Categories
Tutorials

Stable Diffusion: Saving Your Settings in AUTOMATIC1111

I wrote a post about how to save your default settings in Stable Diffusion and posted it to Reddit.

The post gave me an encouraging boost and a few comments stood out because they shared a new process on how you can save your Stable Diffusion settings by using saved information in the generated PNGs.

In particular, this reply from Jaceholt:

What I found to be the easiest is to have Template-Images in a folder. A pic generated with the correct settings/prompts. When you start Auto1111, click “PNG info”, drag the image into the slot, then “send to Text2Text”. (Edit: Text2Img ofc :D)

This takes about 2 seconds and now you loaded all positive/negative prompts you wanted + all the image settings. The only thing this does not load is model, clipskip, VaE, Controlnet image+settings.

The benefit of this is that you can have as many different “defaults” as you like, just give the Template Images names that make sense for the settings. Example “512x768x2KarrasxRealistic” or whatever works for you.

As a bonus tip: For “default” prompts, the easiest is to use the Style selector under the “Generate” button. Write in your standard prompts (both Pos and Neg) and save them. That way you can easily switch from a Photorealistic prompt–> cartoon->water color etc all with the click of 2 buttons.

Additional thought: Specifically create these Template Images (that shows gibberish or nothing but has a pure prompt) instead of using previous “complete” images. Early on I just loaded in an image i did in the past, then modded the prompt. Lets say I had a picture of an angel sitting on a cloud, wearing a cowboy hat. But I now wanted a demon, sitting on a rock wearing a crown.

What I did was that I just used the angel picture and started to mod the prompt. But it’s so easy to forget to remove something when you do it this way and then you spend 10 minutes confused to why your demon is sitting on a cowboy hat.

Since the template is gibberish anyway you can even write in something like

((beautiful)), ((masterpiece)), DESCRIBE CHARACTER HERE, Raytracing, Intricate Details, DESCRIBE SCENE, DAY or NIGHT? (Photorealistic:1.3), Detailed face, diffused pores, etc etc.

I found that the BIG LETTERS make it easy for me to structure the image well, and also helps my a bit scrambled brain to do things in order and not forget things.

It’s a wall of text. Let’s break it down into steps.

Breaking Down the Steps to Saving Stable Diffusion Settings

Jaceholt proposes that you create a folder. So let’s create one.

creating a stable diffusion settings folder

I created a folder called “Default Settings”, and I will put a few generated images in here that have specific settings.

Generating “Blanks” to Save Stable Diffusion Settings

Jaceholt’s method suggests you put in pics generated with various settings and prompts.

But you might find it more efficient to generate “blanks” that specifically contain all the settings you’d want to retrieve in the future.

Open up Automatic1111 and let’s generate the first blank –one that gives you a portrait image with the maximum dimensions your graphics card can produce.

generating stable diffusion save settings file

For me, it’s about 1100 by 768px. Change the dimensions but leave everything else untouched.

I renamed this to “HMAX-1100x768px.png” and saved it into my Default Settings folder.

Here’s the image. Clearly, even if you don’t prompt, you’ll get something.

using png info to retrieve saved settings in stable diffusion

Loading Your Settings by Using PNG Info and Your Blank

using png info to retrieve saved settings in stable diffusion

In order to test whether it worked, let’s go to the “PNG info” tab.

Drag your blank into the box. If all went well, you should see your settings on the right, as below:

reading png info in stable diffusion automatic1111

Send it to txt2img.

An Even Faster Method to Load PNG Info Settings

Thanks to Potential-Back6972, who said that you can actually just drag the blank into the prompt area, and then press the blue button to read the settings.

reading saved settings from png info in stable diffusion

So, drag your image from your file explorer into the prompt (red arrow), and then press the button enclosed in the pink box.

Reset Seed and Model Hash

stable diffusion model hash and seed

Don’t wanna hamper Stable Diffusion’s creativity, so delete all of that.

Set seed to -1 for a random seed. Click the X to delete the model hash.

Use Case: Maximum High-Res Settings

I have a RTX2070 and it’s pretty limited.

But if you have seen my previous articles, I’m keen on seeing how I can generate large images for print media.

I don’t know how large of an image I can generate.

I’d have to inch upwards in dimensions until my graphics card calls it quits.

I managed to get it to 696 x 768px before it won’t even start. Here’s the blank.

In the past, I’d have to write it down somewhere or do the trial-and-error again. But with the blank, I no longer have to guess.

This really helps me when I do my high-res workflow that produces print-sized images.

Basically, I take tell txt2img to generate a 696 x 768px image with a 2x Latent high-res fix. Then I send it to the upscaler for another 2x, producing a print-media worthy 3100px x 2650 px image.

Here it is in action:

stable diffusion rainbow woman high res fix

And the detail at 100%. Click on the image to see it in 100%.

stable diffusion high res upscale detail

Don’t Save Prompts in Blanks; Use Styles Instead

To me, blanks are really meant for setting the sections beyond the prompt.

The styles section will allow you to create compound prompts.

stable diffusion styles

You can create them much more efficiently by going to your Stable Diffusion Automatic1111 folder. Mine’s C:\stable-diffusion\stable-diffusion-webui

There, you’ll find a styles.csv file. Open it and write your prompts.

Remember that these styles can stack. Meaning, you can load two prompts in the Styles section in Automatic1111.

For example, if you can save one for your prompt and another for negative prompt. Then load both.

Jaceholt further recommends that you write IN CAPITALS the stuff that will change. For example:

((beautiful)), ((masterpiece)), DESCRIBE CHARACTER HERE, Raytracing, Intricate Details, DESCRIBE SCENE, DAY or NIGHT? (Photorealistic:1.3), Detailed face, diffused pores

Are There Extensions to Do This?

Yes.

But I haven’t tested them.

Is There Any Way to Save Defaults?

One thing I realized is that this method produces a result that’s different from my Stable Diffusion saving defaults tutorial.

In that tutorial, you set the defaults once and then the same settings will load every time.

My preferred method has the defaults set up just so I can push out images quickly using a reliable, failsafe set of settings.

The blanks method is meant for times when you need a specific image. For example:

  • Max dimension images
  • A specific workflow such as high-res fix

Whereas, my defaults are meant for me to figure out whether my prompt is working, and what CFG Scale I should use.

Therefore, my defaults are meant for speed, and my blanks are configured for quality once I figure out the prompt.

So, to be more efficient, set your defaults now!

Have a project in mind?

Websites. Graphics. SEO-oriented content.

I can get your next project off the ground.

See how I have helped my clients.