MyMedia models v3 — Chinese-CLIP text tower

Chinese-CLIP ViT-B/16 text tower — provenance and license

• Source: OFA-Sys/chinese-clip-vit-base-patch16, revision
  36e679e65c2a2fead755ae21162091293ad37834, official OFA-Sys weights.
  https://huggingface.co/OFA-Sys/chinese-clip-vit-base-patch16
  https://github.com/OFA-Sys/Chinese-CLIP
• License: MIT; see LICENSE-chinese-clip.txt with the upstream notices.
• Exported on 2026-10-06 using torch 2.14.1, transformers 5.18.0,
  ONNX 1.23.1, opset 18. fp32, single file, BERT CLS hidden state projected
  to 512 dimensions. Inputs: input_ids and attention_mask, int64 [1,52].
  Output: text_features, float32 [1,512], unnormalized.
• Same checkpoint/embedding space as the image tower in models-v2
  (image SHA256 6291342114691c97b1690645ef0a74b375fa2aa6ed60350bcf6bbf47c78f5328).
• Parity: eight zh/en queries, cosine > 0.9999 against PyTorch;
  24 text-image similarity comparisons across three deterministic synthetic
  images, maximum absolute difference 2.384185791015625e-7. See parity.json.
  App tokenizer inputs also matched the eight reference queries exactly.
• Used by MyMedia Photos for fully local, optional Labs visual search.

Model mirror v2 — Chinese-CLIP image encoder

MIT-licensed mirror of the Chinese-CLIP ViT-B/16 image tower (ONNX, exported from the official OFA-Sys checkpoint; parity cosine 1.0 vs torch). Used by MyMedia Photos for on-device dish recognition. See ATTRIBUTION.md; verify with SHA256SUMS.

Optional photo models — models-v1

MyMedia model mirrors, models-v1

These are unmodified copies of the exact ONNX files pinned by MyMedia Photos.
They are optional downloads, not a release of the desktop app. No user photos,
credentials, or personal model data are included.

SFace

File: face_recognition_sface_2021dec.onnx

Source: https://huggingface.co/opencv/face_recognition_sface

Original model: https://github.com/opencv/opencv_zoo/tree/main/models/face_recognition_sface

License: Apache-2.0; see SFace-LICENSE.txt (upstream per-model license).

LaMa ONNX, FP32

File: lama_fp32.onnx

Source and export model card: https://huggingface.co/Carve/LaMa-ONNX

Original LaMa: https://github.com/advimman/lama

ONNX export implementation: https://github.com/Carve-Photos/lama

The export model card declares Apache-2.0. Included upstream license copies:
LaMa-LICENSE.txt and LaMa-ONNX-export-LICENSE.txt. Fixed input shape: 512 × 512.

Integrity and scope

See SHA256SUMS.txt. The app independently verifies pinned SHA-256 hashes before
installation; this manifest/release cannot override them. Model provenance
and training-data considerations remain documented in docs/model-licenses.md
in the MyMedia repository. Hosting does not constitute a broader rights
warranty. Buffalo/InsightFace weights are not included in this mirror.

test_html

test