Launch a Text-to-Image Http Service

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 "https://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"
         }'

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!