Image Generator
This project consists of a free highly customizable image generator.
Usage
Structure
This project is structured to to run in declarative paradigm by making use of the files in the configs folder.
configs folder:
- config.json - serves as the run configs source.
- prompt.txt - serves as the run prompt source.
configs.json example
"model_id": "black-forest-labs/FLUX.1-schnell", - huggingface model directory
"lora": true, - whether to enable lora or not
"enable_tiling": true, - whether to enable tiling
"enable_slicing": true, - whether to enable slicing
"enable_sequential_cpu_offload": true, - whether to enable sequential cpu offload
"safety_checker": false, - whether to enable censorhip
"enable_attention_slicing": true, - whether to enable attention slicing
"num_inference_steps":1, - num of inference steps per image
"guidance_scale":8, - the guidance scale for the image generation
"gen_type":"target", - type of seed iteration, if interval, an iteration will be performed for the provided range, if not interval will iterate through the list of seeds provided
"seeds":[1] - list of seeds to iterate according to the preceding parameter. must always be a list
Output
The output will consist in the following structure:
- Output folder:
- images_i - all images here generated have the same hyperparameters apart from num of inference steps and guidance scale
- (guidance_scale)_(num_inference_steps) - folder with the hyperparameters in the name fixed and its content will be the images iterated through the provided seeds list
-image_seed_i - image corresponding to the seed i.
- generation_configs.json - json of configs to which all the images where generated apart from guidance scale and num of inference steps
Usage
The usage consists in altering the configs.json and prompt.txt and then running the ImageGenerator class as per the test notebook
