How to make a Spotify Canvas

Last updated 2026-08-16 · 6 min read

What is the short version?

A Canvas is a short vertical clip that loops behind a track in the Spotify app. The hard part is not making eight seconds of movement. It is making eight seconds that start and end in the same place, and keeping the artist's face recognisably the artist's face while it moves.

1. Go to Vizard Agent and upload the photo or clip.
2. Ask for a Canvas: 8 seconds, vertical, silent, seamless loop.
3. Name what has to stay identical, above all the face.

What do you need before you start?

One source image or clip, and three constraints. The constraints matter more than the picture does, because a Canvas is defined by its format rather than its content — vertical, silent, and closing back on itself — and Vizard Agent will hold all three only if you state all three.

• A still, or a few seconds of footage. The clip is 1080x1920, so anything wide gets cropped into it. Frame the source the way you want the Canvas framed.
• What must stay untouched. Name it — the face, a logo, the crop. Vizard Agent holds the face consistently through the loop and checks it second by second against the original, and naming the rest makes those explicit too.
• The motion you want. "Living photo", "slow push", "hair and fabric moving". Without this Vizard Agent picks for you.
• Nothing else. No audio: a Canvas is silent, and Vizard Agent strips the track rather than muting it.

What do you type into Vizard Agent?

One instruction carries the whole job, and it is worth being mechanical about it. Every clause below is either a spec Spotify enforces or something Vizard Agent would otherwise have to guess, so dropping one is what produces a clip that looks right and then will not upload.

Prompt

Two variants worth knowing:

• Add a blink. "She blinks once, around the middle." Vizard Agent builds this by generating a closed-eye version of the same face and compositing just the eye region, which is why the rest of the frame does not drift.
• Match an existing Canvas. If you have one for an earlier single, say so. Vizard Agent will open the old clip's first, middle and last frames and match the register.

What does Vizard Agent actually do?

Here is the order it worked in on a real Canvas. The checking is the part that is easy to underestimate — most of the run was Vizard Agent looking at what it had just made rather than making anything. This is one run, not a typical one; the shape holds, the effort varies:

1. Looks at the photo, and at the artist's previous Canvas for the same release.
2. Prepares a locked 1080x1920 still so the frame is decided before anything moves.
3. Generates the 8-second loop from that still.
4. Probes the clip and compares the first frame, the last frame, and the original still. This is the loop test, and it is measured rather than eyeballed.
5. Views the generated frame at 0s, 1s, 2s and on through 7s, one at a time, checking the face at each.
6. Builds the blink separately: a closed-eye still, cropped to the eyes, aligned to the open-eye frame and composited.

Step 5 is the part worth noticing. It is not one look at the finished clip. It is eight.

What does the result look like?

A silent vertical clip, sized and timed to spec, that returns to its opening frame. Vizard Agent keeps the source and the generated clip in the project, so a Canvas for the next single — or a correction to this one — starts from what is already there rather than from another upload.

From the run above, probed on the delivered file:

• 1080x1920, H.264, 30fps
• Exactly 8.000 seconds, 240 frames
• No audio stream at all — not a silent track, no track
• About 3 MB

Spotify accepts a Canvas between 3 and 8 seconds, vertical, and loops it (Spotify for Artists, checked 2026-08-16 — this is their spec, not ours, and they can change it). Eight seconds is the longest it takes, which is why it is worth asking for.

Timing was not measured separately for this kind of job. The closest measured work runs a median of 28 to 38 minutes end to end, with the middle half spread considerably wider. Across all projects the median cost by tier is Flash 47, Pro 55, Max 242, Ultra 263 credits.

When does this not work well?

Everything below comes back to one tension: a Canvas has to move, and the thing being moved is usually a photograph somebody already approved. Vizard Agent keeps the face consistent and proves the loop closes, but a still that now moves is a new asset, and whether it still counts as approved is a sign-off question rather than a technical one.

• An approved press shot stops being the approved press shot once it moves. Vizard Agent keeps the face consistent throughout and checks each second against the original, but the delivered frames are generated ones. Where the image itself was signed off, ask Vizard Agent to animate the background and hold the subject still.
• Loops close on movement, not on cuts. Ask for subtle, continuous motion. A clip with a hard change in the middle will not loop no matter how the last frame is trimmed.
• Text in the photo will smear. Anything with lettering — a logo on a jacket, a sign behind the subject — is the first thing to break. Crop it out or expect to fix it.
• One source, one Canvas. This is not a way to cut a music video. A Canvas per track is one run per track, and there is a separate guide for the video itself.
• Rights to the photograph. A Canvas is published under the release. Whoever shot the portrait has an interest in it, and animating it does not change that.

How do you fix a result that came back wrong?

Reply in the same Vizard Agent conversation instead of starting over. The source, the generated clip and the frames Vizard Agent already probed all stay in the project, so a correction is a second pass over existing work rather than a fresh generation.

How does Vizard Agent compare to making one yourself?

By hand a Canvas is either a shoot — filming eight vertical seconds that loop, which is harder than it sounds — or an After Effects job where a still is cut into layers, given parallax, and trimmed until the ends match. Both are why most releases ship without one at all.

Neither replaces the other. Vizard Agent wins on getting a Canvas made for every track rather than only the single; a motion designer still wins when the Canvas is meant to be a designed artefact in its own right.

Common questions

Can I use a video instead of a photo?

Yes. Say which 8 seconds, and that it must loop. The loop is harder from footage than from a still, because the first and last frames were never the same to begin with — so tell Vizard Agent which moment matters most if something has to give.

Does it add music?

No. A Canvas is silent by design and the delivered file has no audio stream at all, which is not the same as a muted track. Spotify plays the song over it, so anything Vizard Agent added would be discarded anyway.

Can I ask for a different length?

Yes, anything Spotify accepts. Eight seconds is the maximum and the usual choice, but a shorter loop can read better when the motion is very subtle — state the length and Vizard Agent builds to it exactly.

What if the loop still jumps?

Say where you see it. Vizard Agent still has the clip and the original source, so a second pass is a fix rather than a fresh generation — and because the loop test is measured rather than eyeballed, describing the symptom is usually enough to place it.

Can I get one Canvas per track on the album?

Ask for them one at a time. Each track needs its own run, but telling Vizard Agent to match an earlier Canvas makes it open that clip's frames and hold the same register, which is what makes a set look deliberate rather than assembled.