feature/uv-migration (#6)

- Migrate to uv for local dev, use pip for HuggingFace Spaces (5f4271a0df17dc82ec5f88b955a285d5855207a6)

.gitignore

@@ -3,4 +3,18 @@ checkpoints/**/*.pt
 checkpoints/**/*.pkl
 checkpoints/**/*.zip
 checkpoints/**/*.safetensors
-checkpoints/**/*
+checkpoints/**/*
+
+# Ignore local dependencies (installed via pip/uv)
+latentsync/
+tigersound/
+FastAudioSR/
+descript-audiotools/
+
+# Python cache and virtual environment
+__pycache__/
+*.pyc
+.pytest_cache/
+.coverage
+.venv/
+uv.lock

AGENTS.md

@@ -4,18 +4,35 @@ This file provides guidance for agentic coding assistants working in this reposi
 
 ## Build/Install Commands
 
+### HuggingFace Spaces (Deployment)
 ```bash
-# Install dependencies (includes LatentSync v1.6)
+# HuggingFace automatically runs:
 pip install -r requirements.txt
+python app.py
+```
+
+### Local Development (with uv)
+```bash
+# Setup (one-time)
+./setup_local.sh
 
-# Download checkpoints from HuggingFace
-./download_checkpoints.sh
+# Or manually:
+uv venv
+uv pip install -r requirements.txt
 
-# Run the application
+# Run app
+uv run python app.py
+```
+
+### Local Development (with pip - standard)
+```bash
+python3 -m venv .venv
+source .venv/bin/activate
+pip install -r requirements.txt
 python app.py
 ```
 
-**Note**: This project currently has NO linting, formatting, or test infrastructure configured. When making changes, ensure the application runs successfully with `python app.py`.
+**Note**: Use uv for faster dependency installation (10-100x faster than pip). This project currently has NO linting, formatting, or test infrastructure configured. When making changes, ensure the application runs successfully with `uv run python app.py` (or `python app.py` after activating venv).
 
 ---
 
@@ -204,3 +221,6 @@ The application uses modular helper functions for better maintainability:
 - **Large Models**: Models are loaded at startup; avoid unnecessary reloading
 - **Session Management**: Each user session creates a unique output directory that should be cleaned up
 - **Clean Code**: Each function should have a single responsibility - prefer creating new helper functions over adding complexity to existing ones
+- **Flash-attn-3**: Only available for Linux x86_64. Comment out in requirements.txt for local testing on macOS
+- **Dependencies Managed via uv (local) / pip (HuggingFace)**: All packages are installed in `.venv/` (gitignored)
+- **Local Dependencies Ignored**: latentsync/, tigersound/, FastAudioSR/, descript-audiotools/ directories are gitignored - packages are installed from git repos via requirements.txt

README.md

@@ -67,17 +67,79 @@ git push https://huggingface.co/spaces/YOUR_USERNAME/YOUR_SPACE_NAME main
 
 ---
 
-## Local Installation
+## 🚀 Deployment
+
+### HuggingFace Spaces
+
+1. Create a Space on HuggingFace
+2. Push this repository to your Space
+3. Done! HuggingFace will automatically:
+   - Create Python environment
+   - Install dependencies from requirements.txt
+   - Start the application
+
+**Requirements:**
+- Hardware: A10G GPU (recommended, 24GB VRAM)
+- Python: 3.10
+
+## 💻 Local Development
+
+### Option 1: Using uv (Fast - Recommended)
 
 ```bash
+# Install uv (macOS/Linux)
+curl -LsSf https://astral.sh/uv/install.sh | sh
+# Or: brew install uv
+
+# Setup and install
+./setup_local.sh
+
+# Run application
+uv run python app.py
+```
+
+**Why uv?** 10-100x faster than pip for dependency management!
+
+### Option 2: Using pip (Standard)
+
+```bash
+# Create venv
+python3 -m venv .venv
+source .venv/bin/activate  # Windows: .venv\Scripts\activate
+
 # Install dependencies
 pip install -r requirements.txt
 
-# Download checkpoints from LatentSync v1.6
-./download_checkpoints.sh
-
-# Run app
+# Run application
 python app.py
 ```
 
+## 📦 Dependencies
+
+- **requirements.txt**: All dependencies for application
+- Packages are installed in `.venv/` (ignored by git)
+- Git dependencies: LatentSync, FastAudioSR, tigersound, descript-audiotools
+
+## ⚠️ Important Notes
+
+### Flash-attn-3 for Local Testing
+
+The `flash-attn-3` package only provides wheels for Linux x86_64:
+- **HuggingFace (Linux)**: ✅ Works automatically
+- **Local (macOS)**: ❌ Will fail during installation
+
+**Workaround for local testing:**
+```bash
+# Comment out flash-attn-3 in requirements.txt for local testing
+# Uncomment before pushing to HuggingFace
+```
+
+### Checkpoints
+
+Checkpoints are automatically downloaded from `ByteDance/LatentSync-1.6` on startup.
+
+### Audio Language
+
+Target audio supports **English only**.
+
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

