New person, same performance

Genjutsu Motion Transfer: Swap the Person, Keep the Moves

With Genjutsu 1.0, the reference video supplies the choreography, camera path and timing. Your photos supply the characters. Select a 4–30 second section and add at least one photo. Faces, hands and fast turns can drift. Genjutsu AI runs this workflow with Seedance 2.5.

  • Image 1 replaces the lead
  • Keeps motion, camera and timing
  • Vertical 9:16 supported
  • Source audio kept
任务类型
参考图0/10

在指令中输入 @ 选择图 1、图 2 或视频 1。 JPG、PNG 或 WebP · 最大 10 MB · 每边 300–6000 px。

Genjutsu 1.0 使用 Seedance 2.5,支持动作迁移和物体替换,可选 480p、720p 或 1080p。

视频时长
10 秒
4 秒17 秒
分辨率

输出:4:3 · 960×720

消耗 117 积分

你上传的内容不会被公开,仅提供给视频服务商用于审核和生成。 隐私政策

片段起始时间00:00 – 00:10
所选片段在时间轴上的画面
00:0000:17

动作迁移参考素材。选好时间段,添加你的照片,想改什么就改什么。

Motion transfer at a glance

ItemValue
Input

A built-in clip or your MP4/MOV (4 s–5 min, 50 MB) + 1–10 images of the new person

Edited length4–30 whole seconds on Genjutsu 1.0
Output

MP4, 480p–1080p, in the source shape (16:9, 4:3, 1:1, 3:4 or 9:16)

AudioThe selected section's original audio is restored
Identity mapping

Built-in clips: Image 1 → first lead, Image 2 → second lead. Uploads: name the person in your instruction

Cost

10 s at 720p: 117 credits (Genjutsu 1.0); failed jobs refund

What changes and what stays

Changes to match your photoFollows the source clip
FaceChoreography, gestures and body motion
HairCamera movement, framing and cuts
Body typeTiming and length
Clothing (from the photo)Background, other people and the audio

Put yourself into a TikTok dance

  1. Upload the dance clip, or pick a built-in clip such as Times Square stroll. Vertical 9:16 stays vertical.
  2. Choose Motion transfer and add a clear full-body photo of yourself as Image 1; add a 3/4 view for turns.
  3. Drag the window to the part you want (4 to 30 seconds).
  4. On your own clip, name the dancer: "The person in Image 1 performs the moves of the dancer in the white hoodie."
  5. Check the Genjutsu 1.0 credit quote and generate.
Dance clip · your upload
The person in Image 1 follows the steps, pauses and turns of the dancer in the white hoodie in Video 1. Keep the camera timing and the other dancers’ positions.

Photos that work

  • Full body when the dancer is shown full body; face clearly visible.
  • No sunglasses, masks or hair over the face.
  • Even light and a sharp image; 300–6000 px per side, up to 10 MB.
  • Similar build to the original performer drifts least.
  • Extra angles (front, 3/4, side) help on fast turns.

Put yourself into a movie-style scene

A short shot with one clearly visible actor is the best case: the camera move and action are followed while the person changes. Fast cuts, crowds, heavy occlusion, smoke or mirrors, and close-up dialogue are the hardest cases, and lip sync is not guaranteed. Edit only footage you have the rights to use, and never impersonate a real actor. The cinematic built-in clips (Gallery escape, Horseback archer, Diner walk) are made for this kind of test.

Edit or reference: how the model treats your clip

Genjutsu 1.0 uses Seedance 2.5 reference generation to follow the performance and camera while rebuilding the characters from your photos. Seedance 2.0, Seedance 2.0 Fast and Mini generate a new shot that follows your cut as a motion and timing reference, up to 15 seconds: they allow bigger transformations but can loosen exact timing.

Where motion transfer can drift

  • Hands crossing in front of the face can merge for a few frames.
  • Long hair and loose clothing may differ around fast spins and head turns.
  • A replacement with very different body proportions drifts more.
  • With several similar-looking people, name the one who changes.
  • The result is generated, not pixel-identical to the source.

Two people, two photos

On built-in duo clips, Image 1 replaces the first lead (left at the start) and Image 2 the second; with one image, the other lead stays as filmed. The duo swap page covers couples, friends and two-person uploads.

Built-in clips for motion transfer

These are reference clips. Pick one in the editor's clip browser.

街头艺人
街头艺人17 秒 · 4:3 · 1 位主角
时代广场漫步
时代广场漫步5.7 秒 · 9:16 · 1 位主角
银发造型
银发造型22 秒 · 9:16 · 1 位主角
画廊出逃
画廊出逃23.3 秒 · 16:9 · 2 位主角
骑射手
骑射手10 秒 · 16:9 · 1 位主角
街头潮服转身
街头潮服转身19.3 秒 · 9:16 · 1 位主角

Consent and rights

Use only videos and photos you have the rights to use. Anyone recognizable must be an adult who agreed to the edit, the content must stay non-sexual, and AI edits should be labelled as AI-made when shared. Images of people go through the provider's asset review before Seedance 2.5 generation; rejected material returns the credits.

最近更新: 2026-11-07

Motion transfer questions

Can I put myself into a TikTok dance video with one photo?

Yes. Upload the dance clip (vertical 9:16 is kept), choose Motion transfer, add a clear full-body photo of yourself as Image 1, pick the part you want (4 to 30 seconds) and generate. One photo works; adding a 3/4 view helps your face stay consistent through turns. Use only dance clips you have the rights to reuse.

Does it keep the exact dance moves?

It keeps the choreography, gestures, camera movement and timing close to the source, with Genjutsu 1.0 using the clip as its motion reference. The result is generated, so it is not pixel-identical: hands, long hair and fast spins can differ for a few frames.

How many photos do I need for a consistent face?

One clear full-body photo with the face visible is the minimum. Two or three angles (front, 3/4, side) give the model more to work with on turns. Up to ten images are accepted.

Can AI put me into a movie scene?

Technically yes for short shots with a clearly visible actor: the camera move and action are followed while the person changes. Fast cuts, crowds, heavy occlusion and close-up dialogue are the hardest cases, and lip sync is not guaranteed. Only edit footage you have the rights to use, and do not impersonate real actors.

Can I replace both people in a duo video?

Yes. On built-in duo clips Image 1 replaces the first lead and Image 2 the second. On your own two-person videos, say who each image replaces. See the duo swap page for details.

What is the difference between motion transfer and object swap?

Motion transfer uses the clip as a performance reference for the people in your photos. Object swap (Product mode) keeps the person and replaces the object they hold or use. Pick one change per generation.

Can I turn my dog or cat into a person or a character?

You can try Motion transfer with an image of the new character, but animals move with very different proportions, so motion drifts more than with people. For a cartoon look that keeps your pet, use Style mode instead.

Which model should I use for motion transfer?

Genjutsu 1.0 is the default for motion transfer and supports up to 30 seconds (117 credits for 10 seconds at 720p).

用任意 AI 助手核实

“Can AI replace the person in an existing video and keep the original moves?”

保留这条镜头,改换其中的世界。

选择片段,添加参考素材,几分钟后即可看到编辑结果。生成前会先显示报价。

打开 Genjutsu AI 编辑器