Guide · Visuals
Making your own Spotify Canvas: format, loop and what really works
A Canvas is the short, silent video loop that plays behind a song on Spotify. It costs nothing, takes an hour to make and is, in our experience, the cheapest way to keep listeners on a song longer – and one of the few places where AI imagery needs no justification.
The requirements
- Portrait 9:16, resolution 720 × 1280 (larger is downscaled, smaller is rejected)
- Length 3 to 8 seconds
- MP4 with H.264, no audio track, under 8 MB
- No text, logos or social handles that amount to a call to action; no flashing (Spotify checks)
- Upload via Spotify for Artists → song → Canvas; it can take up to a day to appear
As of October 2026 – the current limits are in Spotify for Artists.
From still image to loop: the path that works for us
- Choose a motif, don’t copy the cover. The listener sees the cover anyway. A Canvas tells a detail from the song: the cup by the window, the rain on the glass, the empty chair. One motif per song – whoever uses the same figure for ten tracks is boring.
- Generate a start image. With an image generator of your choice, in portrait format, without text. Several variants, pick one. Which model you use depends on hardware and budget – for a Canvas practically any will do.
- Set the image in motion. Two routes: an image-to-video model (local or as a service) or, if that is not available, a slow camera move over the still (zoom from 100 to 105 percent in eight seconds). The camera move looks calmer than you’d think and costs no compute time.
- Close the loop. This is the step that decides quality. Three methods:
- Crossfade: the last 0.6 seconds are blended into the first 0.6 seconds. Works for motifs without directional motion (steam, clouds, flicker).
- Ping-pong: the clip runs forward and then backward. Works for camera moves, fails for anything with a direction (rain does not fall upwards).
- Time stretch: stretch a 5-second clip to 7.5 seconds and have intermediate frames computed; the motion becomes calmer and the crossfade less noticeable.
- Export and check. Scale to 720 × 1280, remove audio, H.264 at moderate quality. Then watch the loop three times in a row in a player. If you can see the jump, everyone can.
The mistakes we have all made
- Too much motion. A Canvas runs on endless repeat next to the lyrics. Anything that looks like a trailer gets annoying after the third pass. The restrained version almost always wins.
- Too little motion. A pure still is accepted by Spotify but looks like a mistake. At least one element has to live.
- Figures doing something. Image-to-video models love to invent hands reaching into frame, or a second car that is suddenly there. Motifs without acting people are more reliable.
- Curtains, fabric, wind in the prompt. That almost always becomes a large bright cloth drifting through the frame. Better “static locked-off shot” and movement through light and particles.
- Vehicles. Once “a taxi drives by”, twice “one car becomes three”. Solve it as a still with a camera move.
One Canvas per song, one afternoon per album
With a practised routine a Canvas including variants takes 10 to 20 minutes of working time – the rendering runs alongside. For a ten-track album that is an afternoon. The same clips can then be reused as backgrounds for release clips, lyric videos and thumbnails – a separate guide covers that.
All guides are deliberately kept general: which models and tools run on your machine depends on your hardware. Questions or corrections? Write to us.