README_LOCAL.md

@@ -0,0 +1,104 @@
+# Local Development Guide
+
+## ⚠️ macOS ARM64 (Apple Silicon) Limitations
+
+If you're using macOS with Apple Silicon (M1/M2/M3), several ML packages **cannot be installed** because they only have wheels for Linux x86_64 and Windows:
+
+### Unavailable Packages:
+- **triton** - GPU compute library (Linux/Windows only)
+- **onnxruntime-gpu** - GPU ONNX runtime (Linux/Windows only)
+- **decord** - Video processing library (Linux/Windows only)
+- **flash-attn-3** - Attention mechanism (Linux x86_64 only)
+
+### Impact:
+❌ **ML functionality will NOT work** on macOS local development
+❌ Cannot run lipsync pipeline locally
+✅ **Can test**: UI, basic imports, code logic
+
+---
+
+## 🚀 Recommended Workflow
+
+### Option 1: Use HuggingFace Spaces (Recommended)
+
+For full ML functionality, deploy to HuggingFace Spaces:
+
+```bash
+# 1. Create Space on HuggingFace
+# 2. Push code
+git push origin main
+
+# 3. Deploy
+# HuggingFace will handle all dependencies
+```
+
+### Option 2: Use Linux/Windows Machine
+
+If you need local ML testing:
+
+1. Use a cloud VM (AWS EC2, GCP, Azure)
+2. Use WSL2 on Windows
+3. Dual-boot Linux
+
+---
+
+## 💻 Local Development on macOS
+
+### Setup
+
+```bash
+# Run setup script
+./setup_local.sh
+
+# This will install:
+# - gradio
+# - huggingface_hub
+# - Basic dependencies
+```
+
+### What You Can Test
+
+- ✅ Gradio UI
+- ✅ File upload/download
+- ✅ Basic imports
+- ✅ Code logic and structure
+- ❌ ML models and inference
+
+### Running the App
+
+```bash
+# Using uv (recommended)
+uv run python app.py
+
+# Or activate venv
+source .venv/bin/activate
+python app.py
+```
+
+---
+
+## 📦 Requirements Files
+
+- **requirements.txt**: Full dependencies for HuggingFace Spaces (Linux GPU)
+- **requirements_minimal.txt**: Basic dependencies for macOS local testing
+- **requirements_local.txt**: Excludes GPU-only packages (deprecated)
+
+---
+
+## 🔧 Troubleshooting
+
+### Error: "Package not available for macOS ARM64"
+
+This is expected for ML packages. Use HuggingFace Spaces for full functionality.
+
+### Error: "CUDA out of memory"
+
+Happens on GPU environments. Try shorter videos or lower resolution.
+
+---
+
+## 📚 Resources
+
+- [HuggingFace Spaces Docs](https://huggingface.co/docs/hub/spaces)
+- [uv Package Manager](https://docs.astral.sh/uv/)
+- [LatentSync GitHub](https://github.com/OutofAi/LatentSync-batch)

UV_SETUP.md

@@ -1,99 +0,0 @@
-# UV Setup Guide
-
-This project uses [uv](https://github.com/astral-sh/uv) for fast Python dependency management.
-
-## Prerequisites
-
-Install uv:
-```bash
-# macOS/Linux
-curl -LsSf https://astral.sh/uv/install.sh | sh
-
-# Or with Homebrew (already installed)
-brew install uv
-```
-
-## Setup Virtual Environment
-
-```bash
-# Create virtual environment with Python 3.10
-uv venv --python 3.10
-
-# Activate environment (Linux/macOS)
-source .venv/bin/activate
-
-# Or use uv run (no activation needed)
-uv run python app.py
-```
-
-## Install Dependencies
-
-```bash
-# Install all dependencies from pyproject.toml
-uv sync
-
-# Install with dev dependencies (if any)
-uv sync --all-extras
-
-# Or install from requirements.txt (backward compatibility)
-uv pip install -r requirements.txt
-```
-
-## Run Application
-
-```bash
-# Method 1: Activate venv first
-source .venv/bin/activate
-python app.py
-
-# Method 2: Use uv run (recommended)
-uv run python app.py
-```
-
-## Dependency Management
-
-### Add new dependency
-```bash
-uv add package-name
-
-# Add with version constraint
-uv add "package>=1.0,<2.0"
-
-# Add from git
-uv add git+https://github.com/user/repo.git
-```
-
-### Update dependencies
-```bash
-# Update lockfile
-uv lock --upgrade
-
-# Sync with lockfile (exact versions)
-uv sync --frozen
-```
-
-### Python version
-```bash
-# Check Python version
-uv python list
-
-# Install Python version
-uv python install 3.11
-```
-
-## HuggingFace Spaces Deployment
-
-When deploying to HuggingFace Spaces, use:
-
-```yaml
-# README.md metadata (already configured)
-sdk: gradio
-python_version: "3.10"
-```
-
-Space will automatically:
-1. Install uv (already in HuggingFace environment)
-2. Create venv
-3. Run `uv sync`
-4. Start application with `uv run python app.py`
-```

app.py

@@ -5,29 +5,12 @@ Main Gradio UI module
 
 import os
 import shutil
-import subprocess
-import sys
-
-import torchvision.transforms.functional as _F
-
-sys.modules["torchvision.transforms.functional_tensor"] = _F
-
 import gradio as gr
-import sys
-
 import torchvision.transforms.functional as _F
-
-sys.modules["torchvision.transforms.functional_tensor"] = _F
-
 from huggingface_hub import hf_hub_download
-sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
 from processing import lipsync_with_audio_target
 
-
-subprocess.check_call(
-    "pip uninstall onnxruntime onnxruntime-gpu -y && pip install onnxruntime-gpu",
-    shell=True,
-)
+sys.modules["torchvision.transforms.functional_tensor"] = _F
 
 checkpoint_dir = "checkpoints"
 os.makedirs(checkpoint_dir, exist_ok=True)
@@ -48,7 +31,7 @@ os.environ["PROCESSED_RESULTS"] = os.path.join(os.getcwd(), "processed_results")
 os.makedirs(os.environ["PROCESSED_RESULTS"], exist_ok=True)
 
 src = "checkpoints"
-dst = "/home/userapp/.cache/torch/hub/checkpoints"
+dst = os.path.expanduser("~/.cache/torch/hub/checkpoints")
 
 os.makedirs(dst, exist_ok=True)
 

audio_processing.py

@@ -48,13 +48,14 @@ def extract_audio_to_wav(input_video: str, output_dir: str) -> tuple:
     ]
     subprocess.run(cmd, check=True)
 
