ByteDance Seedance 2.0 First Last Frame (Channel Edition)

First-last-frame (FLF) generation is the closest thing video models have to a "trust me, it'll connect" mode: you give the model a starting frame, optionally an ending frame, and it invents the motion in between. This node is the Channel Edition wrapper for ByteDance's Seedance 2.0 FLF endpoint - the API-only video line from the same company that brought you Seedream for images. It's your ComfyUI door to "this exact character walks from this pose to that pose," without a single local inference step.

How it works

Cloud execution through BizyAir. Your first-frame image (and optional last frame) upload to api.bizyair.ai, the request hits the seedance-2-0-base/flf-to-video endpoint, and the finished clip is downloaded back into ComfyUI as a native VIDEO tensor. No GPU, no model files, no ComfyUI video-model setup ritual. It's a paid, metered service - your credits get eaten per clip, and your frames spend time on SiliconFlow's servers.

Inputs that matter

• first_frame_url - the required starting image. This is the whole brief's anchor: pick a frame with clear composition, because everything the model invents flows out of it.
• last_frame_url - the optional ending image (one image, up to 30 MB). Supply it when you need a specific destination; leave it out and the model improvises the ending.
• prompt - what should happen in the middle. "She stands up and walks toward the window" - motion descriptions beat static scene descriptions here.
• resolution - defaults to 2k. The tooltip spells out the two tiers: native output (480p, 720p, Native1080p) and super-resolution versions upscaled from a 720p base (1080p, 2K, 4K). Iterate native, upscale for finals.
• duration - seconds, default 5, range 4–15.
• ratio - aspect ratio, default 16:9.
• seed - integer, default 0; set it to reproduce a clip.
• generate_audio - default on: the clip comes back with sound. Flip off for silent drafts.
• return_last_frame - default off; turn on to also get the video's final frame as an image output, handy for chaining into the next clip.

Outputs: video (into a preview/save node), request_id, urls.

Install and get going

ComfyUI Manager → "BizyAirPlus", or:

cd /path/to/ComfyUI/custom_nodes
git clone https://github.com/siliconflow/BizyAirPlus.git
cd BizyAirPlus
python -m pip install -r requirements.txt

Restart, register at bizyair.ai, set the key in Settings > BizyAirPlus > API Key (or BIZYAIR_API_KEY), and flip the BizyAirPlus action-bar toggle ON - every node in the pack stays inert while that switch is off.

Gotchas

Setup failures are the usual trio: missing key, toggle off, deps in the wrong Python (python -m pip show bizyair-cloudberry). The FLF-specific trap is asking for too much between frames - a 15-second, several-beat action from two stills is a lot to ask and shows it. Keep short clips for complex motion, and if your ending frame keeps getting ignored, check that the first and last frames actually match in style and lighting; wildly mismatched anchors produce a confused middle.