EvoLink Wan 3.0 Text-to-Video (Official)

Here's the awkward part of Wan's story: the open Wan you can run locally froze at 2.2, and everything newer from Alibaba shipped API-only. Wan 3.0 is the newest of that closed line, so if you want it, you call it - and the EvoLink Wan 3.0 Text-to-Video node is the official way to do that from inside ComfyUI. It's also one of the most budget-friendly video calls in the pack, which is a genuinely useful thing to have.

What it does

Pure text-to-video: a prompt in, a clip out. Up to 30 seconds, which is longer than most of the pack's video nodes allow, and it's billed by the second with a 480p tier that's the cheapest video generation in this pack. There's a prime model variant (wan3.0-prime-text-to-video) alongside the base wan3.0-text-to-video - prime is the step-up tier if the standard model's quality isn't doing it for you.

Mechanically it's the pack standard: submit to /videos/generations, poll, download before the 24-hour result links expire, return a native VIDEO output. One nice Wan-specific trick: auto_duration. Flip it on and the model picks the clip length based on your prompt and footage instead of making you guess - you're pre-charged at the 30s ceiling but refunded to the actual output length, so it's not a gamble.

The inputs that matter

• prompt - required, up to 20,000 characters, Chinese-friendly.
• model - base vs. prime.
• duration - 2–30s, any integer, billed per second. Your main cost lever.
• auto_duration - let the model choose the length (see above).
• quality - 480p (cheapest billing base), 720p (~2× 480p), 1080p (~4×). The tooltip says it plainly, so there's no mystery about where your credits go.
• aspect_ratio - adaptive (default, follows prompt/footage) or 16:9, 9:16, 1:1, 4:3, 3:4.
• generate_audio - audio track on/off; unlike Veo, the Wan nodes price this the same either way.
• seed - 0 is random; a fixed seed reproduces results.

Outputs

video → Save Video. Plus task_id (job-log receipt), status_info (summary), result_urls (24h links), response_json (full payload) - chainable into the edit/upscale nodes downstream.

Install and gotchas

Part of the comfyui-evolink pack. Manager → search EvoLink → Install → fully restart ComfyUI (close the console; a browser refresh doesn't count). Or:

cd ComfyUI/custom_nodes
git clone https://github.com/deeplearning-goethe/comfyui-evolink

Only requests as a dependency; no models, no GPU. Key from evolink.ai/dashboard/keys, fill once, blank afterward so it stays out of shared workflow JSON.

The duration field has a quirk: the generic EvoLink Video node uses 0/-1 for "model default"/"auto", but this dedicated node uses a plain integer plus a separate auto_duration switch - they're the same feature wearing different clothes. Everything else is pack convention: 401/402/429 error codes, free failed/moderation-blocked tasks, cached-and-unbilled reruns. If you have a first frame in hand, the Wan 3.0 Image-to-Video sibling is the one to grab; the Reference node handles mixed image/video/audio inputs.