Wan2.7 Image Pro

Here's the situation with Wan 2.7: it doesn't exist as open weights, and it never will. Alibaba stopped shipping Wan weights after 2.2 - 2.5, 2.6, and 2.7 are all API-only, and the community has given up asking. If a search result claims "Wan 2.7 open source download links," it's a content-farm trap with nothing behind it. So when you want the newest Wan image model, this node is the honest way in: Wan2.7 Image Pro runs the actual Alibaba model over Runware's cloud, and hands you a native IMAGE.

The "Pro" bit and the settings.thinking toggle tell you what kind of model this is. It's not a hobbyist checkpoint - it's Alibaba's flagship image generator, up to 4096px, with a reasoning pass it can run before drawing. That's the model's whole personality: give it room to think and it plans the composition; turn thinking off and it's faster but flatter.

The inputs that matter

Required: positivePrompt, plus width and height (768–4096, stepped by 16 - so 1024×1024 or 1536×1024 style multiples). Then the interesting stuff:

• settings.thinking (default on) - the quality/thinking-time trade. Leave it on for hero shots; it's the reason the model gets composition right.
• settings.sequential - image-set mode. Turn this on to get multiple related images that hang together as a coherent set (a character turnaround, a consistent lookbook) rather than independent variations.
• referenceImages - plug in a subject or style anchor.
• seed - for reproducibility, and numberResults up to 20 for variations.

The advanced_json escape hatch is documented as accepting settings.colorPalette and settings.editRegions - so if you want a specific palette or region-based edits, that's where they go. Output is one image.

How to actually use it

The workflow is the same shape as any cloud image node: prompt it, or wire a referenceImages input from an existing image in your graph, and send the image output into Preview Image or Save Image. Because it runs remotely, the 4096px ceiling is genuinely available to you - no OOM, no tiling, no VRAM prayer. If you've ever wanted 4K output from a single node without the local upscale dance, this is the shortcut.

Install and gotchas

Standard Runware pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/Runware/ComfyUI-Runware
pip install -r ComfyUI-Runware/requirements.txt

Restart, find it under Runware/Image/alibaba, and drop in your API key (Settings → Runware API key, or RUNWARE_API_KEY). No weights to download - the model is Alibaba's, running on Runware's GPUs.

Two cautions. First, every run costs money and the node prints it on the title bar - thinking mode and 4096px both push the bill up, so don't leave settings.thinking on and numberResults at 20 for a throwaway test. Second, remember this whole "Wan" family is closed; the freedom you got from Wan 2.2's Apache license is not here. That's the trade, and it's exactly why these cloud nodes exist - you get the model, you just don't get the weights.