Motion Transfer guide

Choose a Motion Transfer trend or bring your own clip

If you already know the Bulin 47 performance, you know the job. One photo supplies a person. A motion reference supplies the movement. Motion Transfer is built around that pair: open a prepared trend, or upload a clip of your own, then generate with the model selected for that job.

The photo is the appearance. The video is the gesture, the timing, and the camera. A sharp still does not decide who in the scene should change. Match the photo to the role the page names, then read the model, the length, and the credit estimate before you spend anything.

Start with the motion

Use a prepared trend when you want a performance that is already loaded. The trends directory lets you preview a routine, see how many people change, and open that tool. A fixed-trend page loads the reference clip and a prepared prompt. You add photos for the labeled roles. Browsing, searching, and playing those previews does not submit a generation. The credit estimate appears on the tool page, and it can differ by clip and by output settings.

Several trends take one photo. The directory names Money Spread, Ozil Goal, and Bulin 47 among them. Other trends ask for two or more photos because each photo is a different role. Open the upload panel and follow its labels. Do not repeat one portrait across slots that are meant to be different people.

Bring your own clip when the catalogued performance is the wrong dance, the wrong room, or a shot you recorded. On the homepage workspace, choose or upload one continuous video, then add the character photo. Trim other scenes away first. Keep a beginning and an end you can still follow, including the hands or feet that matter. A cut in the middle, or a second person crossing the action, gives the model two stories at once.

Play the motion before you commit to it. Gallery previews on the trends page start muted; use the sound control when audio is available. Fixed-trend videos in that collection include music or other sound, and they do not ask you to upload a soundtrack. A clip you upload is different. Read the audio control in Settings. A custom job does not automatically keep a trend’s soundtrack.

Match the photo to the role that changes

Include the body parts the motion actually uses. A full-body dance needs the head, hands, hips, and feet, with space around the limbs. A portrait gesture can sit closer, but a tight headshot gives the model little to use once an arm leaves the frame. Use one sharp, unobstructed person. Skip a collage, a group sheet, and a second face leaning into the frame.

Bulin 47 AI is a concrete one-photo case. Photo 1 replaces the main performer: sunglasses, an open light-pink shirt, standing behind the person who holds the portable speaker. The speaker holder, the speaker, and the rest of the crowd are meant to stay. There is no second slot for them. Use Replace if the file is the wrong person. Uploading the photo does not start a paid request.

Choose a sharp picture of the face, hair, clothes, and upper body you want to see. The pink shirt and sunglasses identify the source performer. They should not come back unless they are in your photo. You do not need to hold a speaker or copy the gesture. A picture of the person in front would give that appearance to the lead. A group shot can make it unclear which person the model should follow.

The Bulin 47 guide uses the same role split. It does not currently host an origin trim you can download. If you bring your own footage, use one continuous vertical take you are allowed to transform: the speaker holder readable in front, the performer behind them, and the overlaps inside that same shot.

Read the model, the length, and the credits

On the Bulin 47 page the model is already selected: Seedance 2.5 Edit. The loaded performance is about 23 seconds in a vertical frame, and the result follows that performance.

Original

Uses the full reference, about 23 seconds.

Economy and Standard

Use a prepared excerpt of up to about 15 seconds, with matching reference audio.

The page includes the music. You do not upload a song. Sign in when asked, then read the credit cost on Generate for the resolution you selected. Plan examples elsewhere can use other models, so they are not the price of this clip. Length names and the credit number can change. Read the controls on the page you are about to submit.

A homepage job is the other path: you choose the model. Checked on 6 November 2026, the homepage FAQ named Kling 2.6 Motion Control, Kling 3.0 Motion Control, Wan 2.2 Animate Move, Wan 2.2 Animate Replace, Seedance 2.0, and Seedance 2.5. It described Kling and Wan as one-photo motion transfer, and Seedance as multiple images plus a video. The same day’s studio control also showed Seedance 2.0 Mini. If the menu and the FAQ name different models, follow the control for the job you are about to submit.

Those help texts also gave ranges to compare with your clip. Kling 2.6 and Kling 3.0 accept 3–10 seconds with Match photo, or 3–30 seconds with Match video. Match photo and Match video choose whether character orientation follows the photo or the video. They do not make every background or camera move match. Wan accepts 3–30 seconds. A Seedance reference video can be 2–15 seconds on 2.0, or 2–30 seconds on 2.5. Kling 2.6 and Kling 3.0 take one JPEG or PNG up to 10 MB. Wan 2.2 also accepts WebP. Check the formats shown for the model you selected. The estimate follows the actual clip and the resolution. A shorter excerpt is a practical first test. Do not copy a number out of this article and treat it as the charge.

Change the prepared prompt only for a specific edit

You can generate a fixed trend without writing a prompt. The Bulin 47 page already includes the instructions. In Settings, Custom prompt uses that built-in text by default. Turn it on to view, edit, or replace the text. If you clear the box, enter a prompt or turn Custom prompt off to return to the built-in version. If the switch is not there, that model is not offering a prompt. On the homepage, Wan Move and Wan Replace have no prompt field. Kling 2.6 and Kling 3.0 offer an optional scene note. Keep any note consistent with the photo and the motion. Text does not invent a gesture the clip does not show.

If you edit the Bulin 47 prompt, keep Photo 1 assigned to the performer behind the speaker holder. Do not ask for the foreground person to change. A different reference video belongs on the homepage workspace: upload the clip there and choose a model that accepts it.

For that homepage path, start with this example and adjust it to your clip. Video 1 is the motion clip you supply. Image 1 is the one character photo.

Use Image 1 as the appearance reference for the performer behind the speaker in Video 1. Keep that character recognizable as the camera moves and other people pass in front. Leave the person holding the speaker, the crowd, the setting, and their movements unchanged. Follow the timing of the reference clip.

If that render comes back silent, add only a soundtrack you are allowed to use. Restoring a prepared reference’s audio does not grant a right to post the music.

Check the person, the overlaps, the sound, and the rights

Open the finished clip in My creations, play it through, and download the version you want to keep. Page previews can start muted, so turn the sound on for the review.

Watch the character first. Compare the face, hair, and outfit with your photo in a clear frame, after a turn, and again near the end. On Bulin 47, the source pink shirt and sunglasses should not snap back unless they were in the photo. Hands, identity, and mouth movement can still drift. The edit is asked to follow the source mouth timing. Precise lip sync is not guaranteed, especially when the face is turned or partly hidden.

Then pause on the blocked views. For Bulin 47, check four moments:

1. The first clear view of the lead, before anyone crosses.
2. A frame where the speaker holder overlaps the lead.
3. The moment a phone or another person blocks the face.
4. The first frame after that block, when the lead returns.

The speaker holder should still be the foreground person, with their own face, clothes, and speaker. After the block, the lead should return as the same person. Crowded frames can still mix a face or a hand. If the wrong person changes, restore the prepared prompt and try one clearer photo. Do not add a second photo to a template that has only one role.

Listen to the whole clip. Bulin 47 includes its reference music, so confirm you can hear it in the file you download, not only in a muted preview. A homepage job follows the audio setting you chose. If the result is silent, check the source clip, the audio setting, and your downloaded file. A trend’s song is not automatically attached.

Using the reference, the template, or the music does not give you publishing rights to the footage, the performance, or the soundtrack. Upload a photo of yourself or of someone you are allowed to depict. Post the clip only where you have permission, and label it as AI-made where the destination asks for that.

Questions before you generate