VEED Fabric 47-second talking still: $4.70 at 480p, $8.81 at 720p

A 47-second talking still on VEED Fabric 1.0 costs $4.70 at 480p and $8.8125 (about $8.81) at 720p on Sume as of 2026-10-09. The catalog lists $0.10 per audio second at 480p and $0.1875 at 720p, and the route bills the ceil of the audio second, so a 47.0-second track is exactly 47 billable seconds.

Fabric is the long-form option in Sume's talking-still family. It accepts audio up to 300 seconds, where MiniMax H3 Max Lip Sync stops at 14.8.

The arithmetic

Take the audio length, round up to a whole second, and multiply. The default resolution is 720p. Pick 480p when the clip will be seen small, for example in a picture-in-picture slot.

How this relates to the vendor list price

The fal.ai model page for VEED Fabric 1.0 lists $0.08 per second at 480p and $0.15 per second at 720p (read 2026-10-09). Sume's catalog prices are those figures times 1.25: 0.08 x 1.25 = $0.10 and 0.15 x 1.25 = $0.1875. For this 47-second track the vendor-page rates would give $3.76 and $7.05, so the Sume premium is $0.94 at 480p and $1.76 at 720p. The vendor-page rates are what a direct customer would see on that page, so treat them as a reference point rather than a like-for-like quote.

Input rules to check before you submit

The request needs audio_url and duration_seconds plus exactly one visual source: image_url, or a ready avatar by id or handle. The audio must live on the Sume media host and be under 10 MB, and non-Sume hosts are rejected. Check the file size of your audio before you upload it. duration_seconds runs from 1 to 300.

A speed_tier of fast uses the Fabric fast queue. Prices in the catalog are listed per resolution, so check the live catalog before assuming a tier changes the bill.

• Audio: Sume-hosted, under 10 MB, up to 300 seconds.
• Resolution: 480p or 720p.
• One visual source only: image or avatar, never both.

Planning a batch of these

A single 47-second clip at 720p is $8.8125. Twelve of them, for example one per chapter of a course intro, are 12 x 8.8125 = $105.75, or $56.40 at 480p. Because each job is billed on its own audio length, the cheapest saving is trimming silence at the head and tail of the audio before you measure duration_seconds: every second you cut is $0.1875 at 720p.

Jobs run under your workspace's generation concurrency, and extra jobs wait as queued until a slot opens, as the generation admission docs describe. Submit with an Idempotency-Key so a retry after a timeout does not reserve the balance twice, then poll the status URL from the first response.