Make images and videos with Muse

Three things it does

It also blends references: send two or three photos and ask it to combine them — a person, a place, a style — into one picture. And because it works with Muse Spark, it can look things up before drawing, so a real landmark or a current product looks right instead of invented.

How to make an image, step by step

1. 1Open Muse — the app or muse.ai on the web. No separate image app is needed.
2. 2Describe the picture in plain words: subject, setting, style. "A birthday card for a 7-year-old who loves dinosaurs."
3. 3Or attach a photo and say what to change. "Turn this into an oil painting." "Restore and colorize this old photo."
4. 4Refine in the same chat. Change one thing at a time — "make the sky dusk" — instead of rewriting the whole prompt.
5. 5Save it from the chat, or ask Muse to keep working with it — email it, put it in a doc, turn it into a poster.

If you cannot find the image feature in your version of the app, just ask Muse in chat where it is — Meta is iterating fast and button placement moves between versions.

Tips that actually help

• Be specific about the scene. Subject, setting, lighting, style — "a golden retriever puppy in a sunny kitchen, watercolor" beats "a cute dog."
• One change per turn when refining. Small edits land; rewriting the whole prompt starts you over.
• Use reference photos for consistency. If a character or product has to look the same across images, send the reference each time.
• Let it use the agent. The image is not the end of the job — Muse can email it, add it to a document, or order a print from it. Ask for the next step in the same chat.

What images will not do

• No @-mentioning other people's Instagram photos. That feature launched with Muse Image and was removed three days later, on 13 July 2026, after Meta said it "missed the mark." Use your own photos as references.
• Do not expect exact likeness of real people. It is a generator, not a camera — faces and hands are where it slips most.
• Check the terms before commercial use. Meta's terms for generated media are still being updated — read the current terms on muse.ai before using an image in an ad or a product.
• It is US-first. Muse Image started in the US, with other countries labelled "coming soon" by Meta. If you are outside the US, the feature may not be there yet.

Muse Video vs video inside the Muse agent

Two different things share the name. Muse Video is the video-generation model from Meta Superintelligence Labs — the thing that turns a prompt or a still image into a moving clip with sound. Muse the agent is your personal assistant that does tasks. The model was previewed in July 2026 alongside Muse Image, and Meta says it is coming to creators and Meta AI. That means video generation reaches you through Meta AI (the app and the web) as it rolls out — not as a finished, everywhere feature inside the agent. If a guide promises "ask Muse for a video and it just works," it is ahead of what Meta has shipped.

Where video already fits: a short clip for a social post, animating a product photo, turning a still image into a talking portrait, or a quick scene for a presentation. Where it does not: long videos, exact likenesses, or anything you need to look identical every time.

Two ways to make a clip

A useful prompt formula: subject + action + place + camera + sound. Example: "A red bicycle leaning against a brick wall, wheels slowly turning, quiet street at dusk, slow push-in, soft wind and distant traffic." One or two sentences is enough — long paragraphs do not help.

Generate a clip, step by step

1. 1Open the Meta AI app or meta.ai on the web, where Muse Video is rolling out.
2. 2Pick text-to-video or upload a still image for image-to-video.
3. 3Write the prompt: subject + action + place + camera + sound.
4. 4Set the aspect ratio first — 9:16 for phones, 16:9 for wide. It cannot be changed after.
5. 5Generate, then watch the whole clip with sound on. If the motion or audio is off, change one detail and generate again — small edits beat rewriting the whole prompt.

First-frame and aspect-ratio quirks

These are not in Meta's documentation — they are what users keep reporting in practice. Treat them as field notes, not rules.

• The first frame is the anchor. In image-to-video, the still image you upload is the clip's first frame, and the model works outward from it. A cropped, blurry, or busy first frame gives you a cropped, blurry, or busy clip. Start from the cleanest image you have.
• Aspect ratio is decided before generation, not after. A clip made in 16:9 cannot be re-cut to 9:16 — you regenerate. Decide where the clip will be watched (phone feed vs wide screen) before you press generate.
• Fast motion drifts. Meta says fast physical motion — quick collisions, sports — still needs work. Slow, simple motion (a push-in, a turn, a walk) holds up better than fast action.
• Sound comes with the picture. Audio is generated together with the video, not added later. If the sound is wrong, the fix is a new generation with a clearer sound cue in the prompt, not an edit.
• Faces and hands are the tell. Like every video model, Muse Video is weakest on faces in motion and hands. Keep people small in frame or still if the clip has to look clean.

What it costs

Meta publishes no per-image or per-video token figure, and no per-clip price for Muse Video inside Meta AI. Both draw on your weekly Muse allowance — the free tier is 100M Muse tokens a week, Power is $20 a month for 500M, Maximum is $100 a month for 3B — and a task that hits the weekly allowance stops where it is. That is the honest answer: media costs some of your weekly allowance, Meta does not say how much, and any site quoting an exact price is guessing. See pricing and the token limit for how the allowance works. If a clip stalls near the end, that is the same "stops mid-task" behaviour the troubleshoot page covers — narrow the task and try again.

What video will not do yet

• Long videos. Clips are short — around 10 seconds in current reports. There is no "make me a 2-minute video" path.
• Exact consistency. A character or product can drift between clips. For anything that has to match, generate from the same first frame and expect to re-roll.
• Silent clips. Sound is generated with the video; there is no published way to turn it off.
• Anything commercial without a check. Meta's terms for generated media are still being updated — read the current terms on muse.ai before using a clip in an ad or a product.

For talking to Muse instead of typing, see voice mode.