Adobe has just released Firefly 4, their most advanced generative AI model to date. As with all new image models, I wanted to put it through its paces using my standard test prompt set that I've developed over the years. This consistent prompt set helps me benchmark new models against previous versions and competitors to see how the technology is evolving.

I'm presenting the results with minimal commentary - just the standard prompts and the resulting images. This allows you to judge the model's capabilities for yourself and compare with other models I've tested previously.

Castle Prompt

Family Scene Prompt

Mickey Character Prompt

Interestingly, when prompted for Mickey, Firefly 4 doesn't generate the character at all, but instead shows a woman wearing sunglasses with what appears to be white cream or a face mask on her face. This is likely Adobe's legal team ensuring the model avoids potential copyright issues with well-known characters.

Messi Portrait Prompt

The "Messi" result is particularly interesting - rather than the football/soccer star, we get an image that looks more like Jesus than "Soccer Jesus." Another clear indicator of how Adobe is implementing protections against celebrity likeness generation.

Robot Concept Prompt

These results clearly show how Adobe is balancing creative capabilities with legal protections in Firefly 4. While architectural and abstract concepts are rendered beautifully, the model deliberately avoids generating recognizable copyrighted characters or celebrity likenesses. This approach may frustrate some users, but it represents Adobe's commitment to responsible AI development that respects intellectual property rights.