search

thundaga / SD-webui-txt2img-script

Stable diffusion webui script to batch process images within the txt2img section
41 stars 5 forks source link

Cannot generate same image. #16

Open StellineOrb opened 11 months ago

StellineOrb commented 11 months ago

Sorry, but I am here again. Just started using this script, but it seems I cannot get the same image after the high-res fix. I examined the metadata of the 2nd pic, and found it different. Pls help to see how can I solve this. It seems the 2nd image had added new metadata with a different seed. 微信图片_20231126212538 微信图片_20231126212601

thundaga commented 11 months ago

@StellineOrb can you explain a bit of your workflow. Based on the metadata there isn't any information relating to the high-res fix options being applied. E.g. Hires upscale: 1, Hires steps: 10, Hires upscaler: Latent, etc...

Usually the user has a set of images they created quickly to get an idea of what they want and process the filtered/desired images again but the second time includes the high-res fix option. User would then select the metadata info they would like to retain from the script options. This overall allows the user to not waste a time expensive option like the hires-fix on undesired outputs.

StellineOrb commented 11 months ago

Yeah, I just used these 2 simple images to demonstrate the difference in the metadata. Even without the highres-fix process, the metadata are different.
My workflow is like this: I use dynamic prompts to input a set of prompts to t2i, so each image will have different prompts. And I will handpick some of them to do the High-res fix. This process will take a lot of time. If I can use the metadata in the PNGs and directly batch fix the images, then it will help me a lot.