+    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
     audio, sr = torchaudio.load(audio_file)
-    audio = audio.to("cuda")
+    audio = audio.to(device)
 
     with torch.no_grad():
         from tigersound.look2hear.models import TIGERDNR
 
-        dnr_model = TIGERDNR.from_pretrained("JusperLee/TIGER-DnR").to("cuda").eval()
+        dnr_model = TIGERDNR.from_pretrained("JusperLee/TIGER-DnR").to(device).eval()
         dialog, effect, music = dnr_model(audio[None])
 
     torchaudio.save(vocal_file, dialog.cpu(), sr)

changes.patch

@@ -0,0 +1,771 @@
+diff --git a/.gitignore b/.gitignore
+index 44c5b3f..156b834 100644
+--- a/.gitignore
++++ b/.gitignore
+@@ -3,4 +3,18 @@ checkpoints/**/*.pt
+ checkpoints/**/*.pkl
+ checkpoints/**/*.zip
+ checkpoints/**/*.safetensors
+-checkpoints/**/*
+
++checkpoints/**/*
++
++# Ignore local dependencies (installed via pip/uv)
++latentsync/
++tigersound/
++FastAudioSR/
++descript-audiotools/
++
++# Python cache and virtual environment
++__pycache__/
++*.pyc
++.pytest_cache/
++.coverage
++.venv/
++uv.lock
+
+diff --git a/AGENTS.md b/AGENTS.md
+index 68e12ad..a60779d 100644
+--- a/AGENTS.md
++++ b/AGENTS.md
+@@ -4,18 +4,35 @@ This file provides guidance for agentic coding assistants working in this reposi
+ 
+ ## Build/Install Commands
+ 
++### HuggingFace Spaces (Deployment)
+ ```bash
+-# Install dependencies (includes LatentSync v1.6)
++# HuggingFace automatically runs:
+ pip install -r requirements.txt
++python app.py
++```
++
++### Local Development (with uv)
++```bash
++# Setup (one-time)
++./setup_local.sh
+ 
+-# Download checkpoints from HuggingFace
+-./download_checkpoints.sh
++# Or manually:
++uv venv
++uv pip install -r requirements.txt
+ 
+-# Run the application
++# Run app
++uv run python app.py
++```
++
++### Local Development (with pip - standard)
++```bash
++python3 -m venv .venv
++source .venv/bin/activate
++pip install -r requirements.txt
+ python app.py
+ ```
+ 
+-**Note**: This project currently has NO linting, formatting, or test infrastructure configured. When making changes, ensure the application runs successfully with `python app.py`.
++**Note**: Use uv for faster dependency installation (10-100x faster than pip). This project currently has NO linting, formatting, or test infrastructure configured. When making changes, ensure the application runs successfully with `uv run python app.py` (or `python app.py` after activating venv).
+ 
+ ---
+ 
+@@ -204,3 +221,6 @@ The application uses modular helper functions for better maintainability:
+ - **Large Models**: Models are loaded at startup; avoid unnecessary reloading
+ - **Session Management**: Each user session creates a unique output directory that should be cleaned up
+ - **Clean Code**: Each function should have a single responsibility - prefer creating new helper functions over adding complexity to existing ones
++- **Flash-attn-3**: Only available for Linux x86_64. Comment out in requirements.txt for local testing on macOS
++- **Dependencies Managed via uv (local) / pip (HuggingFace)**: All packages are installed in `.venv/` (gitignored)
++- **Local Dependencies Ignored**: latentsync/, tigersound/, FastAudioSR/, descript-audiotools/ directories are gitignored - packages are installed from git repos via requirements.txt
+diff --git a/README.md b/README.md
+index c52cd64..4af0bbb 100644
+--- a/README.md
++++ b/README.md
+@@ -67,17 +67,79 @@ git push https://huggingface.co/spaces/YOUR_USERNAME/YOUR_SPACE_NAME main
+ 
+ ---
+ 
+-## Local Installation
++## 🚀 Deployment
++
++### HuggingFace Spaces
++
++1. Create a Space on HuggingFace
++2. Push this repository to your Space
++3. Done! HuggingFace will automatically:
++   - Create Python environment
++   - Install dependencies from requirements.txt
++   - Start the application
++
++**Requirements:**
++- Hardware: A10G GPU (recommended, 24GB VRAM)
++- Python: 3.10
++
++## 💻 Local Development
++
++### Option 1: Using uv (Fast - Recommended)
+ 
+ ```bash
++# Install uv (macOS/Linux)
++curl -LsSf https://astral.sh/uv/install.sh | sh
++# Or: brew install uv
++
++# Setup and install
++./setup_local.sh
++
++# Run application
++uv run python app.py
++```
++
++**Why uv?** 10-100x faster than pip for dependency management!
++
++### Option 2: Using pip (Standard)
++
++```bash
++# Create venv
++python3 -m venv .venv
++source .venv/bin/activate  # Windows: .venv\Scripts\activate
++
+ # Install dependencies
+ pip install -r requirements.txt
+ 
+-# Download checkpoints from LatentSync v1.6
+-./download_checkpoints.sh
+-
+-# Run app
++# Run application
+ python app.py
+ ```
+ 
++## 📦 Dependencies
++
++- **requirements.txt**: All dependencies for application
++- Packages are installed in `.venv/` (ignored by git)
++- Git dependencies: LatentSync, FastAudioSR, tigersound, descript-audiotools
++
++## ⚠️ Important Notes
++
++### Flash-attn-3 for Local Testing
++
++The `flash-attn-3` package only provides wheels for Linux x86_64:
++- **HuggingFace (Linux)**: ✅ Works automatically
++- **Local (macOS)**: ❌ Will fail during installation
++
++**Workaround for local testing:**
++```bash
++# Comment out flash-attn-3 in requirements.txt for local testing
++# Uncomment before pushing to HuggingFace
++```
++
++### Checkpoints
++
++Checkpoints are automatically downloaded from `ByteDance/LatentSync-1.6` on startup.
++
++### Audio Language
++
++Target audio supports **English only**.
++
+ Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+diff --git a/README_LOCAL.md b/README_LOCAL.md
+new file mode 100644
+index 0000000..e2b14aa
+--- /dev/null
++++ b/README_LOCAL.md
+@@ -0,0 +1,104 @@
++# Local Development Guide
++
++## ⚠️ macOS ARM64 (Apple Silicon) Limitations
++
++If you're using macOS with Apple Silicon (M1/M2/M3), several ML packages **cannot be installed** because they only have wheels for Linux x86_64 and Windows:
++
++### Unavailable Packages:
++- **triton** - GPU compute library (Linux/Windows only)
++- **onnxruntime-gpu** - GPU ONNX runtime (Linux/Windows only)
++- **decord** - Video processing library (Linux/Windows only)
++- **flash-attn-3** - Attention mechanism (Linux x86_64 only)
++
++### Impact:
++❌ **ML functionality will NOT work** on macOS local development
++❌ Cannot run lipsync pipeline locally
++✅ **Can test**: UI, basic imports, code logic
++
++---
++
++## 🚀 Recommended Workflow
++
++### Option 1: Use HuggingFace Spaces (Recommended)
++
++For full ML functionality, deploy to HuggingFace Spaces:
++
++```bash
++# 1. Create Space on HuggingFace
++# 2. Push code
++git push origin main
++
++# 3. Deploy
++# HuggingFace will handle all dependencies
++```
++
++### Option 2: Use Linux/Windows Machine
++
++If you need local ML testing:
++
++1. Use a cloud VM (AWS EC2, GCP, Azure)
++2. Use WSL2 on Windows
++3. Dual-boot Linux
++
++---
++
++## 💻 Local Development on macOS
++
++### Setup
++
++```bash
++# Run setup script
++./setup_local.sh
++
++# This will install:
++# - gradio
++# - huggingface_hub
++# - Basic dependencies
++```
++
++### What You Can Test
++
++- ✅ Gradio UI
++- ✅ File upload/download
++- ✅ Basic imports
++- ✅ Code logic and structure
++- ❌ ML models and inference
++
++### Running the App
++
++```bash
++# Using uv (recommended)
++uv run python app.py
++
++# Or activate venv
++source .venv/bin/activate
++python app.py
++```
++
++---
++
++## 📦 Requirements Files
++
++- **requirements.txt**: Full dependencies for HuggingFace Spaces (Linux GPU)
++- **requirements_minimal.txt**: Basic dependencies for macOS local testing
++- **requirements_local.txt**: Excludes GPU-only packages (deprecated)
++
++---
++
++## 🔧 Troubleshooting
++
++### Error: "Package not available for macOS ARM64"
++
++This is expected for ML packages. Use HuggingFace Spaces for full functionality.
++
++### Error: "CUDA out of memory"
++
++Happens on GPU environments. Try shorter videos or lower resolution.
++
++---
++
++## 📚 Resources
++
++- [HuggingFace Spaces Docs](https://huggingface.co/docs/hub/spaces)
++- [uv Package Manager](https://docs.astral.sh/uv/)
++- [LatentSync GitHub](https://github.com/OutofAi/LatentSync-batch)
+diff --git a/UV_SETUP.md b/UV_SETUP.md
+deleted file mode 100644
+index d8d46b4..0000000
+--- a/UV_SETUP.md
++++ /dev/null
+@@ -1,99 +0,0 @@
+-# UV Setup Guide
+-
+-This project uses [uv](https://github.com/astral-sh/uv) for fast Python dependency management.
+-
+-## Prerequisites
+-
+-Install uv:
+-```bash
+-# macOS/Linux
+-curl -LsSf https://astral.sh/uv/install.sh | sh
+-
+-# Or with Homebrew (already installed)
+-brew install uv
+-```
+-
+-## Setup Virtual Environment
+-
+-```bash
+-# Create virtual environment with Python 3.10
+-uv venv --python 3.10
+-
+-# Activate environment (Linux/macOS)
+-source .venv/bin/activate
+-
+-# Or use uv run (no activation needed)
+-uv run python app.py
+-```
+-
+-## Install Dependencies
+-
+-```bash
+-# Install all dependencies from pyproject.toml
+-uv sync
+-
+-# Install with dev dependencies (if any)
+-uv sync --all-extras
+-
+-# Or install from requirements.txt (backward compatibility)
+-uv pip install -r requirements.txt
+-```
+-
+-## Run Application
+-
+-```bash
+-# Method 1: Activate venv first
+-source .venv/bin/activate
+-python app.py
+-
+-# Method 2: Use uv run (recommended)
+-uv run python app.py
+-```
+-
+-## Dependency Management
+-
+-### Add new dependency
+-```bash
+-uv add package-name
+-
+-# Add with version constraint
+-uv add "package>=1.0,<2.0"
+-
+-# Add from git
+-uv add git+https://github.com/user/repo.git
+-```
+-
+-### Update dependencies
+-```bash
+-# Update lockfile
+-uv lock --upgrade
+-
+-# Sync with lockfile (exact versions)
+-uv sync --frozen
+-```
+-
+-### Python version
+-```bash
+-# Check Python version
+-uv python list
+-
+-# Install Python version
+-uv python install 3.11
+-```
+-
+-## HuggingFace Spaces Deployment
+-
+-When deploying to HuggingFace Spaces, use:
+-
+-```yaml
+-# README.md metadata (already configured)
+-sdk: gradio
+-python_version: "3.10"
+-```
+-
+-Space will automatically:
+-1. Install uv (already in HuggingFace environment)
+-2. Create venv
+-3. Run `uv sync`
+-4. Start application with `uv run python app.py`
+-```
+diff --git a/app.py b/app.py
+index c9dd4f2..d8948ee 100644
+--- a/app.py
++++ b/app.py
+@@ -5,29 +5,12 @@ Main Gradio UI module
+ 
+ import os
+ import shutil
+-import subprocess
+-import sys
+-
+-import torchvision.transforms.functional as _F
+-
+-sys.modules["torchvision.transforms.functional_tensor"] = _F
+-
+ import gradio as gr
+-import sys
+-
+ import torchvision.transforms.functional as _F
+-
+-sys.modules["torchvision.transforms.functional_tensor"] = _F
+-
+ from huggingface_hub import hf_hub_download
+-sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+ from processing import lipsync_with_audio_target
+ 
+-
+-subprocess.check_call(
+-    "pip uninstall onnxruntime onnxruntime-gpu -y && pip install onnxruntime-gpu",
+-    shell=True,
+-)
