Digital artists and content creators are excited about using text-to-image deep learning platforms to create incredible images. Creating accurate images based on keywords has completely changed the game of digital art. However, there are general complaints from some who use the Stable Diffusion text-to-image conversion platform. While the generated images may be exactly what they are looking for, the image size has become 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 too small to use? Luckily, there are ways to scale an image. In this article, we will discuss how to scale images from Stable Diffusion.
Image size problem
By default, the default Stable Diffusion image size is 512 x 512 pixels. This native resolution is considered small in today’s digital world and creates problems for those who need to use files created with Stable Diffusion in a much larger format. The model was trained on 512 x 512 resolution image datasets, so its output is in the same format. But most social media platforms require a 1080 x 1080 resolution for acceptable viewing, which really highlights how important image resolution is.
Stable Diffusion allows you to create the perfect image that meets all the wishes of the designer. However, if the file size is so small as to render it unusable, a serious dilemma arises. Simply dragging and dropping an image into the application and increasing its size will seriously degrade the quality of the image, most likely worse than the original file. Images with such a low resolution do not print well and also cannot be placed in Photoshop for editing as required by the designer.
How to scale images with stable diffusion
Luckily, there are ways to scale a low resolution image created with Stable Diffusion. Some users have been creative in developing methods to achieve this. One tricky and time-consuming method is to split the image into smaller 512 x 512 pieces and then put them back together. Others use special algorithms to convert AI images to higher resolution formats.
There are many online tools that can be used to scale images. However, most of these tools require you to create an online account using an email address or pay for services. If you’re looking for a free tool that can be used countless times without a subscription, TinyWow is a great choice.
- Go to TinyWow Image Upscaler .
- “Upload” or “Drag and drop the file” you want to enlarge.
- Once uploaded, click “Zoom in” to select your desired image enhancement.
- When you’re done, click “Upscale”.
- Voila! You can “upload” your images and save them on your computer or save by scanning the “QR code” to save on your mobile phone.
Use chainner to zoom in
You can zoom in on Stable Diffusion images in a chain. It is a flowchart/node-based image processing GUI (GUI) that helps to chain image processing tasks. Its strong point is image scaling. You will have full control over the processing pipeline by connecting nodes. This makes incredibly difficult tasks much easier by letting chaiNNer do the work for you.
It works with Windows, macOS and Linux. If you’re new to GUIs, chaiNNer might seem daunting at first. Luckily, using chaiNNer to scale images is easy. By dragging and dropping certain nodes, you can customize the process flowchart to do all the hard work. Here’s how to start using chainNer:
- Using your computer, go to Github and download the appropriate version of chaiNNer.
- After the installation process is complete, launch the application.
- Load the “Image File Iterator” by selecting it from the left panel menu and dragging it to the right.
- Click in the box that says “Select a catalog” and select the image you want to zoom in on.
- Navigate to the desired folder and click “Select Folder”.
It should be noted that any other images located in the directory you choose will also be processed. If you only want to zoom in on one image, you first need to delete all the others located in the same directory. However, since the scaling process takes a long time, it will be useful for the user to have all the images that require scaling in the same folder so that chaiNNer can scale them at the same time. Once you have a suitable image or images in the same directory, you can move on to the next steps.
- Click and drag “Magnified Image” from the leftmost menu and drop it anywhere in the large window on the right.
- Locate the “Upload Image” box and look for the word “Image”. Click on the word “Image” and drag it to the “High Scale Image” field and drop it on the word “Image” to the “High Scale Image” field. Now you will see a line connecting both fields.
- Using the left panel menu, click and drag “Load Model” and drag it to the window on the right.
- In the Load Model field, click and drag the line from the word Model and drop it into the Zoom Image field next to the word Model.
- Return to the Load Model box and click on Choose File located under the Pretrained Model section.
- Select the appropriate model for the image type you are using.
Next, you need to check how much this model will scale your chosen image. The scale sizes are preset, so you must make sure they are large enough before you start the process. If you need an image larger than a given size, you can run the process twice to double its size. Here’s how to do it:
- In the left pane window, click “Upload Image” and drag it to the right pane window.
- Click “Choose File” at the top of the window. Navigate to the image you want to zoom in on and click Open.
- Click on the image and drag the line into the “High Quality Image” field and drop it on the word “Image”.
- At the bottom of the High Scale Image window, you will see the size of the output image.
- If it’s not big enough, you can double its size by duplicating the “High Resolution Image Box” and placing it next to the first one.
- Now click and drag the line from the original field to the new one. You do this by dragging a line from the bottom of the original box that says “High Quality Image” and ending it in a new box that says “Image”.
- From the Load Model field, drag a line from the model to where it says Model in the duplicate field. You can check the new increased size by looking at the bottom of this new box.
- Click and drag “Save Image” from the left panel menu and drag it to the right.
- Using the second field, Zoom Image, drag a line from where it says Zoom Image and drop it into the Save Image field next to the word Image.
- In the same field, enter a file name for the scaled file and the location of the destination folder.
It would be helpful to have a specific folder labeled Zoomed Images so that you can store the original image in one folder and the zoomed image in another. Once you’re happy with everything, click the green arrow at the top of the screen to begin the process. The lines you draw will begin to animate and continue to do so until the process is complete.
The scaling process is very time consuming and can take several hours. However, after the process is complete, you can test your images by opening them from the zoomed-in images folder or from any folder you specified for the final image. Please note that enlarged images will be significantly larger than their originals. It is extremely important to make sure you have enough disk space before starting the upscaling process.
Using chaiNNer may sound complicated, but once you get comfortable with its flowchart features, it’s actually quite easy to use. Using nodes and connecting them with chains (lines) shows exactly how the whole process will function and is great for visuals.
Other problems with stable diffusion
In addition to the default image size of 512 x 512, there are other notable issues reported by Stable Diffusion users.
Face rendering can sometimes be problematic, especially when the desired result is photorealistic. For creators who need an anime or a surreal face, this is usually not a problem. However, if you’re looking for an authentic and natural look, sometimes Stable Diffusion might not work. This is because there is no way to focus an AI-generated image just on the face. However, you can zoom in and remap the face for better results.
Another problem worth paying attention to is the correct drawing of human limbs. Again, this only matters if you want the desired image to be photorealistic. Sometimes the limbs are visualized incorrectly or in an unnatural position. Users have reported that the images are created with extra limbs, and sometimes with extra fingers on the hands.
These issues should become less frequent as the Stable Diffusion researchers add more datasets and tweak their algorithms.
Easily scale stable diffusion images with chaiNNer
Stable Diffusion is an interesting framework for converting text into an image. Although it produces small, low resolution files, they can be scaled. Although this can take a long time, this process is necessary if you want to further edit the image or prepare it for printing. With the help of a chainner, images can be significantly enlarged without loss of quality.
Have you tried scaling the image created with Stable Diffusion? Have you used a channer? Let us know in the comments below.