AI lofi video generator: a looping scene plus a lo-fi track

An AI lofi video generator makes the two parts of a lofi video: a short scene that loops, such as rain on a window or steam rising from a mug on a desk at night, and a calm instrumental lo-fi track. The scene then repeats under the music for as long as the video runs. With Sume, that is an image-to-video clip that starts and ends on the same picture, a Music Router track, and one Timeline 1.0 render of up to 30 minutes.

Facts come from Sume's Video generation, Music Router, and Timeline 1.0 docs and the Timeline schema in the Sume API reference, read on 2026-09-28. In the Agents tab you can describe the video instead: the agent picks the models and asks before it spends.

How do I make the looping scene?

• Make or pick one still of the scene in the video's shape, such as 16:9, and host it at a public HTTPS URL.
• Send it to POST /v1/videos twice in frame_images, as the "first_frame" and the "last_frame", so the clip ends where it starts. Every image-to-video model takes a last frame except grok-imagine-video-1.5.
• Prompt small motion that can repeat, with a still camera: rain running down the glass, a flickering candle, a cat's tail twitching.
• Check the two ends for a jump before you loop it. Seamless loop AI video shows the request and the check.

How do I make lo-fi music with AI?

Send a prompt to POST /v1/music-router/generate. The Music Router docs use a lo-fi brief as their own example: “Warm lo-fi hip hop, 84 BPM, C minor. Dusty Rhodes chords, brushed boom-bap drums, a muted trumpet answer at 0:10. A 30-second track. Instrumental, no vocals.”

Steer the length in the prompt, for example “a 2-minute track”: duration and duration_seconds are rejected, so the length is asked for, not set. When the job finishes, read the audio file from result.artifacts[] where type is audio. AI music prompt examples has more briefs.

How do I loop the scene and the music into one long video?

One Timeline 1.0 render. The track goes in soundtrack, the clip goes in one video[] slot as long as the whole video, and the fields in the table do the looping. How to increase video length by looping covers the clip side in more detail.

The clip's own sound is not used: in current code a render takes sound only from its spine and soundtrack. For several tracks in a row, list them in audio.parts[] instead of soundtrack (up to 20, joined without gaps, and no mode); the parts must add up to at least duration_seconds.

curl -X POST https://api.sume.com/v1/timeline-1.0/render \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: lofi-30-min-001" \
  -d '{
    "audio": { "mode": "silence", "duration_seconds": 1800 },
    "soundtrack": {
      "url": "https://media.sume.com/artifacts/artf_demo/lofi-track.mp3",
      "gain_db": 0,
      "loop": true,
      "fade_out_seconds": 5
    },
    "output": { "width": 1920, "height": 1080 },
    "render": { "pad_mode": "loop" },
    "video": [
      { "source_url": "https://media.sume.com/artifacts/artf_demo/rain-loop.mp4", "start": 0, "duration": 1800 }
    ]
  }'

What does it cost, and what are the limits?

• The track is $0.125 per audio generation, the fixed Music price on every Music Router model.
• The render reserves $0.10 per output minute, rounded up to whole minutes, so a 30-minute video reserves $3.00; it never charges more than its reservation.
• The clip is billed by model, at the provider's list price × 1.25. Each rate here is plus a 5.5% agent fee by default; see API pricing.
• One render runs 1–1,800 seconds, so 30 minutes at most; for a longer video, render parts and join them in your own editor.
• The track, the clip, and every other URL in the render must be this workspace's media.sume.com files, such as the outputs of your earlier Sume jobs.
• Each pass replays the clip from its start, so any jump between its last and first frames shows at every loop.