Muse Image AI Image Generator
Meta Muse Image for fast, low-cost text-to-image and multi-reference edits. Describe the scene, pick an aspect ratio, or upload up to ten images and describe the change.
Keep building
How to Use Muse Image
- Step 1
Describe One Main Scene
State the subject, setting, lighting, and style in one prompt. Keep the description focused on a single scene so the model does not split attention across competing subjects.
- Step 2
Pick the Aspect Ratio for the Layout
Use 1:1 for square cards, 16:9 for wide scenes, 9:16 for portraits, and 21:9 or 9:21 when you need a banner or tall crop. The ratio does not change the gold cost.
- Step 3
Edit with Reference Images
Upload one to ten images and describe only the change you want, such as a new cloak colour or a different background. Leave the aspect ratio unset on Quick Edit to keep the source framing.
- Step 4
Iterate Cheaply
Each image costs 2 gold, so it is a good model for testing prompt direction before you spend on a higher-cost engine.
Example Prompts
Lantern-lit dwarf blacksmith in a stone workshop, sparks from the anvil, warm firelight, oil-painting style
Misty elven forest shrine at dawn, mossy standing stones, pale gold light through leaves, 16:9 landscape
Close portrait of a scarred human ranger, wet cloak, rain on leather, muted greens and greys
Overhead map table in a candlelit war room, miniature armies, parchment maps, 21:9 banner crop
Crystal golem standing in a flooded ruin, moonlight through a collapsed roof, cool blue reflections
💡 Click the copy button to use these prompts in your own generations
Technical Specifications for Muse Image
Frequently Asked Questions for Muse Image
About Muse Image for CharGen
Muse Image is Meta's image model on CharGen, priced for fast iteration. Write a prompt, pick an aspect ratio, and generate. Add reference images when you want to edit an existing picture instead of starting from scratch.
Key Strengths
A flat 2 gold cost makes it cheap to try several directions. Nine aspect ratios cover square cards, portraits, landscapes, and banners. The same model edits up to ten reference images from a plain-language instruction.