Compare Diffusion

This simple tool is built on top of Hugging Face's Diffusers library and helps you visually compare the results of different image diffusion parameters (and models). It performs inference based on the parameters you provide and then displays the results in a grid, outputted as a PDF.

Usage

Compare Diffusion takes the following arguments:

• hf_token (required): Your Hugging Face token.
• output_path (required): The path where you would like to save the output PDF.
• type (required): The type of inference you would like to perform. You can choose from img2img, txt2img, or inpaint.
• rows (required): The dimension you would like to use for the rows of the image matrix. You can choose from model, image, cfg_scale, denoising_strength, prompt, negative_prompt, seed.
• cols (required): The dimension you would like to use for the columns of the image matrix. You can choose from the same options as for rows.
• model (required): The selected model(s) you would like to use. You can provide a path to a CKPT file, or a Hugging Face model name.
• cfg_scale (required): The selected cfg_scale(s) you would like to use.
• denoising_strength (optional): The selected denoising_strength(s) you would like to use. Defaults to [0.0] if not provided.
• prompt (required): The selected prompt(s) you would like to use.
• negative_prompt (optional): The selected negative_prompt(s) you would like to use. Defaults to [''] if not provided.
• seed (optional): The selected seed(s) you would like to use. Defaults to [1] if not provided.
• height (optional): The height of the input/output images. Defaults to 512.
• width (optional): The width of the input/output images. Defaults to 512.
• inpaint_full_res (optional): Whether to inpaint at full resolution. Defaults to False.
• inpaint_full_res_padding (optional): Padding for inpainting at full resolution. Defaults to 35.

An example of how you might use the Compare Diffusion tool:

python compare_diffusion.py * hf_token <your_hugging_face_token> * output_path output.pdf * type img2img * rows prompt * cols model * model <model_1> <model_2> * cfg_scale <cfg_scale_1> <cfg_scale_2> * prompt <prompt_1> <prompt_2>

Additional Notes

• If you choose txt2img as the inference type, the denoising_strength argument will be automatically set to [0.0].
• If you choose img2img or inpaint, you will need to provide input images in the input/images directory. Each image must have the prefix i_ (for example, i_1.png, i_2.png etc.).
• If you choose inpaint, you will need to provide a mask in the input/masks directory. The mask should be a black and white image where the white pixels represent the area you would like to inpaint. Each mask must have the prefix m_ (for example, m_1.png, m_2.png etc.). The mask should be the same size as the corresponding input image.