AI product photography tested: Nano Banana Pro vs GPT Image 2 (3 failures)

This is video 3 in a series of 30, each one walking through a Runflow workflow and how it holds up in a real production pipeline. This one is product isolation: take a lifestyle photo, mask the product, get a clean cutout back. I run the same task through three engines on three inputs: Runflow's product-isolation workflow, Google's Nano Banana Pro, and OpenAI's GPT Image. Same mask, same prompt. Then I chain a second workflow to drop the cutout into a fresh lifestyle scene, which is where this is headed for e-commerce and resale catalogs. Runflow held the product on all three inputs. The closed models drifted on detail, swapped the object, or invented a fake transparent background. Chapters: 0:00 What product isolation does (and what this 30-video series is) 0:51 How I compare it against closed models 1:01 Example 1: the poster with the mirrored logo and double watch 2:10 Where Nano Banana Pro and GPT Image fall apart on it 2:46 Example 2: the wicker basket 3:27 Production case: isolating a side table from a Sklum lifestyle shot 4:55 The same table through Nano Banana Pro and GPT Image 5:38 Step 2: generating a fresh lifestyle background for the cutout 6:58 The honest caveats (wrong table, odd shadows) 7:59 Where this gets useful: secondhand and resale listings Full write-up with the side-by-side images: https://www.runflow.io/blog/product-i... Try product-isolation: https://www.runflow.io/api/product-is... Got a SKU pipeline you want me to run through the comparison? Send the source images to [email protected]. And if you spot something wrong or have a workflow you'd want to see, drop it in the comments.