SF WaveSpeed Wan 2.5 Text to Video Fast

Text-to-video from a prompt, on the fast lane, with zero local compute. That's the whole job of this node: Wan 2.5's hosted text-to-video on WaveSpeed's fast endpoint. Type a scene, pick a size and duration, hit run, and a video_url comes back. It's the node you reach for when you want to riff - generate a dozen scene ideas as short clips, keep the ones that work, and spend real money only on the survivors.

Wan 2.5 is API-only, full stop - Alibaba stopped shipping open weights at 2.2, and 2.5 is one of the versions the community rated as a genuine improvement over its predecessor. WaveSpeed was the first platform to host it. This node is just the front door to that hosted model, and the "Fast" in the name means the cheaper, quicker endpoint rather than a local speed hack.

How it works

Standard pack anatomy: an SF WaveSpeed Client wired into client, a prompt, and the node POSTs to /api/v3/alibaba/wan-2.5/text-to-video-fast, then polls until the render finishes. Output is a single video_url string. No model downloads, no VRAM, no checkpoints - your machine's only job is holding the API key and waiting.

The inputs that matter

• prompt - the scene description. With a fast endpoint, prompt quality matters more than settings, because you're trading quality for speed and the prompt is all you've got left.
• size - four presets: 1280*720 (default, landscape), 720*1280 (portrait), 1920*1080 and 1080*1920 (the HD pair). There's no freeform size field here; pick from the list.
• audio - an optional audio URL (wav/mp3, 3–30 seconds, ≤15MB) that guides the video. Leave it blank for silent motion.
• duration - 5 or 10 seconds.
• enable_prompt_expansion - off by default; turns on the API's automatic prompt enhancement.
• negative_prompt, seed, enable_sync_mode - the standard trio, all behaving exactly as their tooltips say.

Output: video_url, a string you can open in a browser or feed to a URL-consuming node.

Installing it

ComfyUI Manager, search comfyui-stillfront, install. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/razvanmatei-sf/comfyui-stillfront.git
cd comfyui-stillfront
pip install -r requirements.txt

Grab a WaveSpeed API key from wavespeed.ai and set it via config.ini (copy config.ini.tmp first), the WAVESPEED_API_KEY environment variable, or the SF WaveSpeed Client's api_key widget. The README's stillfront/... clone URL is outdated; the repo actually lives at razvanmatei-sf/comfyui-stillfront.

Where people get burned

The classic: you search "wavespeed comfyui", find this pack, and somewhere in the results there's also ComfyUI-WaveSpeed (chengzeyi's local acceleration pack). Totally different project - that one makes your own GPU run diffusion faster; this one talks to an API. Don't install the wrong one and wonder why the nodes don't match.

Otherwise it's the standard hosted-service cautions: pay-per-generation (so test at 1280*720/5s, not 1080p/10s), a bad key fails fast with Unauthorized, and there's no local fallback - if the API's down or your credits run out, the node has nothing to fall back on. And if you need a subject to stay consistent across shots, this text-only node isn't the tool - that's what the pack's image and reference variants are for.