[rekram1-node]

I added Muse Image (muse-image-1.0) to the meta provider. The lab file in models/meta/ already existed, so this is just the provider entry.

providers/meta/models/
+ muse-image-1.0.toml   v2_only, [api.images], $0.01/image

It inherits modalities from the lab file (text plus PNG/JPEG/WebP in, WebP/PNG/JPEG out) and has one [api.images] table on https://api.meta.ai/v1 with sdk.opencodeai = "@opencode/ai/providers/meta/images". That entrypoint doesn't exist in @opencode/ai yet, so opencode just skips this model until it lands.

Sources:

• Models: https://dev.meta.ai/docs/models (muse-image-1.0 is the only Muse Image model, text and image in, image out)
• API: https://dev.meta.ai/docs/image-generation (/v1/images/generations and /v1/images/edits, OpenAI Images compatible, plus a JSON edit body with images[].image_url)
• Pricing: https://dev.meta.ai/docs/pricing-rate-limits (flat $0.01 per generated image, built-in search included)

I checked edits live on /v1/images/edits with a data URL input and it does a real edit of the input image, so image input stays in the modalities.

Meta also serves Muse Image over the Responses API, but I only listed the images API here since that's what the image path in opencode uses. Meta doesn't document input image size limits or a fixed output resolution, so I left those out.

Checks: bun validate passes. bun run v1:check says v1 output is byte-identical. bun test has the same 13 failures as clean dev.

[github-actions]

No actionable findings.