++sys.modules["torchvision.transforms.functional_tensor"] = _F
+ 
+ checkpoint_dir = "checkpoints"
+ os.makedirs(checkpoint_dir, exist_ok=True)
+@@ -48,7 +31,7 @@ os.environ["PROCESSED_RESULTS"] = os.path.join(os.getcwd(), "processed_results")
+ os.makedirs(os.environ["PROCESSED_RESULTS"], exist_ok=True)
+ 
+ src = "checkpoints"
+-dst = "/home/userapp/.cache/torch/hub/checkpoints"
++dst = os.path.expanduser("~/.cache/torch/hub/checkpoints")
+ 
+ os.makedirs(dst, exist_ok=True)
+ 
+diff --git a/audio_processing.py b/audio_processing.py
+index 83b6706..61ec7d9 100644
+--- a/audio_processing.py
++++ b/audio_processing.py
+@@ -48,13 +48,14 @@ def extract_audio_to_wav(input_video: str, output_dir: str) -> tuple:
+     ]
+     subprocess.run(cmd, check=True)
+ 
++    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+     audio, sr = torchaudio.load(audio_file)
+-    audio = audio.to("cuda")
++    audio = audio.to(device)
+ 
+     with torch.no_grad():
+         from tigersound.look2hear.models import TIGERDNR
+ 
+-        dnr_model = TIGERDNR.from_pretrained("JusperLee/TIGER-DnR").to("cuda").eval()
++        dnr_model = TIGERDNR.from_pretrained("JusperLee/TIGER-DnR").to(device).eval()
+         dialog, effect, music = dnr_model(audio[None])
+ 
+     torchaudio.save(vocal_file, dialog.cpu(), sr)
+diff --git a/config.py b/config.py
+index da48f4e..d804c47 100644
+--- a/config.py
++++ b/config.py
+@@ -8,4 +8,4 @@ MIN_DURATION = 5
+ MAX_DURATION = 60
+ 
+ # Processing directory
+-PROCESSED_RESULTS_DIR = "proprocess_results"
++PROCESSED_RESULTS_DIR = "processed_results"
+diff --git a/lipsync_processing.py b/lipsync_processing.py
+index 5e51806..3411fe9 100644
+--- a/lipsync_processing.py
++++ b/lipsync_processing.py
+@@ -2,6 +2,7 @@
+ 
+ import os
+ import traceback
++from lipsync import apply_lipsync
+ 
+ 
+ def apply_lipsync_to_video(
+@@ -19,9 +20,7 @@ def apply_lipsync_to_video(
+     """
+     try:
+         lipsynced_video = os.path.join(output_dir, "output_with_lipsync.mp4")
+-        apply_lipsync(
+-            video_path, audio_16k_path, lipsynced_video, method=lipsync_method
+-        )
++        apply_lipsync(video_path, audio_16k_path, lipsynced_video)
+         return lipsynced_video
+     except RuntimeError as e:
+         if "out of memory" in str(e).lower():
+diff --git a/processing.py b/processing.py
+index d7ee836..61c1a73 100644
+--- a/processing.py
++++ b/processing.py
+@@ -136,7 +136,7 @@ def process_lipsync_with_audio_target(
+ 
+         with timer("Applying lipsync"):
+             lipsynced_video = apply_lipsync_to_video(
+-                video_looped, audio_16k, output_dir, lipsync_method
++                video_looped, audio_16k, output_dir
+             )
+ 
+         with timer("Upscaling video to original resolution"):
+diff --git a/pyproject.toml b/pyproject.toml
+deleted file mode 100644
+index 6d43dcb..0000000
+--- a/pyproject.toml
++++ /dev/null
+@@ -1,62 +0,0 @@
+-[project]
+-name = "outoflipsync"
+-version = "0.1.0"
+-description = "LipSync application using LatentSync v1.6 with video and audio input"
+-readme = "README.md"
+-requires-python = ">=3.9"
+-dependencies = [
+-    "torchaudio==2.8.0",
+-    "torchvision==0.23.0",
+-    "triton",
+-    "deepspeed==0.17.1",
+-    "flash-attn-3 @ https://huggingface.co/alexnasa/flash-attn-3/resolve/main/128/flash_attn_3-3.0.0b1-cp39-abi3-linux_x86_64.whl",
+-    "latentsync @ git+https://github.com/OutofAi/LatentSync-batch.git",
+-    "FastAudioSR @ git+https://github.com/ysharma3501/FlashSR.git",
+-    "DeepCache",
+-    "pydub==0.25.1",
+-    "ffmpeg-python==0.2.0",
+-    "python_speech_features==0.6",
+-    "librosa==0.10.2.post1",
+-    "accelerate==1.8.1",
+-    "transformers==4.52.3",
+-    "tokenizers==0.21.0",
+-    "sentencepiece",
+-    "g2p-en==2.1.0",
+-    "omegaconf==2.3.0",
+-    "munch==4.0.0",
+-    "tqdm",
+-    "diffusers==0.33.1",
+-    "huggingface-hub<1.0",
+-    "imageio==2.27.0",
+-    "decord==0.6.0",
+-    "opencv-python==4.9.0.80",
+-    "mediapipe==0.10.11",
+-    "av",
+-    "torch-fidelity==0.3.0",
+-    "torchmetrics==1.3.1",
+-    "lpips==0.1.4",
+-    "face-alignment==1.4.1",
+-    "insightface==0.7.3",
+-    "kornia==0.8.0",
+-    "numpy==1.26.2",
+-    "matplotlib==3.8.2",
+-    "numba==0.58.1",
+-    "Cython==3.0.7",
+-    "einops==0.7.0",
+-    "ninja==1.11.1.1",
+-    "descript-audiotools @ git+https://github.com/descriptinc/audiotools.git",
+-    "tigersound @ git+https://github.com/OutofAi/tigersound.git",
+-    "hf-xet==1.1.8",
+-    "modelscope==1.27.0",
+-    "onnxruntime-gpu==1.21.0",
+-    "realesrgan>=0.3.0",
+-    "basicsr>=1.4.2",
+-    "gradio==6.4.0",
+-]
+-
+-[build-system]
+-requires = ["setuptools>=64", "wheel"]
+-build-backend = "setuptools.build_meta"
+-
+-[tool.uv]
+-dev-dependencies = []
+diff --git a/requirements.txt b/requirements.txt
+index 7dc40a0..a3a9e67 100644
+--- a/requirements.txt
++++ b/requirements.txt
+@@ -3,9 +3,9 @@ torchaudio==2.8.0
+ torchvision
+ triton
+ deepspeed==0.17.1
+-flash-attn-3 @ https://huggingface.co/alexnasa/flash-attn-3/resolve/main/128/flash_attn_3-3.0.0b1-cp39-abi3-linux_x86_64.whl
+-latentsync @ git+https://github.com/OutofAi/LatentSync-batch.git
+-FastAudioSR @ git+https://github.com/ysharma3501/FlashSR.git
++https://huggingface.co/alexnasa/flash-attn-3/resolve/main/128/flash_attn_3-3.0.0b1-cp39-abi3-linux_x86_64.whl
++git+https://github.com/OutofAi/LatentSync-batch.git
++git+https://github.com/ysharma3501/FlashSR.git
