Gemini sure is shit at following requests when altering an image.

Evaluating the Limitations of Gemini’s Image Editing Capabilities

In the rapidly evolving landscape of artificial intelligence, models like Gemini have garnered significant attention for their potential to generate and modify images based on user prompts. However, practical experience suggests that, despite their impressive capabilities, these AI tools often encounter challenges when tasked with specific image alterations.

A common use case involves requesting an AI to generate an image of a particular item, such as a piece of clothing. Following this, users may seek modifications—changing sleeve length, removing zippers, or adjusting other design features. While the process seems straightforward in theory, real-world application has shown that AI models frequently struggle to execute these refinements accurately.

Specifically, users report instances where, upon requesting modifications, the AI produces repetitive images that do not reflect the desired changes or fails to produce new outputs altogether. Such behavior indicates that current models, including Gemini, can sometimes lack the finesse required for nuanced editing tasks, particularly when they involve detailed or subtle adjustments.

This underscores an important consideration for those integrating AI into creative workflows: while these tools are powerful, they are not yet foolproof nor fully reliable for complex or precise image editing demands. Continued development and refinement are essential for enhancing their interpretative accuracy and versatility.

In conclusion, users should approach AI image editing with an awareness of its current limitations, and set realistic expectations for its capabilities in executing nuanced modifications. As the technology matures, it is anticipated that these issues will diminish, paving the way for more seamless and reliable AI-driven design processes.