Launch an HTTP-based text-to-image service that generates images from textual descriptions (prompts) using the DiT model.
The generated images can either be returned directly to users or saved to a specified disk location.
For example, the following command launches a HTTP service with 4 GPUs, 2 Ulysses parallel degree, 2 PipeFusion parallel degree, and the model path is ./models/FLUX.1-schnell
.
python ./entrypoints/launch.py --world_size 4 --ulysses_parallel_degree 2 --pipefusion_parallel_degree 2 --model_path /your_model_path/FLUX.1-schnell
To an example HTTP request is shown below. The save_disk_path
parameter is optional - if not set, the image will be returned directly; if set, the generated image will be saved to the specified directory on disk.
curl -X POST "http://localhost:6000/generate" \
-H "Content-Type: application/json" \
-d '{
"prompt": "a cute rabbit",
"num_inference_steps": 50,
"seed": 42,
"cfg": 7.5,
"save_disk_path": "/tmp"
}'