I gave five AI video generators the exact same job:
Create a Singapore hawker centre scene at night with char kway teow, crowd, steam, and energy.
No tweaks. No second tries. Just first outputs.
What came back wasn’t just different quality—it revealed how each AI thinks.
The Prompt That Started It All
Here’s the full prompt I used:
A dynamic Singapore street food hawker centre scene at night. Bustling crowd, sizzling wok with char kway teow, steam rising, vibrant neon signs, people eating at tables, cinematic food lighting, smooth camera pan from close-up food to wide shot of busy center, 4K realistic details, energetic atmosphere.
At first glance, it sounds like a strong prompt.
But hidden inside is a subtle mistake: “vibrant neon signs.”
That one phrase changed everything.
Why Most AI Failed This Test
A real Singapore hawker centre doesn’t look like a neon-lit movie set.
It’s practical.
Fluorescent lighting.
Simple signboards.
Functional, not flashy.
So the challenge for each AI was this:
Do you follow the prompt exactly…
Or do you correct it based on reality?
Most tools chose the first option.
Tool-by-Tool Breakdown
CapCut Seedance 2.0 — Literal Interpretation
The food didn’t look convincing. It lacked texture and that slightly oily finish you expect from real char kway teow.
When the camera widened, the neon lighting dominated the scene. It looked stylised, not local.
This is what happens when AI follows instructions without context.
DOLA AI Seedance 2.0 — Good Detail, Wrong Environment
There were moments I liked.
The egg yolk breaking looked natural. The noodles had a fresh, pre-cooked look.
But once the scene expanded, the same issue returned—neon visuals that don’t belong in a hawker centre.
It understands food detail better than environment.
Google Veo 3.1 — Over-Cinematic Output
This one felt like it was trying to impress.
Flames from the wok.
Heavy smoke.
Crowd density pushed too far.
Signboards duplicated or exaggerated.
Everything was amplified.
The problem? It no longer felt real.
Meta Vibes — Early Stage
This tool still feels like it’s learning the basics.
The scene lacked consistency, and it struggled to recreate a believable real-world environment.
It’s not ready for complex lifestyle scenes yet.
Bing Video Creator — Context Over Prompt
This was the only output that felt grounded.
The char kway teow looked properly cooked—dark, slightly glossy, believable.
The environment resembled a real hawker centre.
The signboards were normal, not exaggerated.
And most importantly…
It didn’t fully follow the prompt.
The neon lights were toned down or absent.
It chose realism over instruction.
What This Test Reveals About AI Video Tools
This wasn’t just a comparison of features or quality.
It exposed a deeper difference:
Some AI tools are instruction-driven
Others are context-aware
And for real-world content, context wins.
A Better Way to Think About Prompts
Most creators believe:
“The more detailed your prompt, the better your result.”
But this test shows the opposite can happen.
If your prompt contains inaccurate details,
more detail just amplifies the mistake.
So instead of writing “more,” aim to write “truer.”
For example:
- Replace cinematic assumptions with real-world descriptions
- Use location-accurate details
- Avoid adding elements just to make things look “cool”
Important Context: This Was a First-Run Test
To keep things fair, I didn’t refine anything.
No iterations.
No corrections.
No multiple generations.
That means: Some tools—especially —might perform better with guided prompting or multi-step workflows.
But this test reflects something important:
What happens when a beginner uses AI for the first time.
Final Takeaway
I didn’t expect this outcome.
The tool that stood out wasn’t the most dramatic or cinematic.
It was the one that felt the most real.
And that changes how I’ll be using AI going forward.
Want My AI Video Prompt Guide?
If you want to learn how to write prompts that produce realistic, high-quality AI videos for food, travel, and lifestyle content…
Comment “AI” on my post.
I’ll send you my FREE AI Video Guide.
~ Adrian Lee @AdrianVideoImage
![]()
Discover more from Adrian Video Image
Subscribe to get the latest posts sent to your email.

