Seedance2.0: First-Last-Frame to Video
Upload the first and last frames of your video. Generate a high-quality video sequence using the Seedance2.0 model.
试试一个示例
关于此工作流
Seedance 2.0 brings a boundary‑conditioned approach to video generation: supply the first and last frames, and the model synthesizes the in‑between motion with striking temporal stability. Instead of guessing a trajectory from a single image or relying on text prompts, the pipeline conditions on both endpoints to infer a coherent path for objects, lighting, and camera movement. In ComfyUI, this is encapsulated by the ByteDance2FirstLastFrameNode, which takes two images and produces a full sequence that faithfully lands on your final frame.
This first‑last‑frame to video (FLF2V) method reduces identity drift, flicker, and composition shifts that commonly plague single‑frame or text‑to‑video methods. Unlike basic frame interpolation—which can only warp what already exists—Seedance 2.0 synthesizes novel content between your keyframes while preserving structure and intent. The result is cleaner motion planning, stronger subject consistency, and more predictable outcomes for real productions.
常见问答
Interpolation warps existing pixels from the endpoints. Seedance 2.0 synthesizes new in-between frames using two-endpoint conditioning, enabling plausible motion and detail that aren’t present in either input while still honoring both boundaries.
Yes. The ByteDance2FirstLastFrameNode expects matching resolutions for stable results. Prepare both frames at the same size before loading them with LoadImage to avoid unintended rescaling or artifacts.
It can infer substantial motion, but extreme jumps may increase artifacts or ambiguity. For best results, keep the endpoints visually related and adjust duration (more frames) to give the model room to plan a coherent path.
Use a fixed seed and keep inputs constant. In ComfyUI, set or lock the seed at the graph or node level, then rerun to reproduce the same sequence deterministically.