatory.
• Mask is white-on-black vs black-on-white - conventions differ across nodes. If the edit targets the opposite region, invert the mask.
• "API key is not set" - empty client node, always the first suspect.
• Edits beyond the mask anyway - GPT Image 1 uses the mask as a strong hint, not a hard fence. Keep the prompt scoped to the masked area.
• Costs - quality and size both move the meter; high quality at 1536 is the expensive lane.

The honest take: this is the pack's most surgical image node, and the mask input makes it the one that plays nicely with the rest of ComfyUI - you can build a whole segment-and-edit pipeline around it. If you're comparing it to local inpaint workflows, that's the one advantage the cloud actually can't fully take away: you still get to draw where the edit happens.