EvoLink Video (All Models)

The dedicated EvoLink nodes - Sora 2, Veo 3.1, Wan 3.0, the lot - are the comfortable path: one model, honest parameters, nothing to think about. EvoLink Video is the other door. It's a single generic node that can call any video model on the platform - about 42 choices at the time of writing, including beta budget channels and video-edit / video-extend variants that don't have their own nodes. New model ships on EvoLink's side? This is the node that can reach it before a dedicated wrapper exists.

When to reach for it

Two situations. First, you're using a model variant that has no dedicated node - the beta budget channels, the edit/extend flavors, anything that shipped after the pack did. Second, you want to script across models: one node, swap the model enum, keep the workflow shape. The dedicated nodes win on discoverability; this one wins on coverage.

The catch is that generic has to pay for generic. Model-specific parameters that the dedicated nodes expose as real inputs - sound, seed, edit-target URLs, that sort of thing - go through the extra_params field as a JSON object that gets merged into the request body as-is:

{"quality": "2K", "video_urls": "https://.../clip.mp4", "seed": 42}

Check evolink.ai/docs for what each model accepts. The tooltip's warning is worth repeating: pick a quality tier a model doesn't support and it errors - the generic node can't hide what the model doesn't do.

The inputs that matter

• model - the big enum; that's the whole point of this node.
• duration - 0 = model default; -1 = Seedance 2.5/Wan auto-length. Otherwise seconds (up to 30).
• quality - auto or 480p / 720p / 1080p / 4k, model-dependent.
• aspect_ratio - blank for model default, or 16:9 / 9:16 / 1:1 / adaptive.
• image1 / image2 - reference or first/last frame, where the model supports it. Kling/Wan first-frame fans: the docs steer you to the dedicated I2V nodes instead, since their first-frame handling differs.
• extra_params - the JSON escape hatch above.

Outputs

video → Save Video, plus task_id, status_info, result_urls, response_json - the pack's standard five, all downstream-chainable.

Install and gotchas

Part of the comfyui-evolink pack. Manager → search EvoLink → Install → fully restart ComfyUI (close the console, not just refresh). 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.

The failure modes here are the price of the generic design: malformed extra_params JSON (the node rejects it with a clear message - don't guess, validate), and requesting parameters a chosen model doesn't support. Default timeout_seconds is 1800 because video is slow. Otherwise pack convention: 401/402/429 codes, free failed/moderation-blocked tasks, cached-and-unbilled reruns. If you're doing one model repeatedly, the dedicated node for it is usually the better fit - keep this one in the back pocket for the beta channels.