Meta Muse Image: Text-to-Image, Editing & Composition Guide

Meta Muse Image is Meta’s first image model from Meta Superintelligence Labs. It writes images from a text prompt, edits a photo, or combines up to 10 reference images into one picture, and it is especially good at legible text, posters and infographics. On VdoBloom it costs 1 credit per image in both Text to Image and Image Edit, always at a fixed 2K size of about 2.5 MP.

Disclosure: VdoBloom is our platform.

Muse Image arrived in the Meta AI app in July 2026 and is now reachable outside Meta’s own apps. This guide explains what the model does differently, the three ways to use it on VdoBloom, the exact output sizes, how to prompt it for clean text and layouts, where its content filter draws the line, and how it compares with GPT Image 2, Nano Banana Pro and Qwen Image 2.1.

What Meta Muse Image is

Meta announced Muse Image on July 7, 2026 as the first image model built by Meta Superintelligence Labs. As of September 2026, Meta’s official announcement describes three abilities: generation that follows instructions closely, precise editing, and composition from multiple reference images. It powers image creation in the Meta AI app, on meta.ai and in parts of Instagram and WhatsApp.

The unusual part is how it works. Meta calls the model agentic: before drawing, it plans the layout, can search the web for facts and reference pictures, and can write small pieces of code to draw accurate charts and QR codes. It then checks its own draft and fixes it if needed. In practice that means:

Text comes out readable. Headlines, labels and short copy are rendered cleanly rather than as letter-shaped noise.
Infographics and charts make sense. Numbers, axes and step-by-step layouts are planned, not guessed.
Real-world things look right. Landmarks, products and logos can be checked against real images instead of invented from memory.

Meta reported that Muse Image ranked second on the Arena leaderboards for text-to-image, single-image editing and multi-image editing as of July 5, 2026. Rankings move quickly, so treat that as a snapshot.

Three ways to use Muse Image on VdoBloom

1. Text to image

Open Text to Image, pick Muse Image, write a prompt and choose a shape. In VdoBloom’s integration runs on September 25, 2026, a 16:9 image came back in about 12 seconds.

2. Edit one photo

Open Image Edit, pick Muse Image, upload a photo and describe the change. The shape setting defaults to auto, which keeps your photo’s orientation. A portrait edit took about 18 seconds and came back as a portrait.

3. Combine 2 to 10 photos

Upload between two and ten reference images on the same Image Edit tab and describe how they fit together: a product on a new background, the same character in a new scene, several items arranged in one flat lay. With several photos you can keep auto (it follows the first photo’s shape) or pick a fixed shape for the new composition.

Every option costs 1 credit per image, including the web search and planning steps. New accounts get 10 free credits, which is ten Muse images. If the provider rejects or fails a request after you are charged, the credit is returned.

Output sizes

Muse Image only produces a fixed set of 2K sizes, each around 2.5 megapixels. There is no 1K or 4K option.

Shape	Pixels	Typical use
1:1	1600 x 1600	Feed posts, product tiles, icons
3:2 and 2:3	1920 x 1280 and 1280 x 1920	Article images, portrait photos
4:3 and 3:4	1792 x 1344 and 1344 x 1792	Slides, printed flyers
16:9 and 9:16	2048 x 1152 and 1152 x 2048	Banners, thumbnails, stories and reels covers
21:9	2352 x 1008	Website headers, wide banners

A note on the auto shape. The provider offers a generic “2K” setting that is supposed to follow the reference photo’s shape. In VdoBloom’s integration runs it returned a landscape image for a portrait photo, twice. So VdoBloom does not rely on it: when you edit with auto, VdoBloom reads your photo’s dimensions and requests the closest size from the table above. A 900 x 1359 portrait, for example, is sent as 1280 x 1920 and comes back upright.

What Muse Image is best at

Images with words in them. Event posters, menu boards, book covers, YouTube thumbnails with a headline, product labels. If the words matter, Muse is a strong first pick. Our text-in-images comparison covers other options.
Infographics, charts and diagrams. “A vertical infographic with five numbered steps for repotting a plant” or “a bar chart comparing four values” benefit from its planning and code steps.
Icon sets and simple graphics. Ask for a grid of a fixed number of icons in one style.
Product mockups. Upload a product shot and ask for it on a kitchen counter, in a gift box, or on a billboard.
Consistent characters and products across a series. Reuse the same reference images for each new scene so the subject stays recognisable.
Everyday photo edits. Removing objects, changing a background, or restyling a photo into claymation or a painting; Meta also lists restoring old photos as a use case.

