AI Upscale and Generative Fills
Along with the development of a plethora of text-to-image applications, AI is also being deployed in the image editing space. One of my favourite things is the remove background feature available from sites like Photoroom. I actually think their free tool is better than the Adobe Photoshop equivalent. Gone are the days of painstakingly outlining an object to remove the background. It is such a relief!
Other features that are quite handy are AI tools that allow you to upscale images without seeing jpeg artifacts. PixelBin have a great upscale option with three daily free credits for three upscales. It does seem to vectorise the images - so they look a little like images that have been traced in Adobe Illustrator, but there is a definite improvement on the overall look of lower resolution images.
Screenshot of PixelBin’s upscale.media output comparison.
Upscaling was useful for us to increase the resolution of our AI generated images for printing (so from 72 - 96dpi - to 150 - 300dpi).
Another really useful new AI feature is the generative expand feature. A number of companies are offering expand features including Runway and Extend Image that uses DALL-E and Stable Diffusion. Photoshop has also included a generative expand feature as part of their new suite of software releases. This is available as a ‘Generative Fill’ option within the crop tool function.
You can leave the input field blank if you would like the ‘Generative Fill’ option to rely on the content of the image. However, you can also include a prompt such as ‘flowers’ that triggers the inclusion of additional visual elements (as seen above). It also gives three options which can be handy to identify the preferable composition or hone your prompt based on the outputs. This is great for extending poorly cropped images or altering the image size such as going from square to rectangular image format.
With this in mind, it stands to reason that AI tools will become some of the stock features in many advertising and design spaces. It is worth mentioning that there is still a level of skill involved in crafting the image prompt and making visual decisions (along with generating the AI training content in the first place). So yes, we will still need humans…
…although we may be expected to be even more productive and generate more content more quickly. Would this be a good move forward?