Flux Inference

A flexible script for loading fine-tuned versions of Black Forest Labs' Flux.1-dev model.

Supports loading LoRAs, as well as full fine-tunes from DeepSpeed and Safetensors formats.

Running

First, install the required packages:

pip install -r requirements.txt

Then, run the script:

python inference.py --model_dir=<path_to_checkpoint>

Command Line Args

Required Args:

--model_dir=<path_to_checkpoint> || The directory of the model checkpoint

Optional Args:

--num_images=<number> || The number of images to generate per caption
--output_dir || The directory to save the generated images to
--cache_dir || The directory to save rebuilt DeepSpeed models to
--prompt_file || The path of a file containing prompts to use for inference

Helpful Links

• SimpleTuner: Easy training and tuning of large diffusion models (including Flux)
• SimpleTuner Configs: My SimpleTuner configurations for full fine-tunes and LoRA tunes of Flux to get you off the ground quickly!