video1.0<iframe src="https://www.loom.com/embed/3fb728825f75498d9d5a9db7358dbe17" frameborder="0" width="1920" height="1440" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>14401920Loomhttps://www.loom.com14401920https://cdn.loom.com/sessions/thumbnails/3fb728825f75498d9d5a9db7358dbe17-5fa5b59a710de405.gif400.598This Loom compares an object removal workflow against closed source models, focusing on production reliability. The author tests removing specific items from images using NanoBanana and GPT-Image Pro 2, noticing that closed source outputs often change the image ratio and framing, such as zooming in or out. They also report hallucinations, including adding a second background tower and generating random text, and in other cases the plate or objects appear larger even when removal is correct. Processing time for the closed source models is typically between half a minute and one minute, sometimes taking more than three minutes.