PhotoFlow on-device models

Model files downloaded by the PhotoFlow app on first run (LiteRT .tflite, arm64). manifest.json lists name / size / sha256 / SoC / required.

file what licence
dinov3_convnext_tiny_fp16 / _npu_SM8650 DINOv3 ConvNeXt-Tiny image embedding (grouping, aesthetics, non-photo heads). Built with DINOv3. DINOv3 licence (Meta)
lama_dilated_fp16 / _npu_SM8650 LaMa-Dilated inpainting (Clean Up), from Qualcomm AI Hub Apache-2.0
siglip2_b32_image_fp16 / _npu_SM8650 SigLIP2 B/32 image tower (subject labels, search) Apache-2.0

_npu_SM8650 files are Qualcomm QNN AOT variants for Snapdragon 8 Gen 3; the fp16 files run on GPU/CPU.

Downloads last month
57
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support