Seedance 2.0 Mini: Reference to Video
Generate cinematic video from two reference images using Seedance 2.0 Mini, producing a single video output with AI-driven camera motion and consistent character generation. Ideal for storytelling, branded content, and short film creation where shot composition and visual continuity are essential.
Try an example
About this workflow
Seedance 2.0 Mini: Reference to Video turns two still images into a single, cinematic video with AI-driven camera motion and strong character consistency. In ComfyUI, the workflow is compact and focused: two LoadImage nodes feed the ByteDance2ReferenceNode (the generator that runs the Seedance 2.0 Mini model), and the frames are encoded to a final clip with SaveVideo.
By supplying a pair of complementary references—commonly one close-up for identity and one wider image for pose, outfit, or scene—the model blends cues to maintain the same subject across time while moving the virtual camera for a polished, shot-like result.
Technically, ByteDance2ReferenceNode ingests both reference images and synthesizes a sequence of frames conditioned on those visuals. It handles motion, composition, and temporal coherence internally, so you don’t need a separate sampler or flow node. SaveVideo then assembles the generated frames to your chosen resolution and FPS. The minimal node set makes this ideal for rapid iteration: swap references, adjust motion/length settings in the generator, and quickly export clips for storytelling, branded content, or short-film beats where continuity and shot language matter.
Questions & answers
The workflow is designed for two references: one for identity and another for pose or style. If you only have one, you can feed it to both inputs, but you’ll likely get less dynamic motion or variety compared to using a complementary second image.
Use sharp, well-lit images with similar aspect ratios and framing. For identity, ensure the subject’s face and key features are large and unobstructed. Keep backgrounds simple on the identity image and let the second image carry pose, outfit, or scene cues.
Set them in the ByteDance2ReferenceNode (for how many frames to generate and at what size) and finalize FPS/encoding in SaveVideo. For faster iteration, start with shorter durations and lower resolutions, then upscale or extend once you’re happy with motion and identity.
If ByteDance2ReferenceNode exposes a seed parameter, setting it will help reproduce results. Otherwise, expect small variations between runs due to stochastic generation. Keep references and settings consistent to minimize drift.