Segmind API vs Sume: PixelFlow workflows or saved Formats

What is the difference between Segmind and Sume?

Segmind is a model host plus a visual workflow builder. Sume is a media platform where an agent builds the output and you save the recipe. If you already think in node graphs, Segmind is the closer fit. If you would rather describe the result and have the recipe saved, Sume's Formats are the equivalent.

What does Segmind document?

The Segmind docs describe 200+ models and an AI Gateway across image, video, audio and language generation. Authentication is an API key sent in the x-api-key header. The API supports synchronous and asynchronous inference, with a section on async inference and webhooks. Cost is expressed in billing units, though the page does not list rates. The standout feature is PixelFlow, which converts a visual workflow into an API endpoint, with ready-made templates you can fork.

What are Sume Formats?

A Format is a thread's recipe from the Agents tab, addressable by handle and slug. You call it with POST /v1/formats/{handle}/{slug}/runs and an Idempotency-Key; it returns 202 with a run receipt, and you poll status_url until the run leaves queued or processing, then read result_url. You can also queue up to 100 runs with POST /v1/formats/{handle}/{slug}/bulk-runs and a concurrency window of 1 to 16.

Sume also accepts x-api-key: $SUME_API_KEY as well as a Bearer header, so the auth header from a Segmind client keeps working.

What does Segmind do better, and what does Sume not do?

Segmind gives you deterministic control. A graph runs the same nodes in the same order, and you can see and edit each step. A Format run is an agent turn, so the path can vary between runs, even though structured output rules constrain the result. Sume also has no public endpoint to author a Format through a visual graph; you author in the Agents dashboard or edit the package through the Contents API.

On the other side, Sume handles the planning: it picks the steps and models itself, and every run is bounded by a per-run generation_spend_cap_usd.

What happens when a run fails or must be retried?

With a graph you retry the endpoint. With a Format run you have more handles: the run has a receipt, a phase timeline at GET /v1/format-runs/{run_id}/events, an idempotent cancel, and a webhook redeliver endpoint that re-posts the terminal format.run.terminal receipt to the run's webhook URL. Terminal results return 409 run_not_completed while the run is still going, so poll status first.

The cost model also differs. Segmind lists billing units but I did not find rates on the page I read, so test with a real workflow. Sume bills metered generation (video, image, avatar, voice, timeline work) to the key that made the call, and curated first-party Formats are billed to whichever key calls them. The balance and usage endpoints are the billing record.

How should you choose?

Choose Segmind if you need a fixed pipeline you can inspect node by node and 200+ hosted models behind one key. Choose Sume if the inputs change per run (a new product, a new script) and you want the recipe to adapt. Start with the Format call guide and a single run before you queue a bulk list.

curl -X POST https://api.sume.com/v1/formats/acme/product-promo/runs \
  -H "x-api-key: $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: promo-0042" \
  -d '{"instruction":"Promo for the matte black bottle, 15 seconds vertical"}'