EvoLink Wan 3.0 Image-to-Video (Official)

The open Wan you can run locally stopped at 2.2, so everything newer - including Wan 3.0 - lives behind an API. The EvoLink Wan 3.0 Image-to-Video node is the official wrapper for the first/last-frame flavor of that API: you give it a real frame, optionally a closing frame, and it animates between them. It's the middle child of the Wan 3.0 family - more control than text-to-video, less plumbing than the reference node - and it's where most people end up.

How it works

Standard EvoLink flow: submit to /videos/generations, poll, download before the 24-hour links expire, return a native VIDEO output. first_frame is a required IMAGE input - the tooltip's the authority here: it accepts 240–8000px at ratios from 1:8 to 8:1, which is a wide door. last_frame is optional but must be paired with first_frame. Images upload through EvoLink's own file channel, so a Load Image node straight into first_frame is all the setup there is - no imgbb, no URL paste.

Model pick: wan3.0-image-to-video or the wan3.0-prime-image-to-video step-up. auto_duration is the Wan-specific trick - the model picks the length and you're refunded to the actual output, so letting it decide isn't a gamble.

The inputs that matter

• first_frame - required. This is your starting image.
• last_frame - optional; fixed ending. A closed first→last loop is the demo worth running.
• prompt - required, up to 20,000 characters, Chinese-friendly.
• duration - 2–30s integer, billed per second.
• quality - 480p (cheapest billing base), 720p (~2×), 1080p (~4×).
• aspect_ratio - adaptive default, or 16:9 / 9:16 / 1:1 / 4:3 / 3:4.
• generate_audio - audio on/off; same price either way on the Wan nodes.
• seed - 0 = random, fixed = reproducible.

Outputs

video → Save Video, plus task_id (receipt for the job log at evolink.ai), status_info (summary), result_urls (24h raw links), response_json (full payload). The whole pack shares this shape, so the clip chains into the Topaz upscale or an edit node with no adapter.

Install and gotchas

Part of the comfyui-evolink pack. Manager → search EvoLink → Install → fully restart ComfyUI (close the console window; refreshing the tab isn't enough). Or:

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

Only requests as a dependency; no models, no GPU. API key from evolink.ai/dashboard/keys, fill once, blank afterward so the key doesn't ride along in shared JSON.

This is the rare EvoLink node with a required IMAGE input, so if it errors on missing first_frame, check the wiring before you suspect the API. Otherwise pack convention rules: 401 bad key, 402 no credit, 429 rate limit; failed and moderation-blocked tasks don't charge; unchanged reruns are cached and unbilled. Want subject/style consistency beyond a fixed start/end? The Wan 3.0 Reference sibling handles mixed image/video/audio references instead.