BYOKey Wan Text to Video

Wan is the open-weights video model the community actually loves - the one people run on their own cards and hold up against the closed giants. This node doesn't wrap a secret model; it runs Wan on Alibaba's own DashScope service with your key, so you get the current generation (wan2.6-t2v) without downloading multi-gigabyte weights or budgeting the VRAM. If you can run it locally, local is still free and unfiltered - this is the convenience door for everyone else, and the one that lets you hit the latest version day one.

How it works. The standard submit-then-poll dance. POST to dashscope.aliyuncs.com (default base_url; the international endpoint is dashscope-intl.aliyuncs.com, and relays slot in here too) with your DashScope/Model Studio key, poll every poll_interval seconds until the clip is ready, download the mp4, wrap in ComfyUI's native VIDEO type.

The inputs that matter:

• prompt - the whole creative input. Wan reads English and Chinese natively, so write in whichever you think in.
• size - this is the distinctive control: 13 presets that bundle resolution and aspect ratio into one pick, e.g. 480p: 1:1 (624x624), 720p: 16:9 (1280x720), 1080p: 9:16 (1080x1920). Pick your target platform's format and it's handled in one field.
• model - wan2.6-t2v (default) or wan2.5-t2v-preview. 2.6 is the current quality bar.
• duration - 5, 10, or 15 seconds; 15 is only on 2.6, so don't ask for it on the 2.5 preview.
• negative_prompt - supported, unusually for a video API. Describe what to avoid.
• prompt_extend - on by default: DashScope rewrites your prompt with AI assistance.
• shot_type - single or multi, and only meaningful while prompt_extend is on. multi gives you multiple shots with cuts instead of one continuous take - a cheap way to get an "edited" feel from a single call.
• generate_audio - off by default; on auto-generates an audio track.
• watermark - off by default; on tags output with an AI-generated watermark.
• seed - set one for reproducible rolls.

poll_interval and max_wait are the babysitters; the 10-minute cap is generous.

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 (Alibaba hosts the model), no extra deps beyond ComfyUI's bundled torch, aiohttp, PIL, numpy and av.

Gotchas. The 15-second-only-on-2.6 rule and the model/resolution interactions are the fiddly bits - read the tooltips when you switch models. And the usual fresh-pack hygiene: it's new, no community consensus behind it, so skim the source before handing it a DashScope key.