ClipCanva

Muse Image Workflow: $0.01 API Stills, Multi-Reference Edits, Then Image-to-Video

How to use Meta Muse Image at $0.01 per API still: lock references, edit once, skip withdrawn Instagram tags, then hand the keeper to image-to-video.

August 27, 2026ClipCanva Editorial

Muse Image Workflow: $0.01 API Stills, Multi-Reference Edits, Then Image-to-Video

A Muse Image workflow is not “ask Meta for a cinematic product ad and hope the bottle holds.” Meta Superintelligence Labs launched Muse Image on 7 July 2026 as an agentic image model: it can search, write code for plots and QR codes, self-refine a draft, and compose from multiple references. The news that matters for operators on 26 August 2026 is access, not the July launch copy. Meta for Developers posted that Muse Image is on the Meta Model API at $0.01 per image, with text-to-image, single-image editing, multi-image editing, and multi-reference composition in one model. Vercel’s AI Gateway listing shows the same unit price for meta/muse-image-1.0. Runway posted the same day that Muse Image is available in the Runway app. ClipCanva is not affiliated with Meta, Runway, OpenAI, or Google, and Muse Image is not a live ClipCanva generator. Lock the still on the surface you actually have, then animate a keeper with image-to-video.

Official facts for Muse Image and nearby tools

Source Public claim Workflow implication
Introducing Muse Image and Muse Video (7 Jul 2026) First image model from Meta Superintelligence Labs. Agentic: search, coding tools, self-refinement, test-time compute. Precise edits. Multi-reference composition, including people, objects, clothing, styles, and environments. Consumer access: Meta AI app, meta.ai, Instagram Stories in the US, WhatsApp in limited countries. Facebook “coming soon.” Muse Video is a preview with native audio, “coming soon.” Arena rank as of 5 July 2026: No. 2 for text-to-image, single-image editing, and multi-image editing. Treat consumer Meta AI and a developer API as different desks. Do not brief a Muse Video clip as if it shipped. Arena ranks are dated; re-check before you quote them in a client deck.
Introducing Muse Image (7 Jul 2026; updated 10 Jul 2026) Everyday creation in Meta AI. Multi-photo blending. The Instagram @-mention of public accounts was pulled after Meta said the feature “missed the mark.” Do not plan a brief that tags a public Instagram account as a reference. Use files you have rights to.
Meta for Developers (26 Aug 2026) and Vercel AI Gateway Muse Image on Meta Model API at $0.01/image. Reasons before it renders. Searches the web through iterative passes. Checks precise elements such as charts and QR codes. One model for generate, edit, and multi-reference. Gateway model id: meta/muse-image-1.0. Price the still, not the adjective “agentic.” A $0.01 call is not a 4K SLA.
Vercel AI Gateway, Muse Image 1.0 (listed 26 Aug 2026) Model id meta/muse-image-1.0. $0.01 / image. Optional reference images (playground UI shows up to four). Subject to Meta terms. Gateway access is not Meta AI. Confirm the live model id and terms before you automate.
Runway (26 Aug 2026) “Muse Image from Meta, now on Runway. Available today.” Link: app.runwayml.com. App availability is not the same as a documented Runway Dev model string. As of 27 August 2026, Runway’s public models catalog still lists other image models (gpt_image_2, gemini_image3_pro, Seedream, Gen-4 Image) and does not yet name Muse Image. Confirm the live docs before you write an API job.
GPT Image 2 on ClipCanva Live ClipCanva image generator for prompts and reference edits. Use this when the still has to be made here today.
Nano Banana 2 Lite on ClipCanva Planning page. Exact Lite model is not connected. Do not treat that URL as a generator.

These are vendor-documented limits and ranks, not a promise that your SKU, face, chart, or QR code will scan. Region, safety filters, rights, and credit rules still sit on the live account.

Pick the desk before you attach files

Muse Image fails when every brief is treated as one chat box. Official Meta materials split three jobs that fight each other if you dump them in one pile: a new still from text, an edit of one approved file, and a composition from many references.

