This technique, while recently available in Stable Diffusion (an open-source AI platform), has never been accessible to the average user in an easy to use interface. That’s where Magnific steps in, and perhaps with others on the horizon.
Magnific started as a platform renowned for enhancing image details. It excelled in upscaling AI-generated images that lacked detail, making them more realistic by creating missing detail, and it is also suitable for prints and improving blurry photos. Recently, Magnific introduced a feature that transfers art styles from reference images to your own, completely transforming them. They are now expanding their feature set to include relighting images (ones you've taken or an AI image). It looks like they aren't slowing down!
Here’s an example from Relight’s beta user Julie Wieland:
Currently, the main way of adjusting lighting in a photo is primarily via dodging and burning. This definitely requires significant skill, time, and precision and isn’t meant to entirely relight an image. Now Imagine capturing a portrait indoors and later transforming it to look like it was taken in the golden hour, or enhancing a landscape shot taken on a cloudy day by adding dramatic sunlight filtering through the clouds.
I was made aware of the ability to relight images thanks to Russel Klimas and his expertise in Stable Diffusion. Naturally, I was shocked at the possibility and knew it would marry an easy-to-use interface soon. Today I saw Relight on my feed and was a little surprised how fast that happened. It almost seems like photographers can now revisit their archives with a new perspective, transforming older work with different lighting techniques. This could also mean fewer constraints and more freedom in post, right?
As Magnific gears up to release the Relight feature to the public, it’s going to be fascinating to test its capabilities and see its limitations. At first glance, it appears that Relight may alter the features of subjects to some extent. Moreover, the input images we’ve seen so far seem to be AI-generated, so the output is as well. We haven’t seen the output when real photographs are used as the source yet. How much will it alter the original image?
If you’re working in fashion or genres where the subject’s exact appearance isn’t critical, this might not be an issue. However, for projects where maintaining precise and consistent subject features is essential, the Relight feature might not yet meet your needs.
A notable concern is the file output options. If you’re hoping to save a high-resolution TIFF file, you might find yourself limited to JPG or PNG formats, which don’t offer the same depth of information as the original source files. Additionally, the consistency in color space and bit depth isn’t guaranteed. For instance, those needing a 16-bit AdobeRGB TIFF file might be disappointed.
Another consideration was nuanced adjustments. So far it doesn’t look like you can simply tweak your existing lighting setup in a subtle way in Relight. Let’s say you wanted to move the light source a little more behind the subject or move individual lights in various positions, you can’t do that yet. The output is based on the input reference. So you'd have to be reliant on that for now. However, it's possible outside of relight using other workflows in Stable Diffusion, and Relight may add that with an update when it's launched or in the near future.
For another example of what it can do, here is an example from beta tester Lyson Ober. The middle column is the reference image with the column on the end being the result.
