Sume's experimental Fabric route costs $0.3024 a second vs $0.1875

Short answer

The Sume models docs describe POST /v1/avatar-1.0/fabric as a temporary, test-only route used to compare a different talking-clip backend against POST /v1/veed/fabric-1.0 on identical inputs. Its listed price is $0.3024 per second at 720p. The existing image-to-video route is $0.1875 per second at 720p. That is 61 percent more per second (Models docs).

Side by side

The two routes take the same request body, so a comparison script only changes the path. The differences are in the table.

Do not build on it

The docs say the name fabric is a placeholder that will change before general availability, and that Sume can change or remove the route when the comparison finishes. VEED Fabric 1.0 stays the supported path for talking clips, including Live Commerce and the MCP tools.

If you saw fabric in a catalog or an SDK type and wondered whether it is a cheaper or better tier: it is neither a tier nor a stable product. It is a more expensive A/B route.

What to do instead

• Use the supported veed/fabric-1.0 path for production talking clips.
• Price by clip length: 15 seconds at $0.1875 is $2.8125.
• Re-read the models docs before relying on any route marked experimental; routes in that section can disappear.

Why a more expensive route exists

The docs give the reason: Sume compares a different talking-clip backend against the supported one on identical inputs. A comparison route does not need to be cheaper; it needs to be measurable. Because it takes the same request body, a script can send the same still and audio to both paths and compare output and cost. The differences, such as the 15-second limit and the ignored speed_tier, are the visible results of that different backend.

The practical reading for a buyer is simple: nothing about the experimental route is a deal.

Worked cost of a comparison run

Ten 15-second clips at 720p on the supported route are 10 x $2.8125 = $28.125. On the experimental route they are 10 x $4.536 = $45.36, a gap of $17.235. If you are only evaluating quality, run a couple of short clips; a 5-second test is $0.9375 on image-to-video and $1.512 on fabric.

If you are using the supported route in production, nothing here requires action. The experimental path is for people who want to measure a different backend against the supported one, and the docs say plainly not to build on it. Treat the $0.3024 figure as the cost of the comparison, not as a premium tier with extra features, since the visible differences are limits rather than additions.

When you do compare, use identical inputs, record the captured amount for each job from the ledger, and keep clips short. A handful of five-second pairs tells you what you need to know for under ten dollars.

The route also teaches a general lesson about reading a price list. Two entries for what looks like the same task, a talking clip from a still and audio, can differ by more than half because the backend differs. Always read the model id and the route in the request before you multiply seconds by a rate, and keep the id in your own records next to the amount.

Because the rate is per second, the gap grows linearly with length: at 15 seconds it is $2.8125 against $4.536 for one clip, and the difference of $1.7235 is repeated for every clip in a batch.