Luma Ray 3.2 alternative: one /v1/generations route vs Sume's two

If you are looking for a Luma Ray 3.2 alternative that keeps your code shape, the main difference is routing. Luma's Agents API takes image and video jobs at one endpoint, POST /v1/generations, and you pick the model by name. Sume uses a separate route per media type. Sume's docs do not list a Luma id, so this is a migration of the call pattern to other models, not a like-for-like swap of Ray 3.2.

What Luma's page says

Luma's quickstart, read 2026-10-09, names three models: uni-1 and uni-1-max for image generation and editing, and ray-3.2 for video generation, editing and reframing. Jobs are asynchronous: submit, poll GET /v1/generations/{generation_id}, then download from presigned URLs. The page mentions Pay-as-you-go and Provisioned Throughput plans and an optional user_id for usage attribution. It gives no prices, so this post makes no price comparison.

Moving the code

In practice you replace one generations client with two small ones and a switch on media type. The polling loop stays the same shape; only the path and the status names change. Sume's statuses are pending, in_progress, completed, failed and cancelled.

For model choice, Sume's video catalog is seedance-2.5, seedance-2, wan-3.0, minimax-h3, minimax-h3-max and gemini-omni-flash-1.1. Check GET /v1/videos/models for the live list before you commit, since the catalog changes. A field-by-field migration of the older Ray 2 API is in the Ray 2 to Ray 3.2 post.

When Luma is the right call

If you need Ray 3.2 itself, for example for its reframe job or its look, no catalog substitution reproduces it, and Sume's docs do not claim to. Pick Sume when what you need is the async job pattern across several providers' models with one balance.

A small adapter

Keep your calling code unchanged by adding an adapter with two methods, video(prompt, seconds) and image(prompt). Behind the Luma implementation both call POST /v1/generations with a different model name. Behind the Sume implementation they call /v1/videos and /v1/images. The caller never learns which route it uses, and you can A/B the two on the same prompts.

Because Luma's quickstart gives no prices, measure cost by running a small batch and reading your invoice, then compare it with Sume's usage.cost on the same batch.

What I could not confirm

Luma's older docs URL now points to the Agents API documentation. Pricing, rate limits and Ray 3.2 resolutions are not on the quickstart page read, so they are not quoted. All vendor figures were read on 2026-10-09 and are listed in the sources; Sume figures come from its public catalog and docs the same day.