Muse Image API and Pricing, Explained Through a Real Pipeline

Muse Image is Meta's image generation and editing model, and it is a different kind of model from most of the image catalog. Where a typical image model renders in a single pass, Muse Image works out the request first and then draws. It takes a prompt alone, or a prompt plus reference images, and it decides the framing itself. There is no aspect ratio setting and no resolution setting to pick. We put it on the Image Generation node and on a model page on October 14, 2026, after probing it through the same path the node uses.

The pricing question has a twist, because as of October 2026 OpenRouter publishes no per-image price for Muse Image. So this post cannot quote a rate card. What it can do is report what our two test renders billed and how long they took, describe what came back, explain what the missing settings mean for a pipeline, and show how to call the model as a published API endpoint instead of writing a Meta integration.

The spec sheet

The two empty rows are the story. Every other image model on the node exposes at least an aspect ratio, and most expose a resolution tier. Muse Image exposes neither. You describe the picture, and the model decides how wide and how large it should be. In our tests a plain prompt came back as a 1920 by 1280 landscape, a 3:2 frame, and a reference edit came back as a 1600 pixel square. Both are large enough to crop for most placements, which matters more here than usual because cropping downstream is the only way to force a shape.

The model runs on the images endpoint rather than chat completions, which is a plumbing detail the node handles for you. It does mean Muse Image is a canvas model: it sits in the Image Generation node's model picker and is not one of the models you can ask for by name in Treza chat.

What Muse Image costs per image

OpenRouter lists no price for Muse Image, so there is no rate card to reproduce. What we have is two billed renders from October 14, 2026:

One cent each. The provider reports the charge as output image tokens, 733 for the prompt and 494 for the edit, and both settled at a cent. The reference image added nothing to the bill: the edit's prompt-side cost was zero. Two renders on one day is not a benchmark, and a provider with no published rate card can change the charge without notice, so treat a cent as what we saw rather than a promise. A hundred images at that rate is a dollar.

On the Treza node, Muse Image carries a one cent credit rate, so the estimate shown before a run is $0.01 per image and the charge settles on what the provider billed.

What came back

The text prompt asked for a ceramic cup of coffee on a sunlit wooden cafe table with steam rising and shallow depth of field. Muse Image returned a cream stoneware cup with a leaf of latte art, a ribbon of steam catching the light, a folded linen napkin at the left edge and a few crumbs on worn wood, with cafe chairs dissolving into bokeh behind. Nothing the prompt did not ask for, and every element the prompt named is present. That is what reasoning before rendering looks like: a brief with several parts, all of them landing.

The edit took a product still of a folded navy linen shirt and asked for the same shirt on a white marble counter in a bright studio. The model kept the shirt, its weave, collar and fold, laid it on veined marble under hard sunlight, and came in close on the fabric. The product survives and the set changes, which is what a reference edit is for. It also changed the frame shape, from the landscape still we gave it to a square, without being asked. That is the flip side of a model that picks its own framing.

What no settings means for a pipeline

Most image pipelines lean on two settings: the aspect ratio that matches the placement, and the resolution tier that matches the budget. Muse Image has neither, which changes how you wire it.

1. Describe the shape in the prompt. If a vertical frame matters, say so: a tall portrait composition, a wide banner. The model may honour it. In our tests it chose 3:2 for an unconstrained prompt and 1:1 for an edit of a landscape source, so do not count on it.
2. Crop downstream when the shape is non-negotiable. A Short needs 9:16 and a thumbnail needs 16:9, and Muse Image will not promise either. The renders come back large enough that a centre crop to the target frame keeps most of the picture, so an edit step after the node settles it.
3. Stop thinking about resolution tiers. There is no 1K draft and 4K finish on this model. Every render is one cent and comes back at whatever size the model picks. The draft-then-finish loop that saves money on Seedream 5.0 does not apply here, and does not need to.

Reference images and the edit path

With a prompt alone Muse Image generates. With a prompt plus reference images it edits or composes, on the same node. Wire the images into the Image Generation node's reference input, or drag them onto it, and write the change as the prompt: the same shirt on a marble counter, or this sneaker on a wet city street at night.

