Generative AI for images is rapidly generating more demand for Viesus

Summary

• AI-generated images have become very popular and are driving an increased demand for higher resolution and quality of the output.
• Viesus AI Upscaling solves the issue of low-resolution AI-generated images, and also improves the quality.
• It is easy to try out and to implement the AI Upscaling in your process. For trying out download the app DiffusionBee and upload the AI-generated image on Viesus Cloud for AI Upscaling.

Introduction of Generative AI for images

Many companies are using AI for image creation that can generate a realistic, detailed photo from scratch in minutes. This includes applications such as DALL-E and Stable Diffusion that have been around for the past few months. This blogpost explains why Viesus is so valuable in this process.

What types of Generative AI for images are there at the moment?

• Text to Image - You just need to write what you want to have in the image, what style you like (such as cinematic), etc.
• Image to Image - User uploads an image and provides a prompt to create an image based on the uploaded image.
• Outpainting - A painting continues to be painted outside the frame.

How does Generative AI work?

To demonstrate let's take our hero image, the Matterhorn during sunrise, from the landing page as an example. We use DiffusionBee for Mac which implemented the Stable Diffusion model. In this example, we want the Matterhorn sunrise image to be painted in Van Gogh style. Therefore, I enter the prompt with what we want to generate and I selected the original Matterhorn sunrise image. You can fine-tune options, but for now I'll start with the default option. The generation is ready in a minute!

What is the issue with the generated AI image?

The current limitations of these generative AI models for images is the low resolution. That means the generated image soon looks blurry. The resolution is 512x320 pixels at the moment. It can be enlarged to 1024x640 but that does not improve the quality. It probably looks fine on your mobile phone, but displaying it larger could be problematic. For example, if we want this generated AI image to be a printed canvas on the wall.

What is the solution for the lowres generated AI image?

One thing you can do to increase the quality of your images is to use AI Upscaling. Viesus has created a feature that combines these types of upscaling algorithms together, like Super Resolution. You won't have to compromise on your images with any blurring - these algorithms are designed for high resolution and true quality. In this example, for convenience, we are going to enlarge our image by 4x from 512 × 320px to 2048 × 1280px.

Conclusion

The comparison between the original and upscaled versions demonstrates that even though it might not be immediately obvious on screen, there are a lot of nuanced differences in the look of the two renderings. However, in the field that we operate, this difference is huge. When a person wants to hang this "Van Gogh Matterhorn" as a large print on the wall of their home, high quality is required to avoid disappointment.

That's why our customers say AI Upscaling is an absolute must-have for AI-generated images and more companies want to use it.