Modelverse Flux Kontext Max

This is the node the pack's README calls "flagship," and it earns the title. Flux Kontext Max is Black Forest Labs' editing model with a long context window - it can genuinely hold several images at once and act on them as a set. Through UCloud's Modelverse API, that means you can point it at a batch of frames and say "combine these into a single image" or "make all three the same style" and it will. No ControlNet stacking, no compositing nodes, no 16GB of Flux weights on your drive.

What it is

A single-image and multi-image editor that switches modes automatically. Wire in one IMAGE tensor and it does single-image editing. Wire in several - via the Modelverse Image Packer, which batches up to five images into one tensor - and it runs multi-image mode, treating the whole batch as context. The README's example workflows show exactly this: the Max node with a packer feeding it a group of images for a "combine these" edit.

The inputs that matter

• client - required, from the Modelverse Client / Secret Client node.
• prompt - the instruction. In multi-image mode, describe the relationship you want: "merge these three into a single product shot," "unify the lighting."
• images - required. A single IMAGE, or a batched tensor from Modelverse Image Packer for multi-image mode.
• guidance_scale - 2.5 default (0–10). Max is an editing model; gentle values work.
• seed - -1 for random, or fix it.
• num_requests - 1–10. The bill multiplier.

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

Installing and keying up

Same one-plugin install as the whole pack. 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 files - Kontext Max runs on UCloud's side, which is the only place it realistically runs for a home user. Get a ModelVerse API key from the UCloud console and enter it via the Modelverse Secret Client node (name only in the workflow, key in secrets.json) or a config.ini MODELVERSE_API_KEY=... line.

Common issues

• Multi-image mode doesn't trigger - the node detects it from the shape of the input tensor. If you fed one image that happens to be a batch of one, you get single-image mode. Use Modelverse Image Packer with at least two inputs to force multi.
• "Input image(s) is required" - the images input is required; you can't run this node prompt-only. That's what the T2I variant is for.
• "API key is not set" - empty client node, first thing to check.
• Edits blend more than you expected - multi-image mode assumes you want the images to relate. If you need surgical edits to one image, do it in single-image mode instead.
• Costs - Kontext Max is premium-priced; multi-image runs cost more than single. Draft the prompt, then spend the run.

If you've ever wanted to skip the six-node identity-merge pipeline and just describe the merge, this is the node. It's the reason the pack ships an Image Packer at all.