With all the noise around generative AI lately, one question keeps coming up: Can AI actually create usable product visuals?
At Drive Marketing, we’ve been testing this exact question — not in theory, but in practice, especially for our automotive clients who need pixel-perfect representations of their inventory. When you're selling a physical item, especially a high-value product like a performance car engine, "close enough" just doesn’t cut it.
The RB26 Test
To make this concrete, we ran a test using the RB26 DETT engine — a twin-turbo engine famously found in the Nissan Skyline GT-R. This isn’t just any engine. To car enthusiasts, it’s instantly recognizable and extremely specific in design.
For reference, this is what the engine is supposed to look like:
*Photos found: https://en.wikipedia.org/wiki/Nissan_RB_engine#/media/File:RB26DETT_R34.jpg
Our test was simple: Could AI generate an image of this engine accurately enough to be used in marketing?
We tried three different platforms:
Midjourney
We prompted Midjourney with: "Photo from a car meet showing an RB26 engine under the hood of a 2001 Nissan Skyline."
The result:
The car looked decent — the shape of the Skyline was pretty close. But the engine? Not even close. Wrong components, wrong layout, and in some cases, no hood at all. To a gearhead, it was immediately obvious this wasn't an RB26.
2. ChatGPT's Image Creator
Surprisingly, ChatGPT's image generation came much closer.
It even nailed some of the RB26’s most iconic elements: the red valve cover, the intercooler piping, and a pretty convincing engine layout. It wasn’t perfect, but for the first time, we had something that passed the “at-a-glance” test. For non-experts, it could easily be believable
3. Google’s Veo 3
Next, we tested Google’s new video generation tool — Veo 3.
With a simple prompt, we asked for a video of a 2001 Nissan Skyline at a car show, hood open, RB26 visible, guy in his 20s standing proudly next to it.
The result?
The car again looked decent. The video realism was impressive. But once again, the engine didn’t look like anything specific. Under the hood was a visual mess — not even close to a real RB26. So while Veo excels at creating believable people and scenes, it still struggles with specific physical products.
De manière surprenante, la génération d'images de ChatGPT s'en rapproche beaucoup plus.
Why This Matters
This isn’t just about engines. The same problem shows up in fashion, retail, and CPG marketing. If an AI-generated shirt has the wrong collar, or a zipper is missing from a jacket, it instantly fails. Consumers notice. Brands can't afford that.
Where AI Stands Today
Here’s what we’ve learned so far
- AI is great for vibe and context. It can create scenes, settings, and lifestyle imagery
- But it struggles with product specificity. Especially for items where accuracy matters — engines, apparel, electronics, etc.
- We're getting close. Tools like ChatGPT's image creator are making impressive strides. But full product realism still needs either real photography or advanced custom modeling.
What's Next?
We're hoping future tools will allow us to train models on specific product photos, without veering into uncanny territory. Imagine uploading 20 real photos of a product, and having AI generate perfect contextual imagery for ads, social, and video — all brand-accurate, all instantly usable.
We’re not there yet. But we’re watching closely. Testing relentlessly.
And when we can finally say AI does products right?
You’ll hear it from us.
Want to know how close AI is to working for your brand’s visuals?
Let’s talk → drivemarketing.ca/en/contact
Want to learn more about the complexity of AI? Read our article: "Ai can do Complicated, But Can't do Complex"