https://github.com/dogbertdev/faststablediffusionxl
https://github.com/dogbertdev/faststablediffusionxl
Last synced: 16 days ago
JSON representation
- Host: GitHub
- URL: https://github.com/dogbertdev/faststablediffusionxl
- Owner: dogbertdev
- Created: 2024-08-28T12:32:40.000Z (over 1 year ago)
- Default Branch: master
- Last Pushed: 2024-08-28T22:27:42.000Z (over 1 year ago)
- Last Synced: 2024-08-28T23:36:03.328Z (over 1 year ago)
- Language: Python
- Size: 23.4 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# TGATE SDXL Image Generation
This project implements an image generation pipeline using the Stable Diffusion XL model with TGATE (Text-Guided Attention for Efficient Text-to-Image Generation) and TCD (Temporal Coherence Diffusion) scheduling.
## Features
- Uses Stable Diffusion XL as the base model
- Implements TGATE for efficient text-to-image generation
- Utilizes TCD scheduling for improved temporal coherence
- Supports LoRA (Low-Rank Adaptation) for fine-tuning
- Configurable image resolution, prompts, and generation parameters
## Requirements
- Python 3.x
- PyTorch
- diffusers
- tgate (custom implementation)
- CUDA-capable GPU (for optimal performance)
## Setup
1. Clone this repository
2. Install the required dependencies:
```
pip install torch diffusers
```
3. Place the following files in the same directory as the script:
- `aniversePonyXL_v10.safetensors` (base model)
- `TCD-SDXL-LoRA.safetensors` (LoRA weights)
## Usage
1. Adjust the prompts, negative prompts, and generation parameters in the script as needed.
2. Run the script:
```
python main.py
```
3. The generated image will be saved as `image.png` in the same directory.
## Configuration
You can modify the following parameters in the script:
- `prompt`: The main text prompt for image generation
- `prompt_2`: Additional text prompt (combined with the main prompt)
- `negative_prompt`: Text prompt for features to avoid in the generated image
- `num_inference_steps`: Number of denoising steps
- `guidance_scale` and `guidance_scale_2`: Guidance scales for the prompts
- `eta`: Eta value for DDIM sampling
- `seed`: Random seed for reproducibility
- `width` and `height`: Output image dimensions
## Advanced Features
- TGATE implementation with configurable gate step, intervals, and warm-up
- TCD Scheduler for improved temporal coherence
- LoRA integration for fine-tuned results
## Notes
- The script currently uses CUDA for GPU acceleration. Ensure you have a compatible GPU and CUDA setup.
- Uncomment the upscaling code if you want to use the 4x upscaling feature (requires additional setup).
## Acknowledgements
This project uses components from various open-source projects, including Stable Diffusion XL, diffusers, and custom implementations of TGATE and TCD.