+ DeepCache
+ 
+ pydub==0.25.1
+@@ -45,8 +45,8 @@ einops==0.7.0
+ ninja==1.11.1.1
+ 
+ # --- Model Repos & Tools ---
+-descript-audiotools @ git+https://github.com/descriptinc/audiotools.git
+-tigersound @ git+https://github.com/OutofAi/tigersound.git
++git+https://github.com/descriptinc/audiotools.git
++git+https://github.com/OutofAi/tigersound.git
+ hf-xet==1.1.8
+ modelscope==1.27.0
+ onnxruntime-gpu==1.21.0
+diff --git a/requirements_local.txt b/requirements_local.txt
+new file mode 100644
+index 0000000..4a78a20
+--- /dev/null
++++ b/requirements_local.txt
+@@ -0,0 +1,54 @@
++# --- Core Torch Stack ---
++torchaudio==2.8.0
++torchvision
++triton
++deepspeed==0.17.1
++git+https://github.com/OutofAi/LatentSync-batch.git
++git+https://github.com/ysharma3501/FlashSR.git
++DeepCache
++
++pydub==0.25.1
++ffmpeg-python==0.2.0
++python_speech_features==0.6
++librosa==0.10.2.post1
++
++accelerate==1.8.1
++transformers==4.52.3
++tokenizers==0.21.0
++sentencepiece
++g2p-en==2.1.0
++omegaconf==2.3.0
++munch==4.0.0
++tqdm
++
++diffusers==0.33.1
++huggingface-hub<1.0
++imageio==2.27.0
++opencv-python==4.9.0.80
++mediapipe==0.10.11
++av
++torch-fidelity==0.3.0
++torchmetrics==1.3.1
++lpips==0.1.4
++face-alignment==1.4.1
++insightface==0.7.3
++kornia==0.8.0
++
++# --- Numerical / Scientific ---
++numpy==1.26.2
++matplotlib==3.8.2
++numba==0.58.1
++Cython==3.0.7
++einops==0.7.0
++ninja==1.11.1.1
++
++# --- Model Repos & Tools ---
++git+https://github.com/descriptinc/audiotools.git
++git+https://github.com/OutofAi/tigersound.git
++hf-xet==1.1.8
++modelscope==1.27.0
++onnxruntime==1.21.0
++
++# --- Real-ESRGAN for Video Upscaling ---
++realesrgan>=0.3.0
++basicsr
+diff --git a/requirements_minimal.txt b/requirements_minimal.txt
+new file mode 100644
+index 0000000..e5a2d30
+--- /dev/null
++++ b/requirements_minimal.txt
+@@ -0,0 +1,13 @@
++# Minimal requirements for local testing on macOS (ARM64)
++# Excludes GPU-only packages: triton, onnxruntime-gpu, decord, flash-attn-3
++
++gradio==6.4.0
++huggingface-hub<1.0
++omegaconf==2.3.0
++tqdm
++ffmpeg-python==0.2.0
++python_speech_features==0.6
++numpy
++imageio
++opencv-python
++pillow
+diff --git a/setup_local.sh b/setup_local.sh
+new file mode 100755
+index 0000000..6a20031
+--- /dev/null
++++ b/setup_local.sh
+@@ -0,0 +1,41 @@
++#!/bin/bash
++set -e
++
++echo "🚀 Setting up OutofLipSync with uv (local development)..."
++
++# Check if uv is installed
++if ! command -v uv &> /dev/null; then
++    echo "📦 Installing uv..."
++    curl -LsSf https://astral.sh/uv/install.sh | sh
++    export PATH="$HOME/.cargo/bin:$PATH"
++fi
++
++echo "🔧 Creating virtual environment..."
++uv venv
++
++# Detect OS
++OS="$(uname -s)"
++ARCH="$(uname -m)"
++
++if [[ "$OS" == "Darwin" && "$ARCH" == "arm64" ]]; then
++    echo "⚠️  macOS ARM64 detected"
++    echo "📦 Installing minimal dependencies (GPU packages not supported on macOS ARM64)..."
++    uv pip install -r requirements_minimal.txt
++    echo ""
++    echo "⚠️  WARNING: The following packages cannot be installed on macOS ARM64:"
++    echo "   - triton (GPU compute)"
++    echo "   - onnxruntime-gpu (GPU inference)"
++    echo "   - decord (video processing)"
++    echo "   - flash-attn-3 (attention mechanism)"
++    echo ""
++    echo "   Full ML functionality is NOT available on macOS."
++    echo "   For testing, use HuggingFace Spaces (Linux GPU environment)."
++else
++    echo "📥 Installing all dependencies from requirements.txt..."
++    uv pip install -r requirements.txt
++fi
++
++echo ""
++echo "✅ Setup complete!"
++echo "🎉 Run app with: uv run python app.py"
++echo "   Or: source .venv/bin/activate && python app.py"
+diff --git a/time_util.py b/time_util.py
+index 3f23746..7e90c1e 100644
+--- a/time_util.py
++++ b/time_util.py
+@@ -1,9 +1,11 @@
+ import time
+ from contextlib import contextmanager
+ 
++
+ @contextmanager
+ def timer(name: str):
+     start = time.time()
+     print(f"{name}...")
+     yield
+-    print(f"  -> {name} completed in {time.time() - start:.2f} sec")
+
++    print(f"  -> {name} completed in {time.time() - start:.2f} sec")
++
+diff --git a/video_processing.py b/video_processing.py
+index 0647df9..ab1d19c 100644
+--- a/video_processing.py
++++ b/video_processing.py
+@@ -7,6 +7,7 @@ import subprocess
+ import cv2
+ import numpy as np
+ import torch
++from fractions import Fraction
+ 
+ import torchvision.transforms.functional as F
+ 
+@@ -110,7 +111,7 @@ def get_video_info(video_path: str) -> dict:
+ 
+     width = data["streams"][0]["width"]
+     height = data["streams"][0]["height"]
+-    fps = eval(data["streams"][0]["r_frame_rate"])
++    fps = float(Fraction(data["streams"][0]["r_frame_rate"]))
+     duration = float(data["format"]["duration"])
+ 
+     return {"width": width, "height": height, "fps": fps, "duration": duration}
+@@ -247,6 +248,7 @@ def upscale_video(video_path: str, width: int, height: int, output_dir: str) ->
+             num_grow_ch=32,
+             scale=4,
+         )
++        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+         upsampler = RealESRGANer(
+             scale=4,
+             model_path=model_path,
+@@ -255,7 +257,7 @@ def upscale_video(video_path: str, width: int, height: int, output_dir: str) ->
+             tile_pad=10,
+             pre_pad=0,
+             half=True,
+-            device=torch.device("cuda"),
++            device=device,
+         )
+ 
+         cap = cv2.VideoCapture(video_path)

