Good News: AI Kind of Sucks at Retouching, Study Says

4 min read Original article ↗

With the introduction of AI, many companies have rushed to invest in the medium. This includes photo editing platforms, which have claimed they truly help reduce workload and increase output. What’s more, the brands are working to push it on all of us. While the latter is a great idea, the quality has often been a question that has never seen a proper answer — until now. New research conducted by a Virginia-based company reveals a different truth. One that will hurt the AI companies. So does humanity have hope?

What is the Study?

Color Experts International, who claims to be one of the world’s largest image manipulation service providers based in Leesburg, has recently conducted a study that reveals whether AI is better than human retouchers. In a study titled “AI vs. Human Retouchers: A Comparison in Product Photo Retouching,” the authors reveal that image generation models certainly fail.

Chart outlining performance fields assessment for image quality, editing, and process efficiency with related criteria.
Methology
Close-up of a TAG Heuer watch face; AI and human retouched versions shown side by side for comparison.
Side-by-side example

Speed and Quality

Bar chart showing average completion times; Nano Banana Pro is highest at 28.65 mins, others under 3 mins.
Infographics by Color Experts International,

One of the main areas of the study was the two aspects: speed for reaching the final product vs the quality of the output. In their research, they found that Nano Banana Pro completed edits in 25.4 seconds, while human retouchers averaged 28 minutes and 39 seconds.

Side-by-side of original and AI-retouched BBQ sauce bottle photo, with labels noting visible editing differences.
Example of AI retouching
Side-by-side images of a BBQ sauce bottle, showing original and retouched versions with improved label clarity.
Example of human retouching

Rather uninterestingly, the results revealed that AI tools were about 22.7 times faster than human retouchers. Moreover, Nano Banana Pro was 67.6× faster than human retouchers. But when it came to the quality test, human retouchers scored higher, with an average of 8.85/10. As for AI models, they got 5.16 out of 10, or 41.7% lower overall. Each score was given across 11 performance fields. The finding also revealed that Nano Banana Pro performed best among the various AI tools, scoring 6/10, but still trailed human output by 32.2% in quality.

From a professional standpoint, working with AI like Nano Banana isn’t as good as working with someone with professional skills. With both humans and AI, edits often go through rounds of revisions. And with humans, there are things like client proofing that make the whole process easier. But with AI, you end up sometimes struggling with how to communicate with it. Consider, too, the fact that humans even have trouble communicating with one another and getting our ideas across clearly.

Technical Versatility

Another important part was which of the two is more versatile and produces fewer mistakes. Here, AI failed, with most prompts having errors and misinterpretations on RAW and high-resolution TIFF files. Sharing an example, the report reveals that when asked to remove the backdrop in a RAW image of a slingback heel, “GPT Image 1.5 produced a white-background image of a stainless steel water bottle.” But humans, who have a better understanding of nuances and language, completed the task easily across all formats and file sizes.

Screenshot showing a black water bottle and a blush pink slingback pump, each on a clean white background.
Example of the tests

The Hybrid Future

Another part of the test was to combine AI and humans to see what the results looked like. With a human retoucher using AI, the results were completed faster than a fully manual approach. For instance, in one hair and cosmetics product test, a hybrid workflow took “7.18 minutes, compared to 10 minutes entirely by hand.”

Bar graph comparing retouching times: hybrid retouching at 7.18 min, human retouching at 10 min.
Test result of AI retouching
A yellow Kerastase soap box with AI and hybrid retouching workflows shown side by side using Photoshop layers.
Result of hybrid retoucing

The findings suggest that hybrid results were better due to “specific capabilities and consistency of the AI model used.” The company suggests that AI speed can greatly accelerate simple edits, “freeing” human retouchers to focus on and refine complex tasks.

While the report suggests that hybrid is the future, the question is whether companies are ready to accept it. One of the reasons is that Adobe and VSCO are examples that seem to be creating quick solutions to improve workflow, but the problem is that these tools are not perfect yet. But it is easier to use and invest in them than it is to hire human retouchers. However, if the hybrid solution is the best way forward, then there must also be the demand that human reouchers get paid their fair share. If not, then no amount of tools can change the pay disparity.

Here’s something for photographers to think about: do you want an AI making your next camera? If you use AI for your own personal benefit and don’t see how we should all be protected from it, then you’re being a big part of the problem.

Earlier this year, a company tried to get us to promote them by offering us an article written by AI. The crazy thing is that it was promoting the fact that humanity was still doing a better job than AI. However, they used AI to write the article. We refused to run it.

Do you see the irony?