Clarification of Apache-2.0 coverage for published SDXL weight files #479

Hello maintainers,

The MotionPrep Studio project is evaluating IP-Adapter for paid, hosted SaaS
inference. We are not seeking legal advice; we are asking for confirmation of
the licensing scope intended by the copyright owner or authorized licensor.

The exact candidate source is:

• Hugging Face repository: h94/IP-Adapter
• Revision: 018e402774aeeddd60609b4ecdb7e298259dc729
• sdxl_models/ip-adapter-plus_sdxl_vit-h.safetensors
  • SHA-256 from Hugging Face LFS metadata:
    3f5062b8400c94b7159665b21ba5c62acdcd7682262743d7f2aefedef00e6581
• sdxl_models/image_encoder/model.safetensors
  • SHA-256 from Hugging Face LFS metadata:
    657723e09f46a7c3957df651601029f66b1748afb12b419816330f16ed45d64d

The model card declares Apache-2.0, and the official code repository contains
an Apache-2.0 license, but the Hugging Face weights repository does not expose
a separate LICENSE file at this revision.

Could you please confirm:

1. Are you the copyright owner or authorized licensor for both exact weight
   files above?
2. Does Apache-2.0 apply to both the IP-Adapter checkpoint and the bundled
   image encoder weight, rather than only to the source code?
3. Does that grant permit paid commercial hosted inference, copying the files
   into a private Docker image run by our infrastructure provider (RunPod), and
   providing generated outputs to customers without distributing the weights
   to customers?
4. Are there additional attribution, acceptable-use, output, redistribution,
   or commercial-license requirements that are not stated in Apache-2.0?

If another party owns or licenses the image encoder weight, please identify the
correct upstream model, exact license, and contact or license URL.

Thank you.