Job Where it actually runs What you must supply What the model may invent Fail condition
New still, facts or a QR/chart Meta Model API / gateway, or Meta AI One scene card. Quote the URL or number. Official: coding tools exist for plots and QR codes Layout, type style, extra props Unreadable type, broken QR, extra SKU
Change one approved photo Same surfaces, edit mode One file plus one instruction Local change only A new product, a new face, a new room
Keep identity without locking frame one Multi-reference composition Named stills for person, wardrobe, product, set Camera and in-between lighting Untagged moodboards mixed with the hero still
Hands-on still inside ClipCanva GPT Image 2 The same scene card Whatever that live model documents Pretending Muse Image is on ClipCanva
Animate the keeper Image-to-video or AI video generator Cropped still + one verb Motion and sound only Animating a still whose label already failed
Fast Google still (not on ClipCanva) Google Nano Banana 2 Lite, if you have Gemini API access Short prompt; Google prices Lite at $0.034 / 1K image in the 30 June 2026 developer post A cheap draft, not a Meta agent loop Calling the ClipCanva Lite page a generator

If the brief is “this 30ml bottle must not change,” lock the still first. Approve the crop. Then animate. If the brief is “this person in this shirt, new set,” use reference files you own. Do not rebuild the July Instagram @-mention path. Meta already withdrew it.

Delivery card before you pay $0.01 a still

Write the card before you hit generate. Meta’s public API price on 26 August 2026 is $0.01 per image. That is cheaper than a high-quality GPT Image 2 call on Runway’s published table (20 credits at 1K/2K high, 41 credits at 4K high, with credits sold at $0.01 each). It is also cheaper than Google’s published Nano Banana 2 Lite 1K rate of $0.034. Cheap is not the same as “correct label.”

Job: [new still | edit one photo | multi-reference compose]
Surface: [Meta AI app | meta.ai | Meta Model API | Vercel meta/muse-image-1.0 | Runway app]
Identity lock: [SKU / face / wardrobe / chart numbers]
Refs: [file names you have rights to]. No public Instagram handles.
Type: [exact headline]. Fail if a letter is wrong.
QR / chart: [URL or formula]. Fail if it does not scan or the axis is unlabeled.
Output: still only. No Muse Video. No 15-second clip from this model.
Handoff: crop to 9:16 or 16:9, then image-to-video with one verb.
Success: same object, readable type, no second product.

Starter for a product still (original; not copied from Meta samples):

Job: new still, then edit once.
Subject: one 30ml frosted glass bottle, gold pump, black label “NORTH HARBOR 01”.
Set: pale oak tabletop, north-window key, no extra props.
Type on label must match the quote exactly. No slogan overlay. No second language.
Do not invent a QR code unless I paste the URL.
After the first keep: edit only the light — “north-window softbox, keep everything else the same.”
Fail if a second bottle appears or the label becomes decorative glyphs.

Draft that still on ClipCanva with GPT Image 2 or collect lines on prompt ideas, then move the file to Meta or Runway if you have Muse Image access. Do not paste meta/muse-image-1.0 into a ClipCanva prompt box. That field is not there.

A prompt formula that names the lock

Keep one skeleton. Change only the lock.

Job: [generate | edit | multi-reference]
Subject: [one object or one person]. No extras.
Action in the still: [held / pouring / worn]. Frozen, not a clip.
Camera: [tabletop / 3/4 portrait / packshot].
Light: [window / overcast / practical].
Type: [quoted string] or “no type.”
Refs: hero file = product. File B = wardrobe. File C = set. Not moodboards.
Fail if: [second SKU, new face, extra headline, unreadable type].
Keep everything else the same.

For a talking 9:16 take, do not ask Muse Image to become Muse Video. Official Meta language still lists Muse Video as coming soon. Write the spoken line on the AI script generator, freeze the still, then send the still to image-to-video:

Image-to-video. 9:16. One continuous shot.
First frame is the approved bottle still.
Motion: slow push-in, one pump press, one bead of serum.
Audio: pump click + quiet room. No invented slogan.
Fail if the label warps or a second bottle appears.

Compare the still, then pick a video model

Use the table as a routing card, not a beauty contest. Meta’s own Arena snapshot is dated 5 July 2026. Do not treat it as today’s scoreboard.

