Add /normal16 endpoint: raw float16 normals (.npy) to avoid 8-bit quantisation grain

Return normal map as lossless PNG via gr.File (avoid webp orange-peel)

Return depth as 16-bit PNG via gr.File (preserve full precision)

Fix 16-bit->8-bit depth conversion

Switch to ZeroGPU hardware (model requires float16 = GPU)

Add show_error=True to see inference error details

Fix bool-schema bug in gradio_client (crash on /gradio_api/info)

Bind to 0.0.0.0:7860 so HF proxy can reach the server

Pin transformers==4.44.0 (has FLAX_WEIGHTS_NAME); add torchvision

Add cached_download + HfFolder shims for hub>=0.30 compat

Add HfFolder compat shim for huggingface_hub>=0.30

Fix ResolutionImpossible: drop huggingface_hub pin + torch pin (HF base provides)

No gradio-client pin - let gradio 5.x sdk injection handle it

Remove gradio-client pin (conflicts with gradio 5.x sdk injection)

Bump gradio sdk to 5.6.0 (python 3.10 already pinned, audioop safe)

Override bundled gradio-client to fix json-schema bool bug

Add torchvision; loose torch pin

Remove unused import tensorboard

Pin python_version to 3.10 (audioop removed in 3.13)

Fix syntax errors from escape mangling

Slim requirements: drop torch (preinstalled), drop gradio (sdk handles)

Revert sdk_version to 4.44.0

Match haodongli build deps (torch 2.5.1, gradio 4.44)

Bump sdk_version to gradio 5.6.0

Slim requirements; upgrade gradio to v5

Minimal dual-endpoint app: /depth + /normal

saving as .tiff didn't work - it doesnt seem to calculate the depth

it's .TIFF, not TIFF

Trying to force the image to be saved as TIFF

Revert num_inference_steps=1 from 3. Didnt seem to change much

num_inference_steps=3, from 1 - trying to remove the noise in the generated image

Update to export depth maps for depth estimation as 16bits, not 8 bits

Update utils/image_utils.py

move back

update video depth

huggingface_hub==0.25.0

update lotus-depth-g-v2-1-disparity

add resizing

add resizing

add resizing

update for video depth

update gitattributes

delete old video

update github

update X

update port

remove vid

update for zerogpu

update for zerogpu

add transformers.utils.move_cache()