Photo-realistic composition, color and tonal correction, believable retouching…and a bit of generative AI.
Commercially available generative AI for image editing has made radical advances over the past two years, particularly with the release of Adobe Photoshop v. 25, but it still has limitations as a retouching technology. It works well for creating material (or faces) from ‘scratch’, but doesn’t yet integrate effectively with existing imagery (or faces). The fascinating thing about working with generative AI is the role of language and description in the new discipline of prompt engineering. I look forward to seeing how that role progresses, given my training in visual-to-verbal translation from Art Center College of Design (West Coast sister to RISDI).
A stock image was initially chosen for this campaign but the client later requested a photo that showed less skin; we didn’t want to lose the great mood of this model, so I extended the sweatshirt with layer masking and manual brushwork.
A passable generative AI fill-in. Telltale signs of CG still need to be fixed by a human, and my hand-drawn version better meets the client requirements, but AI is getting better at this.
The texture in this little black dress presented a digital tailoring challenge, and the client also requested a bit of cosmetic retouching.
The background is toned back to allow the candidate image to come forward, and alterations based on facial impression research to make the most of this candidate’s wonderful smile. Generative AI will be able to do the dress corrections soon, but it’s not up to redrawing a convincing human eye.
A candidate with impressive academic credentials but a very limited photo shoot; I wanted to get a warmer, more focused gaze from this one image.
I started by addressing the subject’s right eyelid, redrawing to adjust slightly— without losing the subject’s personality.
In the final image, a series of very small adjustments add up to a warmer smile and eyes that gaze out with clarity and purpose.
Here, the client requested some wardrobe assistance. Additionally, the background detail in the image is quite sharp, distracting from the subjects, the figures are too far apart, the color is grim, and two subjects’ expressions are a little cool.
After fixing the shirt, the figures are moved closer together, some of the background detail is removed to keep focus on the figures, and subtle facial adjustment warms up the expression on two of the figures. A good generative AI prompt can speed up the process, but a human eye is still needed for realistic results, for instance; AI isn’t yet able to redraw the blouse or evaluate subtle facial expression.
The handshake is the conceptual focus for this candidate introduction piece, but its position in the shot is too close to the photo’s edge, leaving a gap before the bleed line. The proportion of the image also isn’t wide enough for the media frame and there isn’t enough “cushion” around the candidate to zoom in further.
Here, the edges of the photo and foreground figures are extended with a combination of scaling, masked layers, and plain old-fashioned drawing to create image material where there was none before. The image and handshake now fit comfortably into the frame around the designer’s lovely text treatment. My extension was all drawn manually, but this is definitely the kind of project at which generative AI excels.
The figures in this composition are a little spread out, limiting how closely the image can be cropped, minimizing the impact of the candidate’s lovely smile. It could also be a little warmer and brighter overall to better convey an upbeat, hopeful mood.
In this version, the figures are grouped closer together and the color corrected. I’ve also brightened up the expression on the right-side supporting figures; while unlikely to be consciously noticed by the casual viewer, they still affect the mood of the composition. These are the kinds of corrections we can’t do with AI… yet.
A different kind of wardrobe adjustment: this time, a color change to the candidate’s dress. The image also needs overall tonal correction.
A different dress and warmer, more vibrant skin tones. In this version, there is also some facial enhancement; the candidate has a terrific smile that can be made even more engaging with subtle expression retouching. Finally, it’s worth paying attention to the figures around a candidate and the mood suggested by their expressions as well. Again, AI isn’t up to this kind of subtlety yet.
The height disparity among these council candidates makes the grouping look awkward and limits how the image can be cropped. The architectural detail in the background is distracting, and the color and skin tones need correction.
Here, the height disparity is evened out a bit and the distracting background elements removed or softened. Skin tones and the image overall are color corrected, and slight adjustments made to facial expressions and eye focus.