Need Muse Image GPT Image 2 (ClipCanva / APIs) Nano Banana 2 Lite
Where it runs today Meta AI, Meta Model API at $0.01/image, Runway app (catalog may lag) Live on ClipCanva Google API / Gemini surfaces. ClipCanva page is planning-only
Strength Meta claims Search, code, QR/charts, multi-reference, iterative edits Instruction-following and text-forward layouts in OpenAI’s line Speed and cost; Google’s 30 Jun 2026 post: ~4s, $0.034 at 1K
Video Muse Video is not a shipping operator tool Still only; hand off to image-to-video Still only; pair with Veo / Omni on Google if you have those
Rights trap Do not @ public Instagram accounts Use files you own SynthID on Google video/image outputs where documented
Fail if you Brief a 30-second ad in this model Expect a Meta search/code loop Open the ClipCanva Lite URL and wait for a generate button

Route the clip after the still is approved. Compare live video models on model comparison. A Veo 3.1 take, a Seedance 2.5 one-take, and a Grok Imagine start-frame job are different packets. Muse Image does not pick that packet for you.

Creator checklist

  1. Decide the job: new still, one-file edit, or multi-reference compose. Do not mix them in one prompt.
  2. Confirm the surface. Meta AI, Meta Model API, a gateway meta/muse-image-1.0 job, and the Runway app do not share one control panel.
  3. Confirm Runway in the live docs if you are automating. An X post is not an OpenAPI spec.
  4. Collect rights-cleared files. Skip public Instagram handles. Meta withdrew that path on 10 July 2026.
  5. Quote every string that must print. “Make the label premium” is not a lock.
  6. If you need a QR or chart, paste the URL or the formula. Then scan or read the axis before you ship.
  7. Budget one edit: “Keep everything else the same.” Stop if the SKU drifted.
  8. Crop the keeper to the delivery ratio before image-to-video.
  9. If you need a still inside ClipCanva today, use GPT Image 2. Use prompt ideas for the scene card.
  10. Legal review still sits with you. Official pages describe capabilities. They do not license a face, a song, a competitor logo, or a scannable campaign QR.

Limits to keep in the brief

  • July launch is not August access. 7 July 2026 is the consumer model. 26 August 2026 is the $0.01 API note and the Runway app post.
  • ClipCanva does not silently add Muse Image. The GPT Image 2 page is a different model. If that changes, re-read the live page.
  • Muse Video is not a production handoff. Official Meta language remains “coming soon” / preview. Do not sell a native-audio Muse Video spot off this still workflow.
  • Arena No. 2 is a 5 July 2026 snapshot. Meta published it. It is not a perpetual ranking.
  • Runway docs can lag the tweet. Do not hard-code a muse_image model id until Runway’s model list shows it.
  • Instagram @-mentions are withdrawn. Plan as if they do not exist.
  • Gateway playgrounds are not Meta AI. Vercel lists up to four references in its UI. Meta’s research post describes “many” references. Use the smaller number until your own account proves more.
  • $0.01 is a still. It is not 4K video, not HDR, and not a 30-second Seedance take.

FAQ

What is Muse Image?

Muse Image is Meta Superintelligence Labs’ first in-house image generation model, announced 7 July 2026. Official materials describe it as an agent: it can search, write code for plots and QR codes, self-refine drafts, edit a still, and compose from multiple references. It is not the 2023 Google “Muse” masked-transformer paper, and it is not Muse Spark (Meta’s text/reasoning API family). ClipCanva is a separate product and does not currently run Muse Image.

How much does Muse Image cost on the API?

Meta for Developers stated $0.01 per image on 26 August 2026 for Meta Model API. Vercel’s AI Gateway listing for meta/muse-image-1.0 shows the same unit price. Runway app credits are a different ledger; confirm the live Runway price board before you forecast a batch.

Can I generate Muse Image stills on ClipCanva?

No. Use GPT Image 2 if the still has to be made here today. Draft the scene on prompt ideas, then take the packet to Meta AI, the Meta Model API, or Runway if you have Muse Image access.

Is Muse Video available?

Not as a general creator tool you should brief. Meta’s 7 July post previews Muse Video with native audio and says it is coming soon. Do not write a shot list that assumes you can extend a Muse Image still inside Muse Video today.

Should I animate the still with image-to-video?

Yes, once the label, crop, and identity pass. Send the keeper to image-to-video or pick a live video model from the AI video generator. One verb. One camera move. Fail the take if the product redesigns itself.

Sources