Depth Motion Transfer · Techwear Dance in Front of a Rural Tractor

SeedanceWorkflowRealisticImage to Video15s
The dance reference is converted into a Depth video, keeping only the movement, rhythm, and body structure, then paired with a composited reference image of a new character in a rural setting and fed to Seedance for motion imitation: a dancer in a silver techwear jacket performs the whole routine on a fixed camera in front of a tractor and a village dog. The author notes that the original choreography and music copyrights still need to be handled separately.
PROMPT · Video Prompt
Create with AI
Generate a 15-second, vertical 9:16, fixed-camera, single continuous-shot high-definition realistic dance video. No cuts, no transitions, no push/pull, and no orbiting throughout. @video_1 is the Depth dance reference, used only to control the movement, rhythm, body center of gravity, turning direction, character position, and shot composition. Strictly follow the choreography order in the reference, restoring each arm swing, hip twist, knee raise, cross-step, jump, and pause as closely as possible, without adding new moves. @image_1 is the only character and scene reference. Keep the roughly 20-year-old adult East Asian female dancer's face, body, high ponytail, and outfit consistent: a silver short techwear jacket, a white sports bra, loose dark blue cargo pants, a metal waist chain, and white sneakers. Preserve the rural farmland scene from the image: a flat drying ground, an old red tractor, golden farmland, villagers, and a Chinese rural dog. The dancer is centered in the frame, dancing seriously, while the background people and the village dog only give slight, natural reactions — they must not dance along, steal focus, or block the character. The character is tall and slender with a narrow waist and long legs, and her movements are cool, sharp, relaxed, and powerful. Use a slightly low-angle, upward-tilted full-body medium-wide shot, but without exaggerated perspective. She should be fully in frame from head to soles, with her feet staying firmly planted on the ground. The high ponytail, jacket hem, pant legs, and waist chain sway naturally with her movements. Authentic afternoon sunlight, realistic skin, and camera texture. Avoid: face swapping, outfit changes, character scaling, foot sliding, body clipping through objects, limb merging, duplicated background people, villagers dancing along, the dog dancing, the tractor moving, scene cuts, subtitles, text, logos, and watermarks.
✍️ Editor’s Notes

This prompt's core idea is splitting 'motion' and 'appearance' into two independent channels: @video_1 handles only movement, rhythm, and composition, with an explicit 'don't add new moves' line to keep the model from improvising; @image_1 handles only character and scene appearance, with hairstyle and outfit details specified down to color and style. Splitting the two channels this cleanly makes them far less likely to interfere with each other than mixing motion and looks into one description. The fixed camera, single shot, no cuts or pans requirement exists to support Depth-based motion transfer — the moment the camera moves, it breaks the spatial correspondence with the reference video. The closing avoid-list specifically calls out 'duplicated background people,' 'villagers dancing along,' and 'the dog dancing' — all common failure points with this kind of composited reference image, clearly written from hard-won experience.

Create with AI
Workflow Prompt
PROMPT · Full Prompt
Create with AI
1. Prepare a dance reference video no longer than 15 seconds 2. Use Claude to build a Depth extraction tool that converts the original video into a depth-map video 3. Generate a new character and a new scene, then composite them into one complete reference image 4. Upload the Depth video and the reference image to Seedance and use motion imitation 5. Once generation is done, re-score it with suitable music and align it to the beat Depth mainly preserves the movement, rhythm, and body structure, keeping the original person, outfit, and background out of the generation pipeline, which also lowers the odds of the source material getting flagged in moderation. The copyright of the original choreography and music still needs separate attention. Note: in the example, the original "tractor dance" video was created by Douyin (China's TikTok) user @周同学.
✍️ Editor’s Notes

This workflow breaks into five steps, and the core logic is 'strip identity first, then composite': step one takes the reference video, step two immediately converts it to a Depth map, stripping out the original person and background early on so only the motion data remains — that way step three's new character and scene won't clash with the source footage, and step four merges both inputs and feeds them to Seedance for motion imitation. This order avoids the 'face and outfit swap failures' that commonly show up when you feed a real person's video straight into a prompt-based generator. It specifically flags that the music and choreography copyrights need separate handling, and credits the source video — for this kind of workflow prompt, the reusable value lies in the step order itself, not any particular line of wording.

Create with AI
Related Works
׋›
↑