Our one-reference edit kept the subject and changed everything around it, and it billed the same cent as a plain prompt. That makes Muse Image a cheap place to try a product in several settings: five sets for five cents.

Where it sits in the price band

Listed and measured prices as of October 2026 for the image models around it, at 1K where the model has a resolution setting:

At a cent a render, Muse Image lands at the bottom of the band next to FLUX 3 Image Klein, and it is the only model in the table with no framing control. What it offers for that cent is the planning behaviour: a long brief with a harbour, a forest and a volcano, or a travel collage with a tram, tiled facades and a pastry, comes back with each part in place. For a fixed-shape slot, Klein or Qwen Image 2.1 Turbo do a similar price with the aspect ratio you asked for. For a multi-part scene or a quick product edit where someone will crop afterwards, Muse Image is the cheapest way to get it.

What you actually pay on Treza

Muse Image through a Treza pipeline is metered from a prepaid credit balance. One credit equals $0.01, packs start at $5, credits never expire, and there is no subscription. A failed generation charges nothing, and that includes a prompt the provider refuses. Building a pipeline on the canvas costs nothing either; only generation spends.

Calling Muse Image as an API

Direct access means Meta's credentials, the request shape of the images endpoint, no published rate card to reconcile against, and retry logic before you have built anything.

Through a Treza pipeline, Muse Image is a node on a canvas. Place the Image Generation node, pick the model, wire references into its reference input if you have them, and publish the graph as a single versioned endpoint. Your app calls the typed /invoke route with JSON in and JSON out, or points an existing OpenAI SDK at the OpenAI-compatible route. Swap the node to a model with an aspect ratio setting later and the endpoint your product calls stays the same.

The same pipeline is reachable from an agent. Treza's hosted MCP server exposes pipelines as tools that Claude and other MCP clients call directly, so an agent can generate the frame and hand back the URL as one step of a larger task.

Putting it in a pipeline instead of a single call

A single image is rarely the deliverable. The pattern that suits Muse Image is a cheap first node with a crop behind it: compose the multi-part key frame for a cent, crop it to the target frame, pass it to a video node as the first frame, then put narration, captions, or publishing behind the video. A fuller AI video pipeline puts a script step ahead of the image and reruns the whole graph on a schedule or from an API call.

The AI image generator page has the image side set up, the AI video generator page puts a video model behind a single generate action, the AI thumbnail generator page covers the frames you will be cropping to, and how to build an AI image generation pipeline walks through image graphs in more depth. For a neighbour with a published rate card, see MAI-Image-2.6 pricing.

Frequently Asked Questions

How much does Muse Image cost per image?

OpenRouter publishes no per-image price for Muse Image as of October 2026. In our tests on October 14, 2026, a text prompt and an edit with one reference each billed $0.01. On the Treza node the model carries a one cent credit rate, and the charge settles on what the provider bills.

Can I choose the aspect ratio or resolution with Muse Image?

No. Muse Image has no aspect ratio setting and no resolution setting, and picks the framing itself. Our prompt came back as a 1920 by 1280 landscape and our edit as a 1600 pixel square. Describe the shape in the prompt if it matters, and crop downstream when the shape is fixed.

Can Muse Image edit an image I already have?

Yes. It accepts image input as well as text. Attach the image as a reference on the Image Generation node and write the change as the prompt. In our probe it kept a folded navy linen shirt intact and moved it from its original still onto a white marble counter in a bright studio, for the same cent as a plain prompt.

Is Muse Image the same as Meta's Muse assistant?

No. Muse Image is Meta's image model, available through OpenRouter. On Treza it is a model in the Image Generation node, separate from Meta's Muse app.

Can I use Muse Image in Treza chat?

Not at the moment. Muse Image is a canvas model: it sits in the Image Generation node's model picker and runs inside pipelines, including pipelines published as API endpoints and MCP tools. Chat image requests use the models that run over chat completions.

Is there a free version of Muse Image?

Building pipelines on the canvas is free, but generation spends real provider money and is charged from a prepaid credit balance. There is no subscription, packs start at $5, credits never expire, and a failed generation charges nothing.