stable diffusion project tools , focused on batch processing image generation variants .
a collection of tools for large scale asset synthesis . each "project" is a folder with a source image ( target img2img ) , a .project file ( settings overrides ) , and a 'selected' subfolder containing synthesized images . the python tools included help to generate such a structure and to then batch process multiple projects .
example for texture synthesis = a low resolution source images of a stone brick pattern , the 'selected' folder contains multiple images of different stone textures synthesized using stable diffusion containing the prompt/negative in their metadata . gen_batch_prompts_in_projects.py gathers the prompt/negative pairs from 'selected' and randomly combines into a new list of variants , a local and project-wide .bat is generated that will use the auto1111 webui api to run img2img with controlnet for all the prompt variants , across multiple checkpoints , saving the variants into each projects 'output' subfolder .
requires auto1111 webui with --api for image synthesis
- install python 3.10
- pip install -r requirements.txt
- gen_batch_prompts_in_projects.py = searchs folders recursively for .project files , for each project gathers the metadata image prompts and generates the local and project-wide batch files for image generation , prompt variants .
- gen_project_files_from_images.py = given a folder of images create named folders and project files for each .
- sd_batch_image_gen_auto1111_webui.py = a basic controller of stable diffusion auto1111 webui api , includes specific arguments to cycle through each prompt/negative pair , each target img2img , each checkpoint .
- image_text_prompt_tools.py = drag drop image to view prompt and negative . copy prompts to work area to merge multiple prompts together without duplicates . options to simplify strengths ( removing all parathesis ) , and scale lora strengths relative to a target maximum .
- lora_variants.py = input a prompt with loras and generate all permutations of strengths , within a range of total lora strength and per lora strengths . the output can then be used for x/y plot or as wildcard .
- image_review_and_rank.py = image viewer from a folderpath , quickly rank images into subfolders using 1,2, or 3 . navigate with arrows . view as tiled texture with T