config.py

@@ -8,4 +8,4 @@ MIN_DURATION = 5
 MAX_DURATION = 60
 
 # Processing directory
-PROCESSED_RESULTS_DIR = "proprocess_results"
+PROCESSED_RESULTS_DIR = "processed_results"

lipsync_processing.py

@@ -2,6 +2,7 @@
 
 import os
 import traceback
+from lipsync import apply_lipsync
 
 
 def apply_lipsync_to_video(
@@ -19,9 +20,7 @@ def apply_lipsync_to_video(
     """
     try:
         lipsynced_video = os.path.join(output_dir, "output_with_lipsync.mp4")
-        apply_lipsync(
-            video_path, audio_16k_path, lipsynced_video, method=lipsync_method
-        )
+        apply_lipsync(video_path, audio_16k_path, lipsynced_video)
         return lipsynced_video
     except RuntimeError as e:
         if "out of memory" in str(e).lower():

processing.py

@@ -136,7 +136,7 @@ def process_lipsync_with_audio_target(
 
         with timer("Applying lipsync"):
             lipsynced_video = apply_lipsync_to_video(
-                video_looped, audio_16k, output_dir, lipsync_method
+                video_looped, audio_16k, output_dir
             )
 
         with timer("Upscaling video to original resolution"):

pyproject.toml

@@ -1,62 +0,0 @@
-[project]
-name = "outoflipsync"
-version = "0.1.0"
-description = "LipSync application using LatentSync v1.6 with video and audio input"
-readme = "README.md"
-requires-python = ">=3.9"
-dependencies = [
-    "torchaudio==2.8.0",
-    "torchvision==0.23.0",
-    "triton",
-    "deepspeed==0.17.1",
-    "flash-attn-3 @ https://huggingface.co/alexnasa/flash-attn-3/resolve/main/128/flash_attn_3-3.0.0b1-cp39-abi3-linux_x86_64.whl",
-    "latentsync @ git+https://github.com/OutofAi/LatentSync-batch.git",
-    "FastAudioSR @ git+https://github.com/ysharma3501/FlashSR.git",
-    "DeepCache",
-    "pydub==0.25.1",
-    "ffmpeg-python==0.2.0",
-    "python_speech_features==0.6",
-    "librosa==0.10.2.post1",
-    "accelerate==1.8.1",
-    "transformers==4.52.3",
-    "tokenizers==0.21.0",
-    "sentencepiece",
-    "g2p-en==2.1.0",
-    "omegaconf==2.3.0",
-    "munch==4.0.0",
-    "tqdm",
-    "diffusers==0.33.1",
-    "huggingface-hub<1.0",
-    "imageio==2.27.0",
-    "decord==0.6.0",
-    "opencv-python==4.9.0.80",
-    "mediapipe==0.10.11",
-    "av",
-    "torch-fidelity==0.3.0",
-    "torchmetrics==1.3.1",
-    "lpips==0.1.4",
-    "face-alignment==1.4.1",
-    "insightface==0.7.3",
-    "kornia==0.8.0",
-    "numpy==1.26.2",
-    "matplotlib==3.8.2",
-    "numba==0.58.1",
-    "Cython==3.0.7",
-    "einops==0.7.0",
-    "ninja==1.11.1.1",
-    "descript-audiotools @ git+https://github.com/descriptinc/audiotools.git",
-    "tigersound @ git+https://github.com/OutofAi/tigersound.git",
-    "hf-xet==1.1.8",
-    "modelscope==1.27.0",
-    "onnxruntime-gpu==1.21.0",
-    "realesrgan>=0.3.0",
-    "basicsr>=1.4.2",
-    "gradio==6.4.0",
-]
-
-[build-system]
-requires = ["setuptools>=64", "wheel"]
-build-backend = "setuptools.build_meta"
-
-[tool.uv]
-dev-dependencies = []

requirements.txt

@@ -3,9 +3,9 @@ torchaudio==2.8.0
 torchvision
 triton
 deepspeed==0.17.1
-flash-attn-3 @ https://huggingface.co/alexnasa/flash-attn-3/resolve/main/128/flash_attn_3-3.0.0b1-cp39-abi3-linux_x86_64.whl
-latentsync @ git+https://github.com/OutofAi/LatentSync-batch.git
-FastAudioSR @ git+https://github.com/ysharma3501/FlashSR.git
+https://huggingface.co/alexnasa/flash-attn-3/resolve/main/128/flash_attn_3-3.0.0b1-cp39-abi3-linux_x86_64.whl
+git+https://github.com/OutofAi/LatentSync-batch.git
+git+https://github.com/ysharma3501/FlashSR.git
 DeepCache
 
