Text, First Frame, and Start-End Prompts
Match the prompt to what the reference images already define
The more the images establish, the less visual setup the prompt needs to repeat. Identify what the source material has already fixed, then use text for motion, timing, and transition.
Source review: . Examples are original editorial drafts, not measured generation results.
- 01Text mode builds the complete shot
- 02First-frame prompts explain what moves next
- 03Start-end prompts design the transition
Divide the work between images and text
Text-to-video needs to define the subject, environment, composition, action, and style. A first frame already supplies identity, clothing, scene, and initial composition, so the prompt can focus on subsequent movement. Start and end frames set both endpoints, leaving the prompt to describe sequence, path, and timing.
Repeating every visible detail can introduce conflicts. Restate only what must remain stable or change.
Three prompts for the same creative idea
Each prompt below depicts a character turning away from a window and walking toward a door. The input material changes what the text must carry.
In a white apartment at dawn, a young man in a pale blue shirt and dark trousers stands with his back to the camera in front of a floor-to-ceiling window. He pauses, turns slowly from his left shoulder, walks toward the wooden door on the right, and stops in front of it. Wide shot, locked camera, soft natural light, realistic domestic cinema. Preserve clothing, room geometry, and walking direction.
Preserve the character, clothing, room, and camera from the first frame. He pauses at the window for one second, turns slowly from his left shoulder, walks in a straight line toward the wooden door on the right, and stops naturally. Keep the camera fixed, the motion steady, and the body proportions and room structure consistent.
Transition naturally from the first frame to the end frame. Hold the starting pose briefly, then turn from the left shoulder and walk along the center of the room toward the door on the right. Complete the main walk in the middle of the clip, slow down near the end, and align precisely with the final position, facing direction, and arm pose. Preserve identity, clothing, room geometry, light direction, and camera position. Do not reach the end frame early.
Check four things before using start and end frames
When the frames conflict, added text cannot reliably repair every mismatch. Simplify the endpoint or split the transition.
- Identity, clothing, and key objects match across both images.
- A clear and physically possible path connects the two poses.
- Lighting direction and camera position do not jump without explanation.
- The duration allows the subject to complete the requested movement.
Choose the input mode
- Use text when you want to explore an undefined shot.
- Use a first frame when appearance and composition already work.
- Use start and end frames when both endpoints matter and the transition can be described clearly.
References
Model features can change. Check the current official pages before relying on a version-specific capability.
- ByteDance Seed. Seedance 2.0 multimodal inputs. Accessed 5 Sept. 2026.
- ByteDance Seed. Seedance 1.0 text and image generation. Accessed 5 Sept. 2026.