How To Upscale Images From Stable Diffusion
Digital artists and content creators are excited about using deep-learning text-to-image platforms to create out-of-this-world images. Generating accurate images based on keywords has completely changed the game of digital art. However, some who use the text-to-image platform Stable Diffusion have a common complaint. Although the images created may be exactly what they’re looking for, image size has been a huge disappointment.
Creating the perfect piece of digital art using Stable Diffusion is fantastic, but what good is it if the image size is far too small to use? Fortunately, there are ways to upscale an image. In this article, we’ll discuss how to upscale images from Stable Diffusion.
The Image Size Issue
By default, Stable Diffusion’s default image size is 512 x 512 pixels. This native resolution is considered small in today’s digital world and presents challenges to those who need to use files created with Stable Diffusion in a much larger format. The model was trained on datasets of 512 x 512 resolution images and therefore its output is in the same format. But most social media platforms require a resolution of 1080 x 1080 for acceptable viewing, which truly highlights how important an image’s resolution is.
Stable Diffusion can create the perfect image that fits all of the designer’s desires. However, if the file size is so small as to render it unusable, it does present a major dilemma. Simply dropping the image into an app and increasing its size will result in severe image degradation, most likely worse than the original file. Images with such low resolution won’t print well and also cannot be placed into Photoshop to be edited to a designer’s satisfaction.
How to Upscale Stable Diffusion Images
Fortunately, there are ways to upscale a low-resolution image created with Stable Diffusion. Some users have been quite creative in developing techniques to accomplish this. One difficult and time-consuming method is to split up an image into smaller 512 x 512 sections and then stitch them back together. Others use custom-made algorithms to regenerate AI images into higher-resolution formats.
There are numerous online tools you can use to upscale your images. However, most of these tools ask you to create an online account by using your email address or to pay for services. If you’re looking for a free tool you can use countless times without a subscription, TinyWow is an excellent choice.
- Navigate to the TinyWow Image Upscaler.
- Upload, or Drag your file that you’d like to upscale.
- Once downloaded, click on Increase to select the image enhancement you wish.
- When done, click on Upscale.
- Voila! You can Download your images and save them on your computer or you can scan the QR code to save them on your mobile.
Use chaiNNer to Upscale
You can chaiNNer to upscale Stable Diffusion images. It’s a flowchart/node-based image processing GUI (graphical user interface) that helps to chain image processing tasks. Its strong point is upscaling images. You’ll have full control over your processing pipeline by connecting nodes. This makes it much easier to perform incredibly complex tasks by letting chaiNNer do the work for you.
It works with Windows, macOS, and Linux. If you’re new to working with GUIs, chaiNNer might seem difficult at first. Fortunately, it’s not a difficult process to use chaiNNer to upscale images. By dragging and dropping specific nodes, you can set up a flowchart of processes to do all the heavy lifting. Here’s how to start using chaiNNer:
- Using your computer, navigate to Github and download the appropriate version of chaiNNer.
- Once the installation process has been completed, launch the app.
- Load the Image File Iterator by selecting it from the left-pane menu and dragging it to the right.
- Tap in the window that reads Select a Directory and choose the image you want to upscale.
- Navigate to the appropriate folder and hit Select Folder.
It should be noted that any other images located in the directory that you select will also be processed. If you only want to upscale one image, you’ll first need to remove any others located in the same directory. However, since the upscaling process takes a lot of time, it would serve the user to have any images that require upscaling to be in the same folder so chaiNNer can upscale them at the same time. Once you have the appropriate image or images in the same directory, you can move on to the next steps.
- Click and drag Upscale Image from the far left-pan menu and drop it anywhere in the large window to the right.
- Locate the Load Image box and find the word Image. Click on your image and drag it toward the Upscale Image box and release it. You’ll now see a line connecting both boxes.
- Using the left-pane menu, click and drag Load Model and drop it into the window towards the right.
- Within the Load Model box, click and drag a line from the word Model and release it in the Upscale Image box next to the word Model.
- Go back to the Load Model box and tap on Select a file located under Pretrained Model.
- Select the appropriate model for the type of image you are using.
Next, you’ll need to check how much this model will upscale the image you’ve chosen. The upscale sizes are preset, so this is why you’ll want to ensure that it’s large enough before running the process. If you need the image larger than the preset size, you can run the process twice to double its size. Here’s how to do it:
- From the left-pane window, click on Load Image and drag it into the window on the right.
- Tap Select a file from the top of the box. Navigate to the image you want to upscale and hit Open.
- Click on the image and drag a line to the Upscale Image box and release it on the word Image.
- On the bottom of the Upscale Image box, you’ll see the output image size.
- If it isn’t big enough, you can double its size by duplicating the Upscale Image box and placing it next to the first one.
- Now click and drag a line from the original box to the new one. You do this by dragging a line from the bottom of the original box at the words Upscale image and ending it at the new box where it says Image.
- From the Load Model box, drag a line from the model to where it says Model in the duplicated box. You can check the newly increased size by looking at the bottom of this new box.
- Click and drag Save Image from the left pane menu and drag it to the right.
- Using the second Upscale Image box, drag a line from where it says Upscale Image and release it on the Save Image box next to the word Image.
- In this same box, type in a file name for the upscaled file and a destination folder location.
It could be beneficial to have a specific folder labeled “Upscaled Images” so that you can keep the original image in one folder and the upscaled one in another. Once everything is to your liking, hit the green arrow on the top of the screen to begin the process. The lines you drew will begin to animate and will continue to do so until the process has been completed.
The upscaling process is very time-consuming and can take hours to complete. Once the process has been completed, however, you can check your images by opening them from within the Upscaled Images folder or whichever folder you designated for the final image. Please note that the upscaled images will be significantly larger than their originals. It’s crucial to ensure that you have adequate disk space before beginning the upscale process.
Using chaiNNer may seem complex, but once you get comfortable with its flowchart functions, it’s actually quite simple to use. Using nodes and attaching them with chains (lines) shows exactly how the entire process will function and is great for visual learners.
Other Issues With Stable Diffusion
Other than the 512 x 512 default image size, there are other notable issues that users of Stable Diffusion report.
The rendering of faces can sometimes pose problems, especially when the desired result is photorealistic. For creators who want an anime or surrealist face, this generally isn’t a problem. However, if you’re looking for an authentic and natural look, at times Stable Diffusion can fail. This is because there is no method to have an AI-generated image focus only on the face. However, it’s possible to zoom in and re-render the face for better results.
Another issue worth noting is the proper rendering of human limbs. Again, this is only of concern when you want the desired image to be photorealistic. Sometimes limbs are rendered improperly or in unnatural positions. Users have reported images being generated with extra limbs, and sometimes extra fingers on hands.
These issues should become less frequent as the researchers at Stable Diffusion add more datasets and finetune their algorithms.
Upscale Your Stable Diffusion Images Easily With chaiNNer
Stable Diffusion is an interesting text-to-image platform. Although it produces small files with low resolution, they can be upscaled. Although it might be time-consuming, the process is necessary if you want to further edit an image or ready it for printing. Using chaiNNer, images can be made significantly larger without losing image quality.
Have you tried to upscale an image created with Stable Diffusion? Did you use chaiNNer? Let us know in the comments section below.