And welcome back to my little slice of the SD Jungle.
If you are new to the series I would suggest you check out the first two parts of this series
Part 1 - Extensions, Configurations, and 1st Wave Generations (Automatic1111)
Part 2 - ADetailer, LoRa Models, Negative Prompts, and Embeddings
If you are returning; It's good to see you again!
In the last part of this series we used a bunch of tools, LoRa, and embeddings to create a nice detailed picture. In fact you may recognize the cover of this article.
This time around we will take our 512x768 image and upscale it to 1024x1536 using 'Ultimate SD Upscale' Script. Not only will we do that, but we will be adding the final bits of detail to the image in the same process.
After we have done that; I will be taking a little time to show you my process of editing my final images in GIMP. Nothing fancy, just some color and level correction to taste.
Let's get to it!
What You Will Need:
Nothing to crazy this time. Just a script, an Upscaler Model and a LyCORIS.
Ultimate SD Upscale by Coyote-A (Script)
Can be installed via the Available Tab under Extensions in the UI. Search "Ultimate"
4x-Ultrasharp by Samael1976
Rename from '4x-Ultrasharp.pt' to '4x-Ultrasharp.pth'
Put in '\webui\models\ESRGAN'
Concept: Perfect Eyes by malcolmrey
Put in '\webui\models\Lora'
Once you have both installed/placed, Restart your WebUI.
Section 1 - Upscaling:
Hopefully, you still have your finished image from the last article. If not, it's fine. Just choose an image from your gallery that you want to upscale and follow along.
The first ting we nee to do is go over to the 'img2img' tab of the WebUI and configure our Generation Parameters and extensions.
Generation Parameters:
Resize Mode: Just Resize
Sampling Method: DPM++ 3M SDE Exponential
Sampling Steps: 40
Width: 512
Height: 768
CFG Scale: 25
Denoising Strength: 0.3
Seed: Random (-1)
DynThres Configuration:
Now that those two are setup let's activate and configure Ultimate SD Upscale.
Scroll down to the bottom of the img2img Accordions and use the dropdown labeled 'Scripts' to select Ultimate SD Upscale.
Configure it as you see above.
Target Size Type: Scale from image size
Scale: 2
Upscaler: 4xUltrasharp_4xUltrasharpV10
Type: Linear
Tile Width: 512
Tile Height: 768
Mask Blur: 8
Double check all your configs to make sure you're goo to go and then let's upload our image.
Great! But we are not quite ready to upscale just yet. We need to tell the Upscaler what that we want to enhance some things, as well as add additional detail. This is where our LyCORIS comes in. We will also be using our 'Add Detail' Style LoRa as well.
The Prompt:
| (Enhance_Detail:1.2), (Enhance_Lighting:1.2), (Enhance_Lighting_Effects:1.2), (Crisp_Clear_Bold_Lines:1.2), (Highly_Intricate_Detail:1.2), (Enhance_Shadows:1.2), (Define_Layers:1.2), (Depth:1.2), (Sharpen:1.2), (Enhance_Color:1.2), (Enhance_Reflections:1.2), (Enhance_Environment1.2), |
| <lora:locon_perfecteyes_v1_from_v1_64_32:0.2>, perfecteyes eyes, |
| <lora:add_detail:0.8>, |Now that we have everything in in place; Let's hit the Generate Button (Ctrl+Enter):
And Behold! We now have a highly detailed 1024x1536 image!
Your picture will not look exactly like mine because we are not using a static seed this time around. If your not happy with the result you got you can regenerate until you get a result to your liking.
You can also adjust the Denoising Strength up (No higher than 0.5) or down depending on how much change you want to see; As well as adjusting the weights in the prompt and of the LoRa/LyCORIS Models.
Simple enough right!?
Section 2 - Post Processing with GIMP:
So this last part is completely optional but sometimes I don't get the "pop" of color and contrast I want from a generation. To correct this I use an Image Editor called GIMP.
If you don't know; GIMP is a free image editor that offers a lot of the same features you'll find in a program like Photoshop. It is widely supported and their are tons of downloadable plugins to achieve what you're trying to do. We won't be needing any of that to do what we are about to do, however.
If you would like to download GIMP you can do so here: GIMP Download Page
Adjusting Levels and Curves:
Let's navigate to our saved image and open it with GIMP.