 pydub==0.25.1
@@ -45,8 +45,8 @@ einops==0.7.0
 ninja==1.11.1.1
 
 # --- Model Repos & Tools ---
-descript-audiotools @ git+https://github.com/descriptinc/audiotools.git
-tigersound @ git+https://github.com/OutofAi/tigersound.git
+git+https://github.com/descriptinc/audiotools.git
+git+https://github.com/OutofAi/tigersound.git
 hf-xet==1.1.8
 modelscope==1.27.0
 onnxruntime-gpu==1.21.0

requirements_local.txt

@@ -0,0 +1,54 @@
+# --- Core Torch Stack ---
+torchaudio==2.8.0
+torchvision
+triton
+deepspeed==0.17.1
+git+https://github.com/OutofAi/LatentSync-batch.git
+git+https://github.com/ysharma3501/FlashSR.git
+DeepCache
+
+pydub==0.25.1
+ffmpeg-python==0.2.0
+python_speech_features==0.6
+librosa==0.10.2.post1
+
+accelerate==1.8.1
+transformers==4.52.3
+tokenizers==0.21.0
+sentencepiece
+g2p-en==2.1.0
+omegaconf==2.3.0
+munch==4.0.0
+tqdm
+
+diffusers==0.33.1
+huggingface-hub<1.0
+imageio==2.27.0
+opencv-python==4.9.0.80
+mediapipe==0.10.11
+av
+torch-fidelity==0.3.0
+torchmetrics==1.3.1
+lpips==0.1.4
+face-alignment==1.4.1
+insightface==0.7.3
+kornia==0.8.0
+
+# --- Numerical / Scientific ---
+numpy==1.26.2
+matplotlib==3.8.2
+numba==0.58.1
+Cython==3.0.7
+einops==0.7.0
+ninja==1.11.1.1
+
+# --- Model Repos & Tools ---
+git+https://github.com/descriptinc/audiotools.git
+git+https://github.com/OutofAi/tigersound.git
+hf-xet==1.1.8
+modelscope==1.27.0
+onnxruntime==1.21.0
+
+# --- Real-ESRGAN for Video Upscaling ---
+realesrgan>=0.3.0
+basicsr

requirements_minimal.txt

@@ -0,0 +1,13 @@
+# Minimal requirements for local testing on macOS (ARM64)
+# Excludes GPU-only packages: triton, onnxruntime-gpu, decord, flash-attn-3
+
+gradio==6.4.0
+huggingface-hub<1.0
+omegaconf==2.3.0
+tqdm
+ffmpeg-python==0.2.0
+python_speech_features==0.6
+numpy
+imageio
+opencv-python
+pillow

setup_local.sh

@@ -0,0 +1,41 @@
+#!/bin/bash
+set -e
+
+echo "🚀 Setting up OutofLipSync with uv (local development)..."
+
+# Check if uv is installed
+if ! command -v uv &> /dev/null; then
+    echo "📦 Installing uv..."
+    curl -LsSf https://astral.sh/uv/install.sh | sh
+    export PATH="$HOME/.cargo/bin:$PATH"
+fi
+
+echo "🔧 Creating virtual environment..."
+uv venv
+
+# Detect OS
+OS="$(uname -s)"
+ARCH="$(uname -m)"
+
+if [[ "$OS" == "Darwin" && "$ARCH" == "arm64" ]]; then
+    echo "⚠️  macOS ARM64 detected"
+    echo "📦 Installing minimal dependencies (GPU packages not supported on macOS ARM64)..."
+    uv pip install -r requirements_minimal.txt
+    echo ""
+    echo "⚠️  WARNING: The following packages cannot be installed on macOS ARM64:"
+    echo "   - triton (GPU compute)"
+    echo "   - onnxruntime-gpu (GPU inference)"
+    echo "   - decord (video processing)"
+    echo "   - flash-attn-3 (attention mechanism)"
+    echo ""
+    echo "   Full ML functionality is NOT available on macOS."
+    echo "   For testing, use HuggingFace Spaces (Linux GPU environment)."
+else
+    echo "📥 Installing all dependencies from requirements.txt..."
+    uv pip install -r requirements.txt
+fi
+
+echo ""
+echo "✅ Setup complete!"
+echo "🎉 Run app with: uv run python app.py"
+echo "   Or: source .venv/bin/activate && python app.py"

time_util.py

@@ -1,9 +1,11 @@
 import time
 from contextlib import contextmanager
 
+
 @contextmanager
 def timer(name: str):
     start = time.time()
     print(f"{name}...")
     yield
-    print(f"  -> {name} completed in {time.time() - start:.2f} sec")
+    print(f"  -> {name} completed in {time.time() - start:.2f} sec")
+

video_processing.py

@@ -7,6 +7,7 @@ import subprocess
 import cv2
 import numpy as np
 import torch
+from fractions import Fraction
 
 import torchvision.transforms.functional as F
 
@@ -110,7 +111,7 @@ def get_video_info(video_path: str) -> dict:
 
     width = data["streams"][0]["width"]
     height = data["streams"][0]["height"]
-    fps = eval(data["streams"][0]["r_frame_rate"])
+    fps = float(Fraction(data["streams"][0]["r_frame_rate"]))
     duration = float(data["format"]["duration"])
 
     return {"width": width, "height": height, "fps": fps, "duration": duration}
@@ -247,6 +248,7 @@ def upscale_video(video_path: str, width: int, height: int, output_dir: str) ->
             num_grow_ch=32,
             scale=4,
         )
+        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
         upsampler = RealESRGANer(
             scale=4,
             model_path=model_path,
@@ -255,7 +257,7 @@ def upscale_video(video_path: str, width: int, height: int, output_dir: str) ->
             tile_pad=10,
             pre_pad=0,
             half=True,
-            device=torch.device("cuda"),
+            device=device,
         )
 
         cap = cv2.VideoCapture(video_path)