Talking Photo From 3 Minutes of Audio: Fabric Cost and File Size

A 180-second recording on a still photo costs $33.75 on Sume's Fabric route at 720p, or $18.00 at 480p. The audio must also fit under the 10 MB limit, which is where a long WAV breaks. This post does the math.

Facts are from Sume's models overview and API pricing, read on 2026-10-05. Fabric is veed/fabric-1.0, called with POST /v1/veed/fabric-1.0.

What does Fabric accept?

The body takes audio_url, duration_seconds and exactly one visual source: image_url or an avatar handle, never both. Duration can be 1 to 300 seconds. Billing is per whole second of audio, at $0.1875 per second at 720p and $0.10 per second at 480p.

The audio must be hosted by Sume (media.sume.com) and be no larger than 10 MB. The preflight checks only the host and the size, so a file that is too large is refused with audio_too_large and a foreign host with unsupported_audio_source; upload through media inputs first.

POST https://api.sume.com/v1/veed/fabric-1.0
{
  "image_url": "https://media.sume.com/still.png",
  "audio_url": "https://media.sume.com/talk-180s.mp3",
  "duration_seconds": 180,
  "resolution": "720p"
}

What does 180 seconds cost, and how big is the file?

The cost is straight multiplication. The sizes are calculated from audio bitrate, not measured on a particular file: a 44.1 kHz, mono, 16-bit WAV is about 88,200 bytes per second, which reaches 10 MB at about 113 seconds. MP3 at 128 kbps is 16,000 bytes per second, so 180 seconds is about 2.88 MB.

Sume's timeline audio docs advise WAV for lip-sync inputs and offer mp3 for smaller files, so for long takes consider splitting the audio into parts under about 100 seconds each and rendering them as separate jobs.

What is the worst case?

Maximum duration is 300 seconds, so the maximum reserve at 720p is $56.25, charged against your balance when the job is admitted. If a take is longer, split it. Remember this animates a still to your audio; it does not make a conversation.