FLUX.2 pose and layout guidance without ControlNet

There is no ControlNet field to set. BFL's guide says that instead of dedicated ControlNet inputs, FLUX.2 uses its multi-reference editing system: you supply a reference image that shows the pose or layout and describe in the prompt how to apply it. On Sume that image goes in image_urls or input_references.

Vendor behavior is from BFL's Pose & Layout Guidance page; Sume request rules from Image generation and Image API, read 2026-10-01.

How many references can I pass?

What should the reference and prompt look like?

BFL recommends clear, uncluttered pose references, simple poses with visible limbs, and referring to images by number in the prompt ("match the pose in image 1"). It notes a single collage can guide layout but quality may be slightly lower than separate input images. BFL says 8 and Sume documents 1–10 for image_urls, so for a FLUX.2 id stay at 8 or fewer and check the model's descriptor in the Sume catalog.

Do all Sume image models accept references?

No. Models whose input_references descriptor is {"min": 0, "max": 0} are text-to-image only and reject references. Reference URLs must be public HTTPS. Read the model's descriptor from GET /v1/images/models before sending an image.

What if I send a ControlNet-style field?

A request that sets a parameter the selected model does not list is rejected with 400 unsupported_parameter rather than silently dropped. So a controlnet or depth-map field will fail; send the pose as a reference instead. More on multi-image edits: FLUX.2 multi-reference editing.