Need to test a camera move, action, or story beat without waiting on a full-quality pass? Seedance 2.0 Fast brings the Seedance 2.0 generation modes into CawCut with a faster-turnaround profile, making it useful for exploring and revising ideas. It is available directly in CawCut’s [Video Generation] node for text-led scenes and reference-guided clips.
1. Seedance 2.0 Strengths
| Strength | Application |
|---|---|
| Quicker turnaround | Explore prompt variations, camera directions, and motion ideas while shaping a scene. |
| Seedance 2.0 family motion and direction | Create action-led clips with clearly described movement, subject behavior, and camera intent. |
| Multiple reference modes | Build from text, images, frames, or combined omni references when the right input mode is available. |
| Iteration-friendly generation | Compare alternate scene beats or visual directions before settling on a final approach. |
Choose Fast when turnaround is a priority and you want to iterate on a video idea, refine the prompt, or preview the intended motion. When the final deliverable calls for the strongest detail or native 4K output, compare with Seedance 2.0 and check the controls available for that model in CawCut.
2. Choose a Mode and Inputs
Choose the mode according to how you want to guide the scene:
| Mode | Use it for | Inputs |
|---|---|---|
| [Text] | Creating a scene from a written brief. | A prompt describing the subject, action, setting, camera, and sound. |
| [Image References] | Animating a still or guiding the subject’s appearance and composition. | Image or character references and a movement direction. |
| [Frames] | Defining the opening and optionally the ending composition. | A first-frame image, optional last-frame image, and a transition description. |
| [Omni References] | Combining reference types in one generation. | Supported image, character, video, or audio references plus a prompt assigning their roles. |
All three Seedance 2.0 variants are listed for these four modes in CawCut. Available inputs and output settings can vary by mode. See Video Generation for the mode details.
3. Prepare References and Prompts
Make each reference’s role explicit:
- Image or character reference: Identify appearance details to preserve, such as clothing, silhouette, color, or product shape.
- Video reference: Say whether it guides movement, camera behavior, framing, pace, or scene structure.
- Audio reference: State whether it guides dialogue, sound effects, music, rhythm, or atmosphere.
- Frame reference: Explain the intended start or end composition and how the scene should move between them.
For a clear prompt:
- Name the main subject and action first.
- Put events in the order they should happen.
- Give the camera one coherent direction for each beat.
- Specify important interactions and what changes as a result.
- Add style, lighting, and sound details only when they support the intended result.
- State which details must stay consistent.
For a fair comparison between model variants, keep the prompt, reference inputs, aspect ratio, and other available settings consistent.
4. CawCut Workflow and Settings
- Open a workflow and add a [Video Generation] node.
- Choose [Text], [Image References], [Frames], or [Omni References].
- Select the Seedance variant that fits the current stage of the work.
- Add references supported by the selected mode and write a prompt that explains the action and each reference’s role.
- Set the available duration, aspect ratio, resolution, and audio options. For Seedance 2.0, choose [4K] when native 4K output is required.
- Review the credit estimate, generate the clip, and check the full result for motion, continuity, sound, and timing.
A useful progression is to test a rough direction in Mini, iterate in Fast, and use Seedance 2.0 for a final pass when detail or native 4K is needed. For workflow connections, see Understanding Nodes.
5. Limitations and Refinement
Fast is positioned for quicker turnaround, so it is a practical choice for exploration and revision. For a final clip where maximum detail or native 4K is central, compare its output with Seedance 2.0. As with other generative video, complex motion, small details, exact timing, and audio alignment may need refinement.
| If you notice | Try this |
|---|---|
| The action is difficult to follow | Break it into a few ordered beats and identify which subject moves. |
| The camera direction is inconsistent | Reduce the prompt to one main camera move and one main action. |
| A subject or object changes appearance | Use a clearer reference and list the visual features that should remain consistent. |
| Audio or timing does not match the action | Describe the timing and sound cues separately, then refine one cue at a time. |
| A draft looks promising but lacks final polish | Keep the prompt and references, then try a higher-detail pass with the model suited to your delivery goal. |
Review a full clip before using it as final output. For current model options, see Supported Models; for node controls, see Video Generation.