Switching Fabric to H3 Max Lip Sync: 6 s, 12 s and 20 s clips compared

VEED Fabric 1.0 and MiniMax H3 Max Lip Sync share the same still-plus-audio body on Sume, so switching is mostly a URL change from /v1/veed/fabric-1.0 to /v1/minimax/h3-max/lip-sync. The catch is the audio window. H3 Max takes 5 to 14.8 seconds, so a 20-second clip cannot move over. For a 6-second clip H3 Max at 768p costs $0.60 against $1.125 for Fabric at 720p.

Three clip lengths

The prices below come from the catalog rows: Fabric $0.10 (480p) and $0.1875 (720p) per audio second, H3 Max $0.0625 (480p), $0.10 (768p) and $0.20 (1080p). Both bill ceil of the audio length.

Switching also changes the ceiling on what you can send. Fabric allows up to 300 seconds of audio, so a 3-minute narration is possible in one Fabric job but needs at least 13 H3 Max jobs of up to 14.8 seconds each (180 / 14.8 = 12.2, rounded up). Splitting audio is a separate step; Sume lists a timeline audio split at $0.01 per job.

Resolution labels are not the same

Fabric's tiers are 480p and 720p; H3 Max's are 480p, 768p and 1080p. Comparing 720p to 768p is the closest pairing, and the table does that: at 12 seconds Fabric 720p is $2.25 and H3 Max 768p is $1.20, a difference of $1.05. Whether the picture quality differs enough to justify it is something to test on your own stills.

Test both routes once on the same still and audio, then compare the mouth shapes and the face detail at your delivery size. The price table tells you what you save; only your own stills tell you what you give up.

• H3 Max: audio 5 to 14.8 s, under 10 MB, Sume-hosted.
• Fabric: audio up to 300 s, under 10 MB, Sume-hosted.
• H3 Max silently clips audio over 14.8 s, so measure before you send.

A routing rule

Send audio of 5 to 14.8 seconds to H3 Max when it is cheaper at the resolution you need, and everything longer or shorter to Fabric. One check in your code decides: if 5 <= measured_seconds <= 14.8, use H3 Max, else use Fabric. Keep the same still and audio so the two routes remain comparable.

A small router function keeps the choice honest. It takes the measured seconds and the target resolution, returns the route and the cost, and refuses audio above 14.8 seconds for H3 Max. Log the route name with every job id so that a later cost review can compare the two over real traffic. The savings per clip are small in dollars, so the value comes at volume: 1,000 twelve-second clips save 1,000 x $1.05 = $1,050 between Fabric 720p and H3 Max 768p.