The first thing we will want to do is adjust or levels. You can do this by clicking on the 'Colors' Option on the top ribbon of the interface and selecting 'Levels' from the dropdown.
You will be presented with the following popup box.
In most cases your color spectrum will already be wide as is the case here. You can confirm this by looking at the white and black pointers at the bottom of the 'Input Levels' section. If those pointers are not at the very edges of the spectrum move them now to reflect the image above.
The grey pointer is currently in the very center of the spectrum. We can adjust this left or right to get different results, however I usually just set to 'Input' to 0.90 to get a darker image or 1.10 to get a brighter image. There is no reason to go too crazy with this number, but it's your image so do as you please and adjust to your taste. I'm going to be setting it to 1.10 for this demonstration.
Great! Our next step will be to adjust the Curves. Click the 'Colors' on the top Ribbon, once again, and select 'Curves' from the drop down.
You will be presented with the following popup box.
Here we can curve the line to adjust the contrast and saturation. I usually don't do too crazy with this as even the smallest adjustment makes a big impact.
You should also tick the 'Split view' option so you can see the changes in real time.
This is my adjustment:
Our resulting image is:
As you can see, the colors really "pop" now and the contrast is high which helps define some of the detail we fought so hard to create. Again, this is my preference and you can adjust these settings to whatever you like to suit your own personal taste.
Now that we've adjusted the levels and curves we have one more step before we save our image. That step is Cropping. With this particular image there is not much I want to crop, if anything, but I will demonstrate the process for the purposes of this Series.
Cropping:
Click 'Tools' on the Top Ribbon, Select 'Transform Tools', and click 'Crop'.
Now, you will notice that the Tools Area on the Left side of the screen has changed to the following:
Tick the 'Fixed' checkbox and make sure 'Aspect Ratio' is selected from the dropdown.
Since we know that out image size is 1024x1536, we will enter that into the field below the checkbox in the following format: 1024:1536.
Now let's select the area we want to crop on our image. You do this by simply clicking and dragging on the image itself. Since I want to preserve the house and the stairway in the image i will select and area that will do just that.
Cool. Now click on one of the 4 boxes in the corner of the selection and the image will be dropped to the selected area.
This is our new image. But wait... Since we cropped the image, the resolution has changed. The image is now 820x1230. That simply won't do. Let's Scale the image to get as close as we can to our original resolution of 1024X1536.
Scaling:
Select 'Image' from the top ribbon and click 'Scale Image'.
You will be presented with the following popup:
The 'Image Size' is in locked mode. This means when you change either the Width, the height will change according to the original aspect ratio. This is ideal since we can't guarantee that we are going to achieve a perfect 1024x1536 scale.
Let's change the Width field from 820 to 1024 and see what our locked Height will be.
We got lucky this time! Often the height will be a pixel or two of the target 1536. In this case we are spot on with or aspect ratio.
Let's click the 'Scale' Button and scale our image and boom! The image is now at a resolution of 1024x1536.
Now that we have that finished, it is time to save the image so we can post it here on CivitAI and/or other image hosting sites.
( NOTE: When going through this process you are stripping the generation data from the image. This means that when you upload it to CivitAI, the website will not detect any data for Prompt, Negative Prompt, or Resources Used. If these things are important for you to have in conjunction with the image than I suggest you either setup Automatic1111 to save a log file for saved images, or create a .txt file yourself that you put the generation info into manually. )
Exporting Our Image:
Select 'File' from the top ribbon and select 'Export As'.
You will be presented with the following popup box:
Navigate to your desired directory to save the image. Once there, in the top dialog box, enter the desired save name for the image and append the desired file type. (i.e.: Example.png, Example.jpeg, Example.jpg).
Click 'Export' and you will be presented with one more popup box:
You can change things here if you know what you are doing, but I recommend keeping all these options As-Is.
Click 'Export' and Bam! Your image is now saved.
If you wish to maintain your original copy of the image simple close the image in the GIMP Editor and Select 'Discard Changes'.
And that's it!
Conclusion:
Well this has been a ride!
We have generated an image; Tweaked that Image, Upscaled that Image, and finally Processed that Image using GIMP. All of this done using my Workflow technique.
I truly hope you found this series informative. If nothing else I hope you got your hands on some cool extensions, LoRa Models, and Checkpoints.
Thanks for joining me here. I really appreciate the read and interest in my process. Leave a comment below if you have any questions! You can also chat me here on CivitAI if you'd like.
Thanks again!