The studio builds the prompt from your photo slots
EasyVideoAI creates the Tiny Desk editing prompt automatically. Your inputs are the labeled portraits and the supported output settings. There is no custom prompt box in this studio. A role that has no photo is explicitly preserved in the generation instructions.
The current workflow uses ReAPI with Seedance 2.5. A model name alone does not define the edit: the source video, reference ordering and role instructions must agree. Copying the text into another tool requires checking that tool’s source-video and multiple-reference support.
Use clothing to find the target person
The lead singer’s navy shirt and the keyboard player’s green sweater describe people in the source. Those details help distinguish targets when faces turn or the camera changes. They are locators for the original cast.
The replacement instruction then gives that role the person from its assigned photo: face, hair, visible body appearance, outfit and accessories. Preserving only a few facial features would leave the old person’s appearance in the result. Performance props such as the headset, microphone and instruments remain attached to the scene.
Example: replace the lead and saxophonist
This readable example describes a two-photo casting plan. It is not API syntax. Keep the fixed role names when some slots are empty; a compact upload order must not change which performer each image belongs to.
Edit the supplied Tiny Desk source video.
Replace the lead singer (black hair, navy collared shirt, headset) with the complete person in the lead-singer reference photo: face, hairstyle, visible body appearance, outfit and accessories.
Replace the saxophonist (wavy brown hair, lavender top, gold saxophone) with the complete person in the saxophonist reference photo.
Keep the keyboard player and both backing singers unchanged. Do not reuse either selected identity for an unassigned role.
Track each assigned person across every cut. Preserve the source performance, camera sequence, timing, room, microphone, headset and instruments.
Audio is restored separately from the source after visual generation.Say what the performance must retain
A reference portrait provides appearance. The source clip provides role position, gestures, movement and shot order. Asking the same edit to invent a new song, change the room and preserve the original sequence creates competing instructions. This template keeps those scene choices fixed.
Appearance replacement still has practical limits. Fine accessories, hands, clothing edges and identity across occlusions can drift. The prompt expresses the target behavior; it does not guarantee every generated frame.
Fix mapping problems before adding more adjectives
If the wrong performer changes, compare the selected slot and review dialog first. If one face appears in multiple roles, inspect whether any photo was assigned twice. If an outfit is unclear, use a reference that shows it clearly.
The soundtrack is restored after the visual generation. Repeating “keep the audio” in a visual prompt cannot replace that media step, and the portrait does not create a new singing voice.