EvoLink Gemini Omni Flash Image-to-Video (Official)

You've got a good image - a portrait, a scene, a frame from a storyboard - and you want to see it move. That's the whole job of EvoLink_GeminiOmniI2V: give it one starting frame and a line of description, and Google's Gemini Omni Flash animates it into a clip of up to 10 seconds. It's Google's fast, cheap image-to-video tier, the budget sibling of Veo, and it's the node to reach for when you're iterating on ideas rather than doing a final render.

The inputs that matter

Two of them actually matter:

• prompt (required) - describe the motion and the shot. Keep it about what happens and how the camera behaves; this is a short-clip model, not a feature-film director. Chinese prompts work fine.
• first_frame (required) - the IMAGE input. Load Image → this socket. That's the entire setup; the aspect ratio follows what you set (below).

Then a couple of dials:

• duration - 3 to 10 seconds, any integer, default 10.
• aspect_ratio - 16:9, 9:16, or auto (default 16:9). auto lets the model adapt to the input frame.

The usual api_key (fill once, stored locally, leave blank after) and timeout_seconds (default 1200) round it out.

How it works

The pack uploads your frame through EvoLink's own file channel (files-api.evolink.ai) - no third-party image host, and the file is auto-deleted after 72 hours - then submits to POST /v1/videos/generations and polls the task until it completes, showing a progress bar on the node. The finished clip is downloaded to your machine as a VIDEO object, so you connect the video output to a native Save Video node to write it to disk. Alongside it you get task_id (for the task logs), status_info (model, runtime, credits used), result_urls (raw links, 24h expiry), and response_json.

One caveat that trips people up: because the video is returned in-memory as a VIDEO type, the pack needs a recent ComfyUI that ships comfy_api's VideoFromFile. On an old install you'll get an explicit error about missing VIDEO support - the fix is updating ComfyUI, not reinstalling the pack.

Installing

No GPU, no model weights, no heavy dependencies. The only Python requirement is requests:

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

then fully restart ComfyUI. Or search EvoLink in ComfyUI Manager. If the nodes don't appear, you either have a nested-folder problem (the pack's __init__.py must sit directly in custom_nodes/comfyui-evolink/) or you didn't fully restart.

Troubleshooting

401 bad key, 402 out of credits, 429 rate limited. A 403 means your account doesn't have access to this model - worth knowing if you're on a fresh account. If you're behind a VPN, whitelist api.evolink.ai and files-api.evolink.ai for direct connection; the pack's own code flags this as a known failure. And remember the "generated but I don't see anything" trap: you must connect the video output to something that previews or saves it. Rerunning after a fix doesn't double-charge - unchanged params reuse ComfyUI's cache.