01
In brief
Meta Muse Image 1.0 is now callable through Vercel AI Gateway, giving developers a straightforward way to add Meta's image generation and editing model to an application without building a separate provider-specific transport layer.
The current Vercel model ID is meta/muse-image-1.0. The same model handles text-to-image generation and instruction-based editing.
02
Before you start
You need a Vercel account with AI Gateway access and an API key configured for your project. Install the current AI SDK package used by your application and store the gateway key in an environment variable rather than hard-coding it into client-side code.
Vercel's current Muse Image model page should be treated as the source of truth for provider availability, pricing and API behavior because these can change after launch.
03
Basic text-to-image generation
Vercel documents Muse Image through the AI SDK image-generation interface:
const result = await generateImage({ model: 'meta/muse-image-1.0', prompt: 'A red balloon on a wooden table.' });
The important pieces are the model identifier and generateImage call. In production, wrap the call in a server route or server action so your gateway credentials never reach the browser.
04
Editing an existing image
Muse Image does not require a separate edit-only model. Vercel's integration accepts an input image alongside an instruction so the same model can modify an existing asset.
The current Vercel changelog describes the input image as being passed through prompt.images. Your application can therefore structure the request around an instruction such as 'remove the background but keep the product unchanged' together with the source image.
Because image-input types and helper signatures can change as the AI SDK evolves, copy the exact current object shape from Vercel's live Muse Image documentation when implementing rather than relying on an older cached example.
05
Using reference images
Reference images are useful when the target should inherit visual information from existing art, product shots or other assets. Vercel says Muse Image can accept reference images through prompt.images alongside the text prompt and blend those references into the generated result.
A practical workflow is:
- Upload or securely fetch the reference asset on the server.
- Pass the image with a specific text instruction describing what should be preserved versus changed.
- Generate the image with <code>meta/muse-image-1.0</code>.
- Save the returned asset to your own storage if the user needs persistence.
- Store the prompt and source references separately if your product needs reproducibility or audit history.
06
Generation vs editing
The main architectural advantage of the current Muse Image integration is that one model covers both jobs. A product can start with a blank-generation flow and then move into edits without changing model families.
For example, an ecommerce creative tool could generate a product scene, then let the user submit follow-up edits such as changing the surface, removing a prop or adapting the same composition for another campaign.
07
Current pricing
Vercel currently lists the Meta provider output price at $0.01 per image for Muse Image 1.0. Check the live model page before production budgeting because provider pricing may change.
The AI Gateway page also says free Vercel users who have not made a payment receive $5 of credits every 30 days. That is a Vercel account-credit policy, not a permanent guarantee that Muse Image itself has an unlimited free API tier.
08
Production considerations
Keep the API key server-side
Never expose an AI Gateway key in browser JavaScript. Route image requests through your own backend.
Validate uploads
If users can provide reference images, enforce file-size and media-type limits before sending assets upstream.
Expect model and provider changes
Vercel currently shows one Meta provider. Build error handling for temporary unavailability rather than assuming infinite provider redundancy.
Track cost by job
At $0.01 per output image today, repeated regeneration can add up. Record generations per user or workflow so you can enforce product-level budgets even if the underlying provider does not expose the quota you want.
Separate prompt intent from references
Tell the model what should change and what should remain fixed. Vague prompts are especially costly in editing workflows because users may repeatedly regenerate to recover details that were accidentally changed.
09
Example product workflows
Social creative generator
Generate a campaign image from a prompt, then let the user upload a brand/product reference and request format-specific edits.
Product-photo editor
Take an existing packshot and ask Muse Image to replace the background or place the product into a new scene while preserving the main object.
Moodboard or concept tool
Use several references to steer color, composition or subject matter, then generate a new visual from the combined direction.
Iterative creative assistant
Store each generated result and send the selected prior image back as the reference for the next edit. This creates an application-level iteration loop even though the exact consumer Meta AI conversation behavior is not automatically provided by the Gateway call.
10
What not to assume
Do not assume every Meta AI consumer feature is exposed through the Vercel endpoint. The API path is a developer integration for the image model; Meta AI's presets, social context, conversation state and app-specific experiences are separate product layers.
Do not invent rate limits, output resolutions, batch limits or commercial-use guarantees that are absent from the current provider documentation. Recheck Meta's terms and Vercel's live model page for production usage.
11
Frequently asked questions
What model ID should I use?
meta/muse-image-1.0 on Vercel AI Gateway.
Can the API edit an existing image?
Yes. Vercel documents instruction-based editing with an input image using the same Muse Image model.
Does Muse Image support reference images?
Yes. Vercel documents reference images alongside the text prompt through prompt.images.
What does the API cost?
Vercel currently lists $0.01 per output image for the Meta provider. Treat that as time-sensitive pricing.
Do I need a separate Meta API key?
The verified workflow covered here is Vercel AI Gateway, authenticated through the Gateway setup. Do not assume a separate direct Meta developer endpoint unless Meta publishes one.
Sources
Primary and supporting sources
Facts were rechecked against the linked sources immediately before publication. Pricing, product availability and rollout status can change.