Guide
The Video Model Behind Tiny Desk AI
How Seedance uses one photo and one reference clip, what the model keeps consistent and what every Tiny Desk AI video generator render includes.
One free preview per account: 5 seconds, silent. Full videos with sound start at $9.90.
Published October 28, 2026
Which model makes a Tiny Desk AI video?
Every paid render uses the main Seedance model from ByteDance through the API provider APIMart. The model receives the consented photo, the reference clip and the direction needed to place the person in the middle of the band as the lead singer.
The free preview uses a lighter, faster Seedance model. Paid 480p, 720p and 1080p videos use the better main Seedance model and run 15 seconds instead of 5.
Why the model matters
A recognizable reference scene puts several demands on the model:
- Keep the new lead connected to the source photo
- Coordinate the lead, band, backing singers and horns
- Follow the reference clip's order of events
- Maintain the crowded office setting across the shot
How a Tiny Desk AI render works
Every render follows the same path:
- The uploaded photo is resized and sent for an automated face and safety check
- Seedance receives the photo and one reference clip of the meme moment
- Your optional extra direction is screened and added to the render request
- The completed MP4 appears under My videos in your account
What every render includes
Every paid render includes:
- Seedance through APIMart
- A 15-second sequence that moves in on the new lead in the middle of the band, then shows the sax solo and band
- Sound from the meme moment
- Your choice of 480p, 720p or 1080p and 9:16 or 16:9
- Automatic credit return if the render fails
Help the model with a good photo
Use one sharp, front-facing photo in even light with the full face visible. Sunglasses, filters, side profiles and group photos make the job harder and can cause refusal. Check the image with the on-device photo checker and read the photo guide.
Safety, privacy and variation
Every photo receives an automated face and content-safety check before rendering. Photos of minors, celebrities, athletes, coaches and other public figures are refused. Source photos are used only for the requested render, are not used to train models and are removed from temporary storage after processing. Every render differs a little, and artifacts, odd motion or likeness differences can still occur.
Tiny Desk AI model FAQ
Is there a free preview?
Yes. Every signed-in account gets one free preview: a 5-second silent 480p close-up made with a lighter, faster Seedance model. The full video uses the better main Seedance model, runs 15 seconds instead of 5 and includes the band, sax solo and song. It costs 10 credits in 480p, 20 credits in 720p or 30 credits in 1080p.
Why does my video differ from the source photo?
AI video is probabilistic. Seedance follows the photo and reference, but each render varies and may include artifacts, odd motion or appearance changes.
Can I choose a different model?
No. Every render uses Seedance through APIMart.
Is my photo used to train the model?
No. The photo is used only for the requested render.
See Seedance at work
Upload one consented photo and make a 15-second Tiny Desk AI video.
More Tiny Desk AI guides
All guides and tutorials →- Tiny Desk AI video settings480p vs 720p vs 1080p, 9:16 vs 16:9 and credit costs.Read the guide
- Tiny Desk AI direction guideShort visual instructions that work with the reference clip.Read the guide
- How to make a Tiny Desk AI videoPhoto, safety check, settings, rendering and posting, step by step.Read the guide