How do I animate my podcast cover art into a 5-second video intro?

To animate podcast cover art into a video intro, send the cover as a first_frame to POST /v1/videos with gemini-omni-flash-1.1, ask for something small to move, and add a sting from the Music Router. On Sume a 5-second 720p clip is $0.63, the sting is $0.125, and the render that puts them together is $0.10: $0.855 for an intro you make once and keep. Rates were read on 2026-10-07.

The first frame is the point. Your cover is a design with a title and a logo, and a first-frame image pins the opening of the clip to that picture. What the model adds is motion, not a new design.

Ask for a small move

Cover art is usually a flat graphic. Ask for a slow push-in, a light sweep across the title, a few particles, or the background drifting. A prompt that asks for a lot (the host turns around, the logo spins) gives the model room to change the title, and text is where video models fail first.

Turn generate_audio off, since the sting should come from the Music Router. The 3 to 10 second range of Omni makes 5 seconds a standard length: long enough to read the name, short enough not to be skipped.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: pod-intro-v1" \
  -d '{
    "model": "gemini-omni-flash-1.1",
    "prompt": "Slow push-in on the cover, soft light sweep across the title, no new text",
    "duration": 5,
    "resolution": "720p",
    "aspect_ratio": "16:9",
    "generate_audio": false,
    "frame_images": [{ "type": "image_url",
      "image_url": { "url": "https://media.sume.com/artifacts/artf_demo/cover.png" },
      "frame_type": "first_frame" }]
  }'

The sting

Music Router takes a prompt, and the prompt controls the length: ask for a 5-second track, and name the mood, the tempo and that it is instrumental. Something like 'a 5-second warm, rising sting at 100 BPM, ends on one clear note, instrumental' gives you something that lands on the title.

Each generation is a flat $0.125, so generate two or three and pick one. Check the returned length. If it is longer than the clip, set duration on the audio part in the render so it fits, or trim the sting in a Timeline audio split ($0.01).

Join and reuse

Render the clip with the sting as the audio. A render under a minute is one started minute at $0.10. Then keep the file. For each episode, the intro is the first slot in your episode render, and you add the intro's audio ahead of the episode audio with Timeline audio concat ($0.01) so the sound and picture line up. The intro itself is never made again unless you redesign the cover.

Check that the sting is not louder than the episode. Listen on earbuds and on a phone speaker before you ship it.

Check the result before you keep it

Look at the first frame and the last frame of the finished clip against the cover. The first should match exactly, because you pinned it. The last may have drifted: if the title letters changed, regenerate, or ask for less motion. You can also pin the end by passing a last_frame image, which holds the clip to a second picture, such as a version of the cover with a different background colour.

If the title keeps changing, take the title out of the animated clip altogether. Animate a cover with no text, and overlay the text still on top with Timeline compose in overlay mode ($0.02). Then the name of the show is always your own file, never the model's drawing of it.

Last, test the sting at the volume of your episodes. An intro that is louder than the show makes people turn the volume down for the next one.

Resolution choices for the clip

• 720p is plenty for an intro that sits in a corner of a podcast app or a video frame.
• 1080p matters if you will show the intro full screen on a TV.
• A retake of the clip alone is the clip price, plus a new render at $0.10.