Creating content can be difficult. Crafting the right words and pairing them with relevant images can take a lot of time. Fortunately, with the developments in both text-based AI and image AI, you can now create content faster than ever before. You can also create content that is truly unique to your brand.
You don't need to use a stock image available on a website that thousands of others are using. You can create your own image using technology and we're going to show you how to do this in seconds.
This all becomes possible with Image 2 Image generations. Image 2 Image involves taking an original image and providing a text-based prompt along with it to get something truly unique but based on that original image. You can further tweak the settings by modifying the init_strength, cfg_scale, and more. Each of these will have an impact on what the output image looks like. By using this method, websites like Unsplash and Envato Elements turn into goldmines of high-quality images that you can use as a foundation for creating something truly unique.
Taking this specific portrait image from Unsplash, we can feed it into the latest in image AI technology and get something unique out. Riku has all of these settings built in and makes it super simple to run. We simply upload the image, change the settings as we like for our needs, and hit the generate button. Here are the initial test settings that we have inputted to get something unique out of this image.
You can see here, that we are wanting a red shirt and a large beard. Our init_strength setting is a little high at 0.4 which may mean we do not get much change but that is ok, it is a process we can tweak and get better at understanding over time. Let's generate some images and see what we get out.
Taking down the init_strength means that the AI is going to take less direction from the image provided and mix it up more with the other settings including the prompt. This should help in getting variations from the original image which are different enough to be truly unique. Let's see how 0.2 affects things!
One of the best upscalers available for use is the real-ESRGAN upscaler which is available through Replicate or other services. Taking the 4 images above and upscaling them by 2x with the face fixing setting, we get the following outputs which we could then use for our content needs.
So what have we learned from this exercise? Can you use Image AI to take a stock image and tweak it slightly to get something truly unique? Yes, you can. I really like the output with the red collar and could use that in future content. Using that image in my own content would be truly unique and I could be confident that nobody else on the whole entire internet is using this image, that's pretty awesome!
It is of course also prudent to look at some of the limitations with Image AI in the current form, we really wanted a big beard and a red shirt and could not get both of these visualizations in the images to change as much as we'd want to. This would lead us to the belief that really if you are looking for an image of a person with blonde hair, it would make sense to start with an image of someone with blonde hair, likewise, if you want a specific fashion style, starting close to that fashion style would reap the best results.
Should stock image websites start to worry about new businesses coming along and creating unique stock images for users putting them out of business? I don't think we're at that point quite yet but if they don't adapt to this new future we're finding ourselves in, then it could be a possibility in the future!
If you have found this article helpful and are interested in building with all of the best large language models in a single place consider signing up for Riku today. We offer the best in both text and image-based AI in an easy-to-use and no-code environment.
