Which Foundry models add C2PA and a watermark: GPT Image, FLUX.2, MAI
Microsoft's Foundry page lists the image, audio and voice models that get C2PA credentials and watermarks. What that does and does not tell you about Sume.

Microsoft states which of its hosted models mark their output. The Foundry provenance page, dated 2026-09-17 with an update on 2026-10-01 and read on 2026-10-05, says supported models add C2PA Content Credentials and invisible watermarks to what they generate. It is one of the few vendor pages that gives a model-by-model list.
If you use any of those models through another API, the useful question is whether the same marks survive. This post lists the models and then says what I can and cannot say about Sume.
The list on the page
| Type | Models named on the page |
|---|---|
| Image | GPT Image (gpt-image-1-mini, 1.5, 2), MAI-Image-2.5 variants |
| Image, Black Forest Labs | Flux-1.1-Pro, Flux.1-Kontext-pro, Flux.2-flex, Flux.2-Pro |
| Audio | GPT Audio, MAI-Voice-1, MAI-Voice-2, Azure AI Speech TTS |
| Audio formats | MP3 and WAV are supported |
What this does not tell you about Sume
Sume's image docs list models including openai/gpt-image-2.5 and black-forest-labs/flux.2-pro. The Foundry page names GPT Image up to version 2 and Flux.2-Pro, not 2.5, and it describes Microsoft's own endpoints. It does not say what happens when a model is reached through a different provider.
Sume's docs say nothing about carrying, adding or stripping C2PA data, so I cannot tell you a Sume FLUX.2 Pro image has credentials. The two facts to hold apart are Microsoft's claim for Foundry and the absence of any Sume claim.
What to do with that
Microsoft also says customers are responsible for their own transparency obligations, which is the same conclusion from the other side: the model vendor's mark does not replace your disclosure. Sume's docs describe a metadata field on video, image and music jobs that Sume stores on the job and does not send to the provider. I found nothing in the docs about Sume adding, keeping or removing C2PA credentials, watermarks or platform labels, so treat that as undocumented and check the delivered file.
- Do not repeat "it carries C2PA" in your own disclosure copy unless you inspected the delivered file.
- Download one Sume output and check for credentials with a verifier; use the vendor's own tool where one exists.
- Remember Microsoft's own caveat: provenance does not prove authorship or trustworthiness.
- Disclose on the platform where you publish, because the platform's toggle is what viewers see.
Sources
Related posts
More in Models
- MiniMax H3 API: the cheapest Sume job is 5 s at 480p, $0.3125
On Sume the minimax-h3 model takes 5 to 15 seconds at native 480p or 768p, with 2K and 4K upscales billed on request. Price table by length inside.
- MiniMax H3 Max: five free fal runs a day, then the Sume cost per clip
fal's page offers five free H3 Max generations a day of up to 15 seconds. Use them to test a prompt, then price the paid clip on Sume at 480p, 768p or 1080p.
- MiniMax H3 open weights: what is released and what stays hosted
MiniMax released H3-Base, FL2VA and Ref2VA checkpoints at 768p. Context-IR and the 2K regeneration module stay hosted. What that means for a Sume API user.
- MiniMax says H3 is 2K: what Sume lists, 768p to 1080p
MiniMax's own page puts H3 at 2K and 15 s. Sume lists native 768p for H3 and H3 Max, 1080p refinement on Max, and priced 2K/4K upscale SKUs on H3. Prices.
Written by Sume