Modelverse GPT Image 1

GPT Image 1 is OpenAI's flagship image generator - the one behind those absurdly coherent text-in-image results that made people screenshot their chats. This node hands it to you inside ComfyUI through UCloud's Modelverse API, which means you get the model's talent for typography and literal prompt-following without an OpenAI key or a second billing relationship. One UCloud key runs the whole pack, and this is one of the two GPT Image 1 nodes in it (the Edit variant handles masked edits; this one is pure generation).

What it is

Text-to-image, OpenAI style. Prompt in, image out. Where most diffusion models mangle text that appears in images, GPT Image 1 is famously good at rendering readable words, logos, and signage - it's the node to reach for when your prompt contains actual typography or when you need the model to take a long, literal instruction at face value.

The inputs that matter

• client - required, from the Modelverse Client / Secret Client node.
• prompt - the image description. GPT Image 1 rewards long, explicit instructions; if it involves text in the image, spell the text out in quotes.
• size - 1024x1024, 1024x1536, or 1536x1024.
• quality - low/medium/high. This is the real dial: quality moves output and price together.
• output_format - png or jpeg.
• output_compression - 0–100, default 100.
• num_requests / num_images - batch knobs and bill multipliers.

Output is a single image (IMAGE tensor), straight into preview or save.

Installing and keying up

The whole pack installs together. ComfyUI Manager → search "ComfyUl-UCloud" → Install → restart, or:

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

restart ComfyUI. Dependencies are light (requests, pillow, numpy, openai, GitPython); no model downloads, since GPT Image 1 runs on UCloud's servers. Get a ModelVerse API key from the UCloud console and enter it via the Modelverse Secret Client node (secret name in the workflow, key in secrets.json) or a config.ini MODELVERSE_API_KEY=... line.

Common issues

• Text in the image comes out wrong - GPT Image 1 is good at text but not psychic. Quote the exact string you want and keep the request short; long paragraphs are where it slips.
• "API key is not set" - empty client node. The eternal first check.
• Low quality is genuinely low - it's a draft tier, not a "you won't notice" tier. Use it for layout exploration, medium+ for anything you'll show anyone.
• Costs - quality and size both multiply the bill. High quality at 1536 is the expensive lane; iterate at medium.
• Different look from local models - obviously. Don't expect SDXL/Flux aesthetics; GPT Image 1 has its own editorial, slightly photographic house style.

The workflow that makes the most of it: generate the typography-heavy hero image here, then pass it to the Edit node (or a Kontext node) for refinement. Together they're the closest thing this pack has to an OpenAI studio.