Modelverse Flux Kontext Max text2image

Every pack has a node that answers "what's the best image this thing can make with zero input beyond a prompt?" - for this pack it's Flux Kontext Max text2image. It's the text-to-image face of the same flagship Kontext Max model the editing node uses, stripped of the image inputs so all it needs is a sentence and an aspect ratio. Through UCloud's Modelverse API, of course: the model runs on their hardware, you run the per-generation meter.

What it is

Pure text-to-image with Kontext Max's quality ceiling and a deliberately simple interface. Where the editing node asks for images and a prompt, this one asks for just a prompt plus aspect ratio and batch knobs. It's the node to reach for when you want the best-looking single image the pack can produce from text alone - product hero shots, editorial-style illustrations, that one image you're going to look at all day.

The inputs that matter

• client - required, from the Modelverse Client / Secret Client node.
• prompt - everything rides on this. Kontext Max follows long, descriptive prompts well; write the image, not a shopping list.
• aspect_ratio - 21:9 down to 9:21, default 1:1. This is your composition control since there's no width/height pair here.
• num_images (1–4) and num_requests (1–10) - batch knobs and bill multipliers.
• guidance_scale - 2.5 default (0–10). Keep it gentle; Kontext is an editing-family model and responds to lower CFG.
• seed - -1 for random; set to reproduce a winning composition.

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

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. Dependencies are light (requests, pillow, numpy, openai, GitPython); no weights to download, because the model lives server-side. The setup step is a ModelVerse API key from the UCloud console, entered via the Modelverse Secret Client node (name in workflow, key in secrets.json) or a config.ini MODELVERSE_API_KEY=... entry.

Common issues

• Aspect ratio surprise - there's no width/height, so the aspect_ratio dropdown is the only shape control. Set it before you queue.
• "API key is not set" - empty client node. Always the first suspect.
• Results feel different from Flux Dev - it's a different (stronger) model family. Don't expect Dev's exact look; expect more editorial polish.
• Soft output at extreme ratios - 21:9 and 9:21 are at the edge of the model's comfort zone; if they come back soft, crop instead of stretching.
• Costs - Kontext Max is the premium tier; num_requests and num_images both multiply the bill. One good run beats five mediocre rerolls.

The workflow that earns its keep: use the T2I node to generate a hero image locally-cheap, then hand that image to the Kontext Max editing node for refinement. It's the cheapest way to spend flagship credits.