🤖 Leon Wan Image API

You probably know Wan as the video family - Alibaba's Apache 2.0 video models that became the base for half the open video work of 2026. But the Wan brand now covers an image model too, and Leon_Wan_Image_API_Node is how you get wan-2.7-image (or its -pro sibling) into a ComfyUI graph: type a prompt, pick a size, and the HyprLab aggregator routes it for you. No local weights, no VRAM, no wan-2.7 checkpoint hunt - just a remote render that lands in your workflow as a normal image tensor.

That "no local weights" bit is doing a lot of work, so be clear-eyed about the trade. The Wan image models in this node are the newer, API-only generation - the KB's Wan-video panel tracks the open lineage as frozen at Wan 2.2. If running Wan locally matters to you, the open weights are still out there; this node is for the other 90% of the time, where you just want the result without the infrastructure.

How it works

Identical plumbing to the pack's other API nodes: the shared HyprLabImageGenerationNodeBase builds an OpenAI-style payload, POSTs to https://api.hyprlab.io/v1/images/generations with a Bearer token, retries with exponential backoff (five attempts max), and converts the response to an RGBA tensor. The Wan-specific parts are the size enum and the image inputs.

Inputs that matter

• model - wan-2.7-image (default) or wan-2.7-image-pro. The pro tier costs more for a quality bump; start on the base.
• size - 18 choices, and here's where beginners get confused: the enum mixes raw pixels (1024*1024, 2048*2048, 1280*720, 720*1280…) with tier labels (1K, 2K, 4K). They're all valid selections; 1024*1024 is just the explicit spelling of 1K.
• image / image_url / input_images_array - Wan Image is a capable editor, so it takes reference images three ways: a single tensor, a hosted URL, or up to nine images through the pack's Image Array Builder (IMAGE_ARRAY). Give it at least one reference when you're editing rather than generating fresh.

Outputs are the pack standard: image, image_url, seed.

Installing

Same pack as every Leon node - ComfyUI Manager (search Leon) or clone:

cd ComfyUI/custom_nodes/
git clone https://github.com/l3ony2k/comfyui-leon-nodes comfyui-leon-nodes
pip install -r requirements.txt

Restart, paste a HyprLab key. Light dependencies, zero downloads.

Common issues

The defaults will get you 90% of the way; the usual suspects for the other 10% are a left-at-default api_key and the "give me either a tensor or a URL, not both" refusal when you double-wire the image inputs. Sizing is the genuinely Wan-specific gotcha: if you pick 4K on a plan or model that only serves lower tiers, expect a 4xx - check the console log for the exact error body. And as always with these nodes, it's metered, so settle on your size and reference before running, not after two paid 4K experiments.