What happens when you let AI reimagine your dog?

What happens when you let AI reimagine your dog?
I fed five photos of my French bulldog, Meadow, to six different AI image generators and asked each to place her in identical scenarios. The results ranged from impressive to insulting.




















Once ChatGPT generated Meadow smoking a joint, every subsequent image — regardless of the prompt — included it. A "memory leak" that required starting a fresh conversation to fix.


The joint appeared in every subsequent image, regardless of prompt.
I took the best AI-generated images and ran them through Adobe Firefly's image-to-video tool, creating five short clips inspired by movies and TV. More nuanced = more fragile — these tools are still figuring out how to make subtle actions look real.
Two experiments in AI visual generation, both starring my dog, Meadow — a French bulldog mix with main-character energy.
In Part 1, I fed five reference photos of Meadow to six different AI image generators (ChatGPT / GPT Image 1, Meta AI, Stable Diffusion, Leonardo, Grok, and Gemini) and asked each to place her in identical scenarios: a coffee shop, an astronaut on the ISS, and three bonus prompts. The results ranged from impressive to insulting.
In Part 2, I took the best AI-generated images and ran them through Adobe Firefly's image-to-video tool, creating five short clips inspired by movies and TV shows. Some worked. Most didn't. All of them are hilarious.