← Back to the main Geek Mode article
What was tested: successive precise edits to an existing WordPress Media Library screenshot, given to Gemini 3.8 Flash (Flash Extended) and GPT-5.6 Sol (Extra High) in early September 2026. Outcome: both could edit the interface image, but ChatGPT followed successive surgical edits more reliably; Gemini’s second output inserted incorrect location/business information.
The mock-up involved changes to the WordPress Media Library interface on jamiedolan.com. Both systems received the same source screenshots and instructions: add new admin-only fields while preserving the existing interface, then in a second pass add a further field along with sample EXIF data and location information.
This is much closer to the kind of image work actually needed day to day: don’t redesign everything, don’t get creative, just change these specific things while preserving the rest.
ChatGPT did this well. The edited interface remained recognizably the original interface, the requested fields landed where asked for, and successive edits generally preserved previous work.
Gemini’s output was weaker. This is a recurring problem with Gemini image editing in firsthand use: a good first image, but once precise corrections or successive edits are requested, it’s more likely to drift, fail to preserve something, or simply refuse to fix a detail no matter how explicitly it’s described. In this case the second edit introduced incorrect Neenah location and business information that was never requested.
Two screenshots don’t prove a universal property of an image model. They do match what has been repeatedly encountered in normal use.
← Previous: identical-prompt image generation · Back to the main Geek Mode article · Next: ZIP code ingestion →