AI can undeniably help small businesses that don't have large budgets for lifestyle photography to create visuals for their communication. I have been playing with a product photography-focused AI service for a couple months now, and i wanted to share my experience with it, ranging from light edits to more substantial modifications, at least when it comes to our products.
AI-edited Images
There's a wide spectrum here, from simple edits such as changing the studio background or lighting in a still-life product shot, to modifying a product's color—something that's traditionally been done in Photoshop—to replacing the model wearing the bag or placing them in a completely different setting.
Changing the color or fabric texture of a bag, whether the original image is a detailed still-life product shot or a styled image featuring a model, is relatively easy and definitely useful to create mockups for internal use. With some additional post-production work, these images could also be used for marketing purposes or even as product images on a website, although I haven't done that yet.
For example, starting from the image above, I asked ChatGPT (in this case) to change the color according to specific Pantone and HEX codes. After several successful attempts, it also started making color mistakes. These are some of the better results:
I also asked it to make the backpack appear as if it was made from waxed canvas in a Carhartt Brown color and to replace the model with an Asian woman in her 30s.
On the first attempt, the model had the face and hairs shown here but retained the same body proportions as the original male model. I had to ask for a more realistic result before arriving at this image.
Using a good still-life pic and changing the studio setup, lighting or background color–or even animating the image into a short cinematic product clip with subtle camera movement and natural motion–can also be useful. I have already created some visuals that I will definitely use.
Here are a few example of the white-backed products photos we use on the website, edited with AI:
In these examples, we begin to see that, without a model and with simple backgrounds, the AI tends to focus more effectively on the product itself and maintain higher image quality. Adding distractions such as complex environments or models often leads to mistakes in the bag's details, like in the two images below, where side-release buckles appear instead of the G-Hooks used on the compression straps.
Natural and realistic backgrounds can work, but only if two conditions are met: no close-up shots and very simple, uncluttered settings, as in the examples below.
On the other hand, when using a styled image and asking the AI to change the model, alter the environment, modify the pose, or make other substantial edits, the results become less convincing. In my experience, two issues consistently emerge:
- The more elements the AI has to focus on, the more likely it is to get product details wrong, especially hardware.
- Models often still look artificial.
In the example below, I changed the model's ethnicity, some clothing details, and the environment. The image is almost entirely generated. While the backpack itself looks acceptable, I wouldn't use this image without additional editing because the AI-generated appearance is still too noticeable.
AI-generated Images
The services currently available offer a wide range of preset studio environments, outdoor settings, and standard models. However, I still find it difficult to achieve consistently good results, at least for our bags.
Even when I provided the AI with detailed, high-resolution images of the bag from every angle, there were simply too many details for it to reproduce accurately. Hardware components, in particular, are almost always altered or simplified, or messed up. The results improve somewhat with simpler products, such as the ARB Tote, but they are still far from reliable when applied to our backpacks.
If a detail—a buckle, a G-Hook, or another component—is not perfectly visible in the reference images, or if it's significantly different from what the AI considers "standard," it will often be replaced with a more common-looking part. As we've already seen, adding complex backgrounds or models further increases the likelihood of these errors.
Here are some examples generated from preset scenes, where the ARB Tote turns into a backpack, where excess webbing appears where it shouldn't, or where elements of the original preset bag remain visible on top of our product.
And here are some examples that work reasonably well, thanks to the larger number of product images provided to the AI as references. Once again, simpler settings allow the model to focus more effectively on the product itself.
Leave a comment