Alibaba Wan 2.7 Text To Image (Official Edition)

The Wan image model is Alibaba's best answer yet to the question "what do you run when you want quality but don't have the VRAM for it?" It's closed-source and API-only - exactly the model that makes sense to rent per-generation rather than curse at on your 6GB card. This node drops it into a ComfyUI workflow as a drop-in text-to-image block: type a prompt, get an image back, no local model files anywhere.

It's part of BizyAirPlus, SiliconFlow's cloud-execution pack for ComfyUI. You keep editing the graph locally, flip the BizyAirPlus button ON in the action bar, and the node sends your prompt to the cloud and pulls the finished image back into your session as a normal image output you can wire into SaveImage - or into anything else downstream.

How it works

The image output is the thing you'll actually use. The node also returns request_id (for chasing failed jobs) and urls (hosted file links). Bilingual prompts are fine - Chinese and English, up to 5000 characters - which is a nice nod to where this model comes from.

Two required inputs: prompt and size. The size tooltip lays out the tiers: 1K = 1024², 2K = 2048², and 4K = 4096², with 4K flagged "(Pro)" - the base tier stops at 2K. Default is 2K, which is honestly the sweet spot for most work; 1K is fine for iteration, and you pay more as you climb.

The inputs that matter

• prompt - the whole job. Up to 5000 characters, Chinese or English.
• size - 1K / 2K / Custom (4K is Pro-only). Start at 1K while you're learning the API, bump to 2K for finals.
• thinking_mode - on by default. The tooltip describes it as enhancing the model's reasoning to improve image quality at the cost of time. Turn it off when you're iterating fast; leave it on for the images that matter.
• seed - 0–2147483647, with -1 meaning server-random. Same seed keeps output "relatively stable."
• custom_width / custom_height - only take effect when size is Custom. Watch the pixel budget.
• color_palette - an optional JSON array of 3–10 hex colors with ratios summing to 100% (8 colors recommended). This is the color-theming knob; it only applies when group mode is off.

Plus watermark, skip_error (continue with a placeholder on failure), and api_config (override endpoint/key for one request).

Installation

One install for the whole pack. ComfyUI Manager, search BizyAirPlus, or:

cd ComfyUI/custom_nodes
git clone https://github.com/siliconflow/BizyAirPlus.git
cd BizyAirPlus
python -m pip install -r requirements.txt

Restart ComfyUI, then add a bizyair.ai API key - via the prompt when you turn BizyAirPlus ON, Settings > BizyAirPlus > API Key, or the BIZYAIR_API_KEY environment variable. Nothing heavy downloads; execution is cloud-side.

Common issues

• "4K" isn't available - that's expected on the base tier; the tooltip marks 4K as Pro-only. Use the Pro variant of this node if you need it.
• Color palette rejected - the ratios must sum to 100%, and it's ignored in group mode. Validate your JSON before blaming the node.
• Slower than expected - thinking_mode on and big sizes add time. Both are tradeoffs, not bugs.
• Cloud failures - key validity and network, then the ComfyUI console for details.