How to batch sending PNG Info to txt2img

[Z-nonymous]

What I'm trying to do:

I have a series of images that I generated (let's call those source images). When I download a new model, I want to generate new images with same parameters as those source images.

So what I do is I put the "source" image in the PNG Info tab, and then use Send to txt2img, then I click generate.

Since I have many images this is taking too long to do manually. I've tried a few things to automate:

1. I've tried to manually recreate a "prompts file" for the "prompt from file or textbox" script, but it's long as well and error-prone. Also it seems the hires settings can't be "prompted" with that script
2. Looked into adding a "Send to promptfile" in PNG Info tab. But same as above, hires settings won't work.

So at that point, I thought that the best way would be to create a txt2img script that uses a file (or textbox) with the list of the source images' path. And then for each file get the info, and generate the image.

So I started a script for txt2img, using prompts_from_file.py as a starting point, I'm able to setup the UI properly, I'm able to get the textinfo with image_data() to retrieve the generation parameters for each file.
But then I'm stuck at how to use that to add jobs to generate the images.
Should I (for each image) use register_paste_params_button from generation_parameters_copypaste.py to update UI and then append a job ?
Or should I be doing something else ?

[Z-nonymous]

I made some progress, I think I'm almost there.

I just need to map parameters properly.

I get the source image parameters using parse_generation_parameters. But I suspect I need to convert the keys to have the proper labels.
i.e. in my case I get this:

{'Prompt': 'myprompt', 'Negative prompt': '', 'Steps': '20', 'Sampler': 'Euler a', 'CFG scale': '9.5', 'Seed': '189775263', 'Size-1': '640', 'Size-2': '640', 'Model hash': '0f7f264114', 'Clip skip': '1', 'Hires resize-1': 0, 'Hires resize-2': 0}

and I suspect I need to send

{'prompt': 'myprompt', 'negative_prompt': '', 'steps': '20', 'sampler_name': 'Euler a', 'cfg_scale': '9.5', 'seed': '189775263', 'width': '640', 'height': '640', 'model_hash': '0f7f264114', 'CLIP_stop_at_last_layers': '1', 'hr_resize_x': 0, 'hr_resize_y': 0}

I can easily write the code to do that, but I'm wondering if there's already the function to do that, or the mapping. I've only found partial mapping info.
I know it may not exist because the infotext labels depends on the tool that wrote them, but in my case all my source images were created with this webui, and I believe I don't have any images with the old hires syntax.

[Z-nonymous]

Ok, it was actually faster to write my own mapping conversion than looking up if one exists.

So I was able to create a custom script that:

• Will take a file input or textbox (like with prompt_from_file) containing the list of paths to the images you want to re-create.
• will regenerate the files with the current model using the same parameters (including hires if it was used)

I've also added a checkbox to reuse same model as well, but I need to test if that actually works.

I didn't implement face restoration as it seems stable-diffusion_webui doesn't store that information in the image.

Maybe I'll share my script after testing the checkbox feature if some are interested.

[johneliot1978]

any way you could share what you have working so far? pretty plz? :D

[Z-nonymous]

I shared my repo here: https://github.com/Z-nonymous/sd_webui_batchscripts

[johneliot1978]

thanks a bunch :) much appreciated

[johneliot1978]

@Z-nonymous I've installed to extensions, but cant seem to find it in gui? any ideas?

[Z-nonymous]

@johneliot1978 It's a script to it goes to script directory.