One face per upload
Choose three separate, well-lit portraits and put them in performance order. A group photo gives the model several possible faces for one role.
Cast the bench, doorway and final close-up.Recreate Spin That Coupe Like What with three photos. A seated performer keeps time, the camera swings to the doorway and a third person leans in for the final line.
The test used Sam Altman on the bench, Dario Amodei by the door and a corgi for the final line. The first two people were recognizable, though the opening beat was subdued. The dog appeared at Dario's feet and the original human speaker kept the close-up. The third replacement failed. This was the first version. The revised instructions have yet to be tested.



Dario Amodei · TechCrunch / Kimberly White, Getty Images for TechCrunch · CC BY 2.0
The first person sits by the bench and taps the beat. A handheld pan reveals the second person already dancing beside the closed door.
The final close-up belongs to the third person. Assign photos by action so each appearance has the face you chose.
Watch the locker room clip from @urbanmusicukThe changing-room video attributed to @urbanmusicuk dates to 2024. A developer described a separate nine-second, three-photo remake in November 2026. This page follows the eleven-second cut of the earlier performance, including the handheld pan and on-screen caption. DEV Community · @guihaihu

The first performer sits on the bench, marks the beat with both hands and points. Other teammates are already at the edges of the shot.

At about four seconds the camera swings right. The second performer is already bouncing by the closed white door when the camera finds them.

From about 7.5 seconds the performer in the dark jacket moves toward the phone. By the end it is their face, close to the lens, that has to match the third photo.
Assign three individual portraits in order: bench drummer, dancer beside the door, then the speaker nearest the camera. We recommend a human portrait for the last role. In our corgi test, the dog appeared by the door while the original speaker delivered the final line.
The clip runs for 11 seconds. Choose 480P or 720P to see the credit cost before generating.
Check the face at the doorway and again on the last line. The person nearest the camera should match the third photo.
Choose three separate, well-lit portraits and put them in performance order. A group photo gives the model several possible faces for one role.
Use a front-facing portrait with the mouth and chin visible. Check this person's face in the last second, when the camera is close enough to reveal a mismatch.
Choose a resolution to see the credit total on the button. Uploading photos does not start generation.
Bench drummer first, person by the door second, final speaker third. Assign the roles by action even when two people share the frame.
The selected footage reveals the second performer beside a closed door. The camera's move from the bench creates the entrance.
The corgi test did not succeed. The dog appeared as an extra by the doorway and the original speaker approached the phone. A human face is the better-supported input for this role.
The earlier changing-room clip circulated in 2024. The November developer write-up describes recreating it with uploaded photos. Its nine-second implementation and this eleven-second cut have different running times.
The original has text across the center and an uploader mark lower down. Both were still visible in the test output, so plan your own captions around them.