Recraft V4.1 Flash API: Sume lists V4, webp, text-only

Recraft V4.1 Flash is live in Recraft Studio and its API. Sume lists recraft/recraft-v4, not a Flash id: webp output, text-to-image only, no references.

4 min readSume
All posts

Recraft announced V4.1 Flash on September 23, 2026, available in Recraft Studio and through the Recraft API. Sume lists no Flash id. Its Recraft entry is recraft/recraft-v4, which takes a prompt and returns webp; reference images are rejected.

Vendor facts are from Recraft's post; Sume facts from the Image API docs and catalog code. Read 2026-09-30.

What does Recraft say about V4.1 Flash?

The post calls Flash a separate model, not a speed mode, and quotes about 1.3 seconds from prompt to image, a median generation time measured on a larger prompt set than its 100-prompt comparison, which Recraft ran on 21 September 2026 (Recraft models through the Recraft API). It notes that download time is excluded from that comparison and that default settings, including a prompt enhancer where the provider turns it on, were left as shipped. Those are Recraft's own figures and say nothing about Sume.

Its Refine button regenerates an image with the Pro model at 2048x2048 inside Studio.

Which Flash claims carry over to Sume?

None of them, because the Flash model is not in the catalog. The table separates what Recraft announced from what Sume lists.

Flash announcement against Sume's catalog, read 2026-09-30.
TopicRecraft postSume
Flash modelLive in Studio and the Recraft APINo Flash id; the Recraft entry is recraft/recraft-v4
Refine with V4.1 Pro at 2048x2048A Studio buttonNo Pro id listed

What should I send to Sume instead?

Use recraft/recraft-v4 with a prompt, and leave out input_references; the docs say a parameter the model does not list returns 400 unsupported_parameter. For vector work see Recraft V4.1 and SVG.

How long can a call take?

POST /v1/images blocks for up to 30 seconds. If generation is still running, Sume returns a 202 job envelope and you poll GET /v1/jobs/{id}/status. You can also send mode: "async", or mode: "webhook" with a webhook_url, to get the job envelope straight away and fetch the finished images from GET /v1/jobs/{id}/result. Latency figures from a vendor post do not tell you which of these paths a given call will take. Choose the model from the catalog, not from a vendor launch post; which image model lists what is there.

Sources

Related posts

More in Models

All Models posts

Written by Sume