What’s new
- It's Meta's model, served through Runway: Muse Image was built by Meta Superintelligence Labs and became available through Runway's image API in 2026 — that's the route Nomi uses to call it.
- Plans first, self-checks while generating: instead of mapping a prompt straight to pixels, it reads every input image and works out the overall layout first — what goes where, how many parts, how they relate — then reviews its own output mid-generation and may make a local fix, redo the whole thing, or switch approach.
- Composes up to 10 reference images: feed it a character, a product and a background separately and it treats each as its own ingredient, keeping identity and setting consistent.
- More trustworthy text in the image: for baked-in text like prices, signage or taglines, Meta says rendering accuracy is stronger in this version.
What it’s good for
- Keeping an identity consistent: the same product across different scenes, or the same person in different looks, without drifting.
- Posters, price tags or signage-style assets that need legible in-image text.
- Stitching existing assets together: combine a character, a background and props you already have into one new image instead of generating from scratch.
Prompting tips
- Say who or what sits where: "the lamp on the desk, angled toward the keyboard, shot straight on at desk height" pins the composition far more than "there's a lamp on the desk."
- Give every reference image a job: name it in the prompt — "the tan leather tote from the second image" — instead of a vague "that image," so the model knows exactly whose look to keep.
- Write down anything you need to survive the next run: there's no seed, so two identical prompts return two different images and you can't get back to yesterday's version — anything you didn't put into words gets re-decided every time.
- Spell out exact numbers and copy for in-image text: for prices, taglines or labels, write the exact digits and wording rather than letting the model guess.
- When you're not sure, generate a batch and pick: with no seed to pin a version, a batch is how you get a choice — not repeated micro-edits of one run.
Combine the woman from reference image 1 with the tan leather tote from reference image 2 into one autumn sale poster:
she's walking down a street scattered with fallen leaves, the tote slung over one shoulder, waist-height shot, side backlight.
Leave space at the top for the headline "Autumn Arrivals - 20% Off", clean and legible type. Keep the warm brown tone from reference image 2 throughout.
Known limits
- No seed parameter: the same prompt gives a different result every time, and you can't return to a specific earlier generation.
- In multi-image composition, layout can reflow based on the inputs, and exact text rendering varies run to run — for text-critical assets, generate a few times and check.
Using it in Nomi
On the generation canvas, add a image card, pick “Runway Muse Image” in the model picker and choose a mode (Text to image, Image to image). Add your references, write the prompt and generate. In the storyboard you can pick it per shot. The first time, connect any one of the providers above under Settings → Models.
