EvoLink Seedance 2.0 Image-to-Video (Official)

Text-to-video is a lottery; image-to-video is a bet on a known frame. If you have a still you like - a Nano Banana render, a photo, a frame from something else - Seedance 2.0 I2V is how you give it motion. EvoLink Seedance 20 Image-to-Video (Official) takes a required first_frame, optional last_frame, and a prompt, and returns a 4-to-15-second clip that starts from your image. Connect a last_frame too and it becomes a first-and-last-frame animation: the video has to move from A to B, which is the closest thing video models offer to editorial control.

This is a closed ByteDance model you can't download, so the API is the whole access path. The mechanism is the same cloud flow as every EvoLink node: submit, poll (default 30-minute timeout - it's a wait node), download the result into a VIDEO object.

What it is

The inputs:

• first_frame (required) - the starting image. Wire any image output in. Your local image uploads through EvoLink's channel and is auto-deleted after 72 hours.
• last_frame (optional) - the ending image. The moment you connect it, you're in first/last-frame mode. If the aspect ratios differ, the node auto-aligns the last frame to the first.
• prompt (required) - the motion/description. Chinese works; keep it under ~500 chars.
• duration - 4–15 seconds, default 5, billed by duration.
• quality - 480p / 720p / 1080p / 4k. In I2V the aspect ratio follows your first frame; quality is your resolution choice on top of it.
• generate_audio - on by default, no extra cost.

Outputs: video, task_id, status_info, result_urls (24-hour expiry), response_json. The video output must connect to a Save Video node or you'll get nothing on screen.

Install and key

Standard EvoLink install. ComfyUI Manager → search EvoLink → install → full restart. Manual:

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

Dependency is just requests; no models, no GPU. Key from evolink.ai/dashboard/keys (sk-), pasted once - saved locally, blank after, never in a shared workflow JSON.

Where this node earns its keep

The first/last-frame mode is the feature to plan around. Want a loop? Generate a first frame, run it as the last frame, and the animation has to come home - the oldest trick in the book, now with a cloud model that can actually honor it. Want a transition? Pick a second frame and write the motion prompt. One warning from the pricing side: this is the same meter as every Seedance call, and the community has clocked exactly how fast "a few clips for testing" becomes a month of subscription cost. Draft at 480p, promote the good ideas to higher quality, and treat the 4K dropdown the way you'd treat a very large tip.