StreamDiffusion-NDI

NDI & OSC extension for real-time Stable Diffusion interactive generation with StreamDiffusion.

Features:

• NDI (video) streaming over the network or localhost.
• OSC communication for prompt and FPS.

Installation:

Supported version Python 3.10

1. Install StreamDiffusion with TensorRT.
2. Install SteamDiffusion-NDI in StreamDiffusion environment with pip install -r requirements.txt

Usage:

1. Accelerate model with an example script (temporal solution). Change acceleration to tensorrt.
2. Configure config.json
3. Run in StreamDiffusion environment python sd_ndi.py
4. Add NDI output to send images in Stable Diffusion and NDI input (SD-NDI) to receive processed images
5. Send string with OSC at /prompt address change the prompt during the inference
6. You can get inference FPS at /fps address on client side

Config:

Look in config.json for an example configuration.

"sd_model": "path to diffusers model",
"t_index_list": [Number of inference steps],
"engine": "path to the folder with the accelerated model",
"min_batch_size": depends on your configuration,
"max_batch_size": depends on your configuration,
"ndi_name": "NDI client name to recieve from",
"osc_out_adress": client address for receiving FPS value,
"osc_out_port": client port for receiving FPS value,
"osc_in_adress": server address for receiving commands.,
"osc_in_port": server port for receiving commands.