The practical guide
How to make a Tiny Desk AI video
Start with the cast, not the prompt. Choose one clear portrait for a Solo lead, or assign up to five portraits to the fixed Band roles. A reference-based generator handles the performance structure; your job is to provide usable photos, choose the quality and review the result.
On this page
Choose the route that matches your goal
A finished desk-concert edit, a manual AI experiment and a photo-overlay template are different jobs. Check what a tool actually changes before spending credits. A template name alone does not tell you whether it animates a new person, replaces only a face or adds a static image.
| Route | You provide | Work to plan for |
|---|---|---|
| Reference-based generator | Portraits assigned to roles and a quality setting | Review the cast, submit a configured generation and inspect the finished video. |
| DIY video-model workflow | Rights-cleared reference material, portraits and instructions | Test identity consistency, edit timing, handle sound and assemble the output yourself. |
| Video-editor template | The photos or clips required by that specific template | Check its actual preview: overlays and transitions do not necessarily recast a performance. |
Give each role a face the model can read
Choose a front-facing or gently turned portrait with even light and a clearly visible face. A chest-up photograph is a useful starting point because it gives the model both facial and clothing information. Crop out other people before assigning the image to a role.
Use photos you have permission to upload and transform. Avoid tiny faces, sunglasses, strong beauty filters, heavy motion blur and hands covering the mouth. More pixels do not fix an obscured face; a simpler, clearer image is usually the better input.
- One person in each uploaded image.
- Keep the full face within the crop, including the chin and forehead.
- Use the same person consistently if you want to reuse a portrait across roles.
- Let the uploader finish its size and format checks before submitting.
Solo changes the lead. Band gives you individual roles.
On this site, Solo uses one lead-singer photo. It does not automatically assign that person to every member of the band. Choose Band to upload one to five portraits into the lead, saxophone, keyboard and two backing-vocal positions.
Band slots can remain empty, and the uploaded role indices stay fixed. Removing the saxophone portrait does not move a backing singer into its place. Switching to Solo retains your Band draft in the current page, but Solo sends only the lead photo.
Choose quality before you create
The configured output is a 19-second vertical performance. Quality determines the credit charge; adding Band portraits does not introduce a separate per-person charge. Review the cost displayed by the generator and both consent checkboxes before submitting.
Sign in and choose an available credit pack to purchase. You can inspect the form and upload local photo selections, but these are not a free AI render or a generated preview. A generation also needs a configured service, a signed-in account and enough credits.
| Quality | Credits per video | Format |
|---|---|---|
| 480p | 100 credits | 19 seconds · 9:16 |
| 720p | 300 credits | 19 seconds · 9:16 |
| 1080p | 700 credits | 19 seconds · 9:16 |
Follow the task through to a finished download
My creations keeps the task history so you can return after leaving the page. A submitted task may still be processing or completing audio and storage work. Download only when the finished video is available.
A task that needs delivery recovery can continue on the same job. This is different from a confirmed failed generation: a successful provider render waiting for finalization is not automatically refunded or submitted again. Check the task state before attempting another creation.
Watch the entire result for identity changes, hands, instruments, transitions and audio timing. AI results can vary even with good inputs. This site does not promise a separate silent export or a fixed completion time.
If you want to build the edit yourself
Begin with reference material you are allowed to use. Write down who appears in each shot, where the camera cuts and what movement needs to continue. Define a cast map before generating anything, rather than hoping the model will infer identities from a folder of images.
A model that accepts video references may edit a complete clip. Other tools require short segments that you assemble afterward. Neither approach guarantees consistent faces. Keep the same character descriptions, compare overlapping shots and test a small part before committing to a full run.
Treat sound as a separate deliverable. A visual generation does not automatically grant rights to a recording, reproduce the intended track or preserve its sync. Plan your permitted audio source and check the platform rules where you will share the video.
Avoid the mistakes that cost another attempt
| Problem | What to change |
|---|---|
| A group photograph in one slot | Crop to the intended person and confirm the role assignment. |
| Expecting Solo to cast every seat | Use Band and assign the roles you intend to change. |
| Expecting an uploaded image to be an AI preview | Local photo selection is an input check, not a generated result. |
| Resubmitting while delivery is recovering | Open the existing task and use its available recovery action. |
| Assuming every editor template performs a face replacement | Watch the actual template demonstration and inspect its input requirements. |
The step-by-step checklist
- 01
Pick Solo or Band
Use Solo for a single lead photo, or Band for one to five fixed role assignments.
- 02
Prepare and assign portraits
Upload clear individual portraits you have permission to use. Wait for the uploader checks and confirm each role.
- 03
Review quality and credits
Choose 480p, 720p or 1080p. Review the displayed credit charge and confirm both photo-permission and AI-result acknowledgements.
- 04
Sign in and submit when available
Generation requires a signed-in account, enough credits and a configured service. Submit once and follow the task status.
- 05
Review your completed video
Return to My creations, wait for completion and inspect the downloaded video for faces, motion and audio.
Questions, answered
- Do I need five photos?
- No. Solo needs one lead portrait. Band accepts one to five portraits assigned to fixed roles; unused Band slots may remain empty.
- Can I make a free generated preview here?
- No free generated preview is currently offered. You can view the reference sample and select photos locally, but neither is a new AI render of your cast.
- Does Solo put my face in every role?
- No. Solo submits only the lead photo. Use Band to assign the additional roles you want to change.
- How long does generation take?
- No fixed completion time is promised. Follow the task status in My creations; rendering, audio processing and delivery can take different amounts of time.