How to prompt Muse Image

The provider’s prompting guide recommends writing a short brief rather than a list of keywords. These habits follow from it:

Name the deliverable first. “An A4 poster for a Saturday farmers’ market” tells the model what the image is for, and it composes toward that.
State counts and positions. “Three glass bottles on a shelf, the tallest in the centre” is followed closely; “some bottles” is not.
Quote the exact text, and keep it short. Write the headline reads “FRESH EVERY SATURDAY”. Short lines render far more reliably than long paragraphs.
Add shot direction for photos. Light source, camera angle, depth of field, and where to leave empty space for text.
Give each reference image a job. “Use the woman from image 1, the jacket from image 2, and the street from image 3.”
Run a few versions. Muse has no seed, so the same prompt gives a different image each time. At 1 credit a run, making three and keeping the best is cheap.

Example prompt: A 9:16 infographic titled “How to brew pour-over coffee”. Six numbered steps from top to bottom, each with a small flat illustration and a one-line caption. Warm beige background, dark brown text, plenty of spacing.

Limits and content filtering

Strict filtering. Meta’s own moderation stays on for every request, and Muse carries a “strict” label in VdoBloom’s model picker, like Nano Banana and GPT Image. Revealing or suggestive requests, violence and anything involving public figures are more likely to be refused than on models marked for fewer restrictions. VdoBloom also runs its own check before any credit is charged.
2K only. Every image is about 2.5 MP. For 4K, use GPT Image 2 or Nano Banana Pro, or upscale the result afterwards.
Fixed shapes. Only the eight shapes in the table. Custom sizes like 5:4 are not available.
No repeatable seed. You cannot regenerate exactly the same image.
Real people. Only use photos of yourself or of people who agreed. Never use photos of minors, never create intimate or sexual images of real people, and do not make deceptive images of public figures. VdoBloom blocks sexual content involving minors and non-consensual intimate edits.

Muse Image vs GPT Image 2, Nano Banana Pro and Qwen Image 2.1

Model	Credits per image	Max size	Photos per edit	Filtering
Muse Image	1	2K (about 2.5 MP)	up to 10	Strict
GPT Image 2	3 (1K), 5 (2K), 8 (4K)	4K	up to 10	Strict
Nano Banana Pro	18 (1K or 2K), 24 (4K)	4K	up to 8	Strict
Qwen Image 2.1	3 (1K), 5 (2K)	2K	up to 10	Fewer restrictions

Muse Image is the cheapest 2K option on VdoBloom and the one to try first for text-heavy graphics, charts and posters. The Muse Image model page sums up its specs.
GPT Image 2 costs more but offers 1K, 2K and 4K and has a wide range of aspect ratios, including auto.
Nano Banana Pro is the premium choice when you want maximum detail at 4K and do not mind paying for it.
Qwen Image 2.1 is the pick for transparent-background PNGs, mask inpainting, or content that strict filters refuse. See our Qwen Image 2.1 guide.

Worked example: ten poster drafts cost 10 credits with Muse, 50 credits with GPT Image 2 at 2K, and 180 credits with Nano Banana Pro at 2K.

Frequently asked questions

Is Meta Muse Image free?

On VdoBloom the 10 free credits on a new account give you ten Muse images. After that it is 1 credit per image, with no subscription needed: credits from one-time packs never expire, and the price is shown before you generate. Inside Meta’s own apps, Muse is part of Meta AI under Meta’s own usage limits.

Can I use Muse Image outside the Meta AI app?

Yes. On VdoBloom it is available in both Text to Image and Image Edit, with the same 1-credit price.

How long does Muse Image take?

In VdoBloom’s integration runs on September 25, 2026, a 16:9 text-to-image took about 12 seconds and a portrait photo edit about 18 seconds.

Can Muse Image edit my photo without changing its shape?

Yes. Leave the shape on auto and VdoBloom matches the closest 2K size to your photo, so portraits stay portrait and landscapes stay landscape.

Does Muse Image make 4K images?

No. All outputs are fixed 2K sizes of about 2.5 MP. Use GPT Image 2 or Nano Banana Pro for 4K, or run the result through the Image Upscaler.

Why was my Muse Image prompt blocked?

Meta’s moderation is strict, and it acts on revealing, violent or public-figure content more readily than some other models. Rephrase an ambiguous prompt in neutral terms, or pick a model with fewer restrictions for allowed content. For how filters work in general, see why AI image generators refuse prompts.