Google Gemini ad recreated with kid’s stuffed toy; AI calls it a puppy
When Google rolled out its Gemini showcase, the clip featured a plush toy that the model instantly identified as a dog, turning a simple visual trick into a headline‑grabbing moment. Curious, I swapped the advertised cuddly figure for my own son’s favorite stuffed companion and set out to see whether the demo’s ease of use held up under less‑polished conditions. I photographed the toy from several angles, fed the images into Gemini, and then ran a reverse‑image check to compare the system’s description with what a human would see.
The experiment quickly revealed a gap between the polished marketing piece and the everyday reality of feeding an LLM a homemade snap. What followed was a mix of surprising confidence and a few missed cues that made the task feel more like a puzzle than the ad suggested. The result?
A candid glimpse into how the model labels what it “sees,” captured in the line that follows.
The AI blurb when I do a reverse image search on one of my photos confidently declares him to be a puppy. Gemini did a better job with the second half of the assignment, but it wasn't quite as easy as the ad makes it look. I started with a different photo of Buddy -- one where he's actually on a pla
The AI blurb when I do a reverse image search on one of my photos confidently declares him to be a puppy. Gemini did a better job with the second half of the assignment, but it wasn't quite as easy as the ad makes it look. I started with a different photo of Buddy -- one where he's actually on a plane in my son's arms -- and gave it the next prompt: "make a photo of the deer on his next flight." The result is pretty good, but his lower half is obscured in the source image so the feet aren't quite right.
The ad doesn't show the full prompt for the next two photos, so I went with: "Now make a photo of the same deer in front of the Grand Canyon." And it did just that -- with the airplane seatbelt and headphones, too. I was more specific with my next prompt, added a camera in his hands, and got something more convincing. I can see how Gemini misinterpreted my prompt.
Was the effort worth it? The recreation shows that AI can splice a stuffed animal into exotic backdrops, yet the process proved messier than the original Gemini spot suggests. The reverse‑image search mislabelled the plush as a puppy, highlighting that the model still struggles with basic visual discrimination when fed a toy photograph.
Gemini managed the latter half of the assignment more competently, but the author notes it required more fiddling than the ad implies, leaving the workflow opaque. Consequently, while the technology can generate charming visuals, the practical overhead and occasional misclassifications raise questions about its suitability for casual projects. The piece underscores that the novelty of AI‑enhanced travel montages may be outweighed by the effort needed to achieve a clean result, and it remains unclear whether such tricks will become streamlined enough for everyday use without compromising accuracy.
In short, the experiment offers a glimpse of possibility, tempered by the reality that the tool is not yet as effortless as its marketing portrays.
Further Reading
- Google’s Gemini AI ad is so fake it’s basically sci‑fi - The Verge
- Google admits its heavily promoted Gemini AI video was edited - CNN
- Google’s impressive Gemini AI demo shows a conversation that never happened - WIRED
- Google’s Gemini AI demo raised the bar — and some eyebrows - MIT Technology Review
- After Gemini backlash, how honest should AI demos be? - TechCrunch
More in LLMs & Generative AI
LWiAI Podcast #229: Google defaults Gemini 3 Flash ChatGPT app launch Nemotron 3
The latest episode of LWiAI, numbered 229, dives into three headlines that have been buzzing the AI...
Hollywood embraced AI in 2025, but the partnership has yet to deliver results
Hollywood rolled out the red carpet for generative AI in 2025, betting that algorithms could...
Z.AI releases GLM-4.7, open-source model boosting coding, reasoning, text+vision
Why does this matter now? The AI community has been watching open‑source efforts intensify,...
AI firms scrape the web and millions of books without permission to fuel LLMs
The surge of large‑language models has turned data into a commodity, and the market’s appetite...
Anthropic reports Opus 4.5 awareness under 10% versus OpenAI in red team
Why does this matter? Because the latest red‑team exercises are exposing a gap that many thought...
Common Questions Answered
How did Google Gemini's ad misidentify the plush toy in the reverse‑image search?
In the author's test, Gemini's reverse‑image search confidently labeled the stuffed companion as a "puppy," even though the image showed a plush toy. This mislabeling highlights the model's difficulty with basic visual discrimination when presented with non‑photographic objects.
What differences did the author notice between the original Gemini showcase and their own test with a personal stuffed animal?
The author found that while the original ad made the process seem effortless, replicating it with a personal toy required more fiddling and multiple angle shots. Gemini managed the second half of the assignment better than the ad suggests, but the overall workflow was messier and less straightforward.
Did Gemini successfully generate a new image of the stuffed toy in an exotic setting, and what limitations were observed?
Gemini was able to splice the plush toy into an exotic backdrop, producing a fairly good result. However, the lower half of the toy was obscured in the source photo, which limited the quality of the final image and demonstrated that the model still struggles with incomplete visual inputs.
What conclusion does the author draw about the practicality of the Gemini demo based on their recreation?
The author concludes that while the Gemini demo showcases impressive AI capabilities, the actual process is more cumbersome than the advertisement implies. The need for careful image preparation and the reverse‑image mislabeling suggest the workflow is not yet as seamless for everyday users.