BYOKey Wan Image to Video

Here's the thing about Wan that makes this node different from most of the pack: Wan has open weights. It's the community's favorite open video model, the one people run locally and swear looks better than those early Sora demos. So this node isn't giving you access to a model you can't have - it's running an open model on someone else's GPU, Alibaba's, via the DashScope API. If you've got the VRAM, the local route is free and uncensored. If you don't, or you want the latest wan2.6 without a multi-gig download, this is your door.

How it works. Submit-then-poll against DashScope. POST to dashscope.aliyuncs.com (the default base_url - note the international endpoint is dashscope-intl.aliyuncs.com, and any relay can slot in here), with your DashScope/Model Studio key and your image sent as a base64 data URI. Poll until the clip is ready, download, wrap in ComfyUI's native VIDEO type.

The inputs that matter:

• image - required, and exactly one. Your video's first frame.
• prompt - supports English and Chinese; Alibaba's model reads either fine.
• model - wan2.6-i2v (default) or wan2.5-i2v-preview. Pick 2.6 for the current quality; the 2.5 preview exists for compatibility.
• resolution - 480P, 720P, 1080P - but Wan 2.6 doesn't do 480P, so that option quietly becomes an error on the default model. Mind the matrix.
• duration - 5, 10, or 15 seconds. 15 seconds is only on Wan 2.6; asking for it on 2.5 will fail.
• negative_prompt - genuinely supported, unusually for a video API. Use it.
• prompt_extend - on by default: DashScope rewrites/enhances your prompt with AI assistance.
• shot_type - single or multi, and it only takes effect when prompt_extend is on. multi means multiple shots with cuts rather than one continuous take - effectively a little edit baked into the generation.
• generate_audio - off by default; on means "if no audio input is provided, generate audio automatically." Turn it on if you want a track baked in.
• watermark - off by default; on tags the result with an AI-generated watermark.
• seed - standard; set one for reproducible rolls if the API honors it.

poll_interval and max_wait are the usual babysitters; the 10-minute cap is fine for a 15-second clip.

Output: one output (VIDEO) socket.

Install. Shared pack, one clone:

cd ComfyUI/custom_nodes
git clone https://github.com/MeteorAndy/comfyui-byokey.git
# restart ComfyUI; it's under api/byokey/video

No model downloads here - the model is Alibaba's, on their machine. No extra deps beyond ComfyUI's bundled torch, aiohttp, PIL, numpy and av.

Gotchas. The 480P-on-2.6 and 15s-only-on-2.6 constraints are the real traps; read the model-specific notes before queuing. And the pack is young with no community footprint - skim the source before feeding it a DashScope key.