Hi everyone,
I’ve built a workflow in n8n to help create high-quality product photos for e-commerce listings.
The idea is to take basic photos sent by suppliers, which usually aren’t professional, and enhance them using AI so they look polished and ready for an online store. I’ve been using gpt-image-1, and so far the results have been promising. As shown in the image on the right, the output looks much more refined, and I’ve been able to apply some elements of the brand’s visual identity.
The main issue I’m running into is that the AI often alters the clothing too much. The final image doesn’t stay true to the original, details like fabric texture, patterns, or design elements (like in the blouse shown) often get lost or modified.
So my question is: How can I achieve this kind of image enhancement while preserving the original clothing details? Is there a better AI model for this use case, or is this just a current limitation of AI image generation?
I’d really appreciate any guidance or suggestions. Thanks!
Hi there, the problem you’re seeing is pretty common when using image generation models like gpt-image-1. These models are trained to enhance or “imagine” improvements and end up sacrifice accuracy (like pattern or texture) in favor of a cleaner look.
Try using a model that focuses on editing only parts of the image (like background cleanup or lighting), while preserving the actual product. Tools like Runway ML, Replicate (with ControlNet or InstructPix2Pix), or even Photoshop’s Generative Fill are better suited for this.
If you’re open to more advanced setups, look into ControlNet with Canny or LineArt models. These help the AI understand outlines and preserve structure. You can send original photos + edge maps to maintain fabric cuts, folds, and seams.
Hope this helps, cheers mate
Thanks @Abrar_Sami! I did some tests with Runway ML looks promising!