Meta's Muse Image, its first image model, comes to Vercel's AI Gateway

Muse Image lands on Vercel's AI Gateway

Vercel has added Muse Image to AI Gateway, making Meta Superintelligence Labs' first image model available through its unified model API. According to the Vercel blog, Muse Image sits in a separate family from Meta's Muse Spark and returns images rather than text, which makes it the lab's first model to produce a visual output.

Generation and editing in one model

Muse Image covers two jobs with a single model. Send a text prompt and it returns a generated image; send an image together with an instruction and it edits that image. Because both modes live in the same model, developers can move from creating an image to revising it without swapping model identifiers or rewiring their pipeline.

The model is reached through the generateImage function in Vercel's AI SDK using the identifier meta/muse-image-1.0:

js import { generateImage } from 'ai';

const { images } = await generateImage({ model: 'meta/muse-image-1.0', prompt: 'A conference poster for a talk on database indexes.', });

To steer output toward visuals you already have, Vercel says you can pass reference images in prompt.images alongside the text, and the model blends those references into what it draws. The same input channel drives editing: supply the image you want changed together with an instruction — the announcement's example moves a date on a poster to the bottom right and enlarges it — and the model alters what you asked for while leaving the rest of the image untouched.

What the Gateway wraps around it

AI Gateway positions itself as a single endpoint in front of many models, with usage and cost tracking, failover, and performance optimizations that Vercel says deliver higher uptime than the underlying provider alone. It also ships built-in custom reporting, budgets scoped to individual API keys, and routing rules.

On pricing, Vercel says the gateway reflects provider pricing with no markup and charges no platform fee on inference, including on Bring Your Own Key requests where developers use their own provider credentials. A model playground lets you try Muse Image before writing any code, and the gateway lists other image models alongside it for comparison.

Why it matters

For teams already building on the AI SDK, adding image generation is effectively a one-line change to the model string, and editing requires no second model. That consolidation is the practical draw: image pipelines often pair a generator with a separate editor, and Muse Image folds both roles into one call with one billing line.

It is also a useful data point about Meta Superintelligence Labs itself. Muse Image is described as the group's first image model, so it marks a step beyond text-only output under the Muse name. Debuting it through an aggregator like Vercel — at provider pricing, with failover and budget controls in front of it — is a low-friction way for the lab's newest model to reach developers who would otherwise never touch a Meta API directly.