ByteDance Seedance 2.5 First-Last-Frame to Video

First-last-frame generation is the closest thing to "directed" video you'll get from an API. Instead of hoping the model animates your whole idea from a prompt, you hand it the exact first frame and (optionally) the exact last frame, and Seedance 2.0 choreographs a believable path between them. ByteDance2FirstLastFrameNode is ComfyUI's built-in wrapper for that, and it's the node to reach for when you know how a shot should start and end and want everything in between to respect both.

It's a partner/API node like the rest of the ByteDance family: nothing runs locally, you need a logged-in Comfy account, and each call is billed through credits. The KB's ByteDance panel is the useful context here - Seedance is ByteDance's flagship, API-only video line, so this is the paid-but-polished route while the open Wan-derivatives cover the free local ground.

The inputs

• model - the interesting one. It's a dropdown that carries its own sub-options: Seedance 2.0, 2.0 Fast, 2.0 Mini, and 2.5. Pick one and it exposes a prompt (with resolution, ratio, duration, and audio toggles) that describes the motion between your frames. The model choice is quality-vs-cost: 2.0 for quality, Fast for speed, Mini for the cheapest quick takes.
• first_frame - the image the video starts on.
• last_frame - the image the video ends on. Optional: with only a first frame, it behaves like plain image-to-video.
• first_frame_asset_id / last_frame_asset_id - the pair that connects to ByteDance Create Video Asset. Use a registered asset ID instead of an image here; they're mutually exclusive with the image inputs, and this is how you keep a specific person consistent.
• seed - controls re-running, not results (nondeterministic, per the tooltip).
• watermark - toggles an "AI generated" watermark on the output.

What comes out

A single VIDEO output - the animated bridge between your frames. Save it, preview it, or feed it onward.

Gotchas

• Asset IDs and image inputs don't mix. Put both a first_frame image and a first_frame_asset_id and the node rejects the run. Choose one source per end.
• Good frames make good video. The first/last frames are the strongest signal you have; mismatched framing or lighting between them and the middle gets weird. Generate the two frames with the same style and character.
• Watch the model's duration ceiling. The 2.5 model's sub-options go longer, but 2.0 tiers have their own limits - and longer + higher res = pricier. A failed validation still shouldn't bill you, but a successful long render will.
• Don't expect the seed to reproduce. Same story as every partner node: seed just decides whether the node re-runs.

If character consistency across shots is the goal, the full pattern is: Create Video Asset once → this node with *_asset_id → consistent faces, shot after shot.