Muse Spark breaks on image inputs

I’m encountering a reproducible AI Gateway error when sending image inputs to both:

meta/muse-spark-1.2
meta/muse-spark-1.2-contributor

Text-only requests succeed. Every request containing a valid JPEG fails with HTTP 400:

Storage limit exceeded for this team.

I reproduced this using:

OpenAI-compatible Chat Completions with image_url
Native Responses with input_text and input_image
Vercel’s official AI SDK using the current file content part
Base64 data URLs and publicly hosted image URLs
Regular and Contributor versions of Spark

The Gateway metadata shows that the request reaches resolvedProvider: "meta" using Vercel’s system credential before Meta returns the storage error. The image is small, and other vision models through the same Gateway key can process it successfully. My AI Gateway account is paid and has available credits.

Is it possible that the Meta integration’s upstream team or system credential has exhausted a file-storage quota? Spark’s model metadata lists vision and file input as supported, so this appears to be an integration/account-storage issue rather than an unsupported modality.