LateNiteSoft, a company with 15 years of experience in mobile photography apps, conducted a study comparing AI image generation models for various photo edits. They ran over 600 tests using models like OpenAI's gpt-image-1, Gemini, Seedream, and others. Their goal was to assess which models worked best for common editing tasks, relying on user-friendly prompts that typical users might employ. The study revealed significant differences in how each model handled specific prompts. For instance, Gemini excelled in preserving details and maintaining realism, particularly for photo filters, but often lacked creativity and could avoid applying edits altogether, especially on images of people. OpenAI's model, in contrast, tended to alter details excessively, resulting in a less desirable "AI slop" effect. Seedream's performance varied, showing promise in specific tasks, like creating visually appealing effects, but struggled with others that required a nuanced understanding. The article also highlighted challenges with prompts. For example, when tasked with isolating an object, Gemini asked for clarification on which object to focus on, indicating limitations in handling ambiguity. Furthermore, while Gemini and Seedream produced results that stayed true to the original images, OpenAI's model frequently generated hallucinated details. The findings aim to guide users in choosing the right AI tools for their editing needs, emphasizing the importance of understanding each model's strengths and weaknesses.