Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

20 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

HueFlow 🎨🤖

License: MIT [Zentalic Labs](

The Zentalic AI Photo Color Grader is a high-performance tool that generates custom .cube LUTs by utilizing AI to analyze the color profile of a reference image and apply that look to other footage. By automating the creation of 3D LUTs, the tool streamlines post-production for software like DaVinci Resolve, Adobe Premiere, and Final Cut Pro.

Developed by Zentalic Labs, this project is specifically engineered to run advanced Vision LLMs on consumer-grade hardware using a unique Layer-by-Layer Inference architecture via AirLLM.


✨ Key Features

  • Layer-by-Layer Inference (AirLLM): Run 11B+ parameter models on just 6GB of RAM by swapping transformer layers from SSD to memory in real-time.
  • Hybrid Model Support: Toggle between local execution (AirLLM) and Cloud APIs (Anthropic Claude, OpenAI GPT-4o, or Google Gemini).
  • Professional .cube Export: Generates industry-standard 33x33x33 3D LUTs for professional color grading workflows.
  • Zentalic "Style ID": Every AI-generated grade is assigned a unique sharing number (e.g., Z-1024) for instant style replication.
  • Dark Immersive UI: A minimalist, high-end interface inspired by the aesthetics of Claude.ai and Cursor.

🏗️ The Tech Stack

Component Technology
AI Backend Python 3.11+, AirLLM, PyTorch
Vision Models Moondream2 (Local), Llama-3.2-Vision (AirLLM)
APIs Anthropic (Claude 3.5), OpenAI, Google Gemini
Image Engine OpenCV, NumPy, Pillow
Frontend React / Next.js (Web), PyQt6 (Desktop EXE)
Compiler Nuitka (Standalone Executable)

🔄 The "Layer-by-Layer" Workflow

To support low-RAM environments (6GB), this project utilizes the AirLLM architecture. Instead of loading the entire model into VRAM, it follows this automated cycle:

  1. Sharding: The model is split into atomic transformer layers on the disk.
  2. Sequential Loading: Layer 1 is loaded into RAM → Processed → Deleted.
  3. Cyclic Processing: This repeats for all layers (1 to 32) until a color analysis is produced.
  4. Parameter Extraction: The AI outputs a JSON object containing Gain, Gamma, Lift, and Saturation values.
  5. LUT Generation: Python calculates the 3D cube map and saves the .cube file.

🚀 Getting Started

Prerequisites

  • Python 3.10+
  • 6GB RAM (Minimum)
  • SSD (Highly recommended for AirLLM swapping speed)

Local Development

# Clone the repository
git clone [https://github.com/zentalic/ai-photo-color-grader.git](https://github.com/zentalic/ai-photo-color-grader.git)
cd ai-photo-color-grader

# Install dependencies
pip install airllm opencv-python pillow anthropic openai

# Run the application
python app.py

About

HeuFlow is a offline ,low end ,photo color grading AI created by Adarsh A

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages