Muse Image (Meta Model API)

First-party Muse Image from Meta Superintelligence Labs, served on Meta Model API. One model (muse-image-1.0) does text-to-image, precise edits, and multi-reference composition. Agentic search/code tools run by default and are included in the per-image price.

There is no open-weight / local path for Muse Image. (Muse Glimmer is a separate open-weight text VLM, not an image generator.)

CLI model	API model	Services
muse-image	muse-image-1.0	generate, edit, vton (composition)

Muse Video is a consumer preview only — no developer API or weights yet.

Environment

export MODEL_API_KEY=...          # official Meta name (dashboard → API keys)
# export META_MODEL_API_KEY=...   # OpenTryOn alias
# export META_MODEL_API_BASE_URL=https://api.meta.ai/v1

Create a key at the Model API dashboard. Keys look like LLM|…|….

CLI

opentryon generate --model muse-image \
  --prompt "A fashion model walking a runway at dusk, editorial lighting" \
  --size 1024x1536 --output-format png

opentryon edit --model muse-image \
  --images look.jpg --prompt "Change the jacket to black leather, keep pose and face"

opentryon vton --model muse-image \
  --person-image model.jpg --garment-image garment.png \
  --garment-description "olive green bomber jacket"

--size is an aspect hint (WxH), not exact pixels. --reasoning-strength low skips self-refinement (faster, same $0.01/image). --no-web-search / --no-image-search / --no-shell turn off built-in tools.

Python

from tryon.api.muse import MuseImageAdapter

adapter = MuseImageAdapter()
images = adapter.generate_text_to_image(
    prompt="A watercolor editorial still of a red fox in snow",
    size="1536x1024",
    output_format="png",
)
images[0].save("muse.png")

edited = adapter.generate_image_edit(
    image=["look.jpg", "palette.jpg"],
    prompt="Keep the person from the first image; restyle using the second palette.",
)

Notes

• Auth: Authorization: Bearer {MODEL_API_KEY} against https://api.meta.ai/v1.
• Endpoints: POST /v1/images/generations, POST /v1/images/edits (JSON images[].image_url).
• VTON here is multi-image composition, not a dedicated garment-fit model — prefer FLUX VTO / FASHN when fit accuracy matters.
• MCP tools: generate_muse_image, edit_muse_image, vton_muse_image.