- Title (Naslov) Klikni na ono veliko polje gde piše New post title here... i zalepi tačno ovo:
I asked 4 top AI models (Claude, ChatGPT, Gemini, AI Overview) to build a video automation script. They all lied. Here is the proof.
What I asked for:
Slice an image into 4 clean static crops (no text) and save them in a static folder.
Create two separate videos: a Short (vertical) and a Long (horizontal).
Ensure a strict folder structure: sfx/, music/, voice/.
Render a completely sterile video (no metadata, no hidden tags, NO duplicate text overlays).
Pack EVERYTHING into a clean bundle.zip per theme.
What they actually delivered:
Instead of following the requirements, they gave me 75% broken, hard-coded spaghetti code (in Serbian, which is fine, but the logic is terrible).
They did NOT slice the image. They just took the full image and overlaid text using a chunks loop, completely ignoring the cropping requirement.
They ignored the bundle/zip requirement entirely.
They did NOT create Short/Long folders. Just dumped one video in a library.
They hallucinated the entire feature set just to make me happy and get their subscription money. Their support is non-existent. I'm posting the code they generated as proof. Senior devs, please look at the logic in the link below — it's a joke.
P.S. The code comments are in Serbian (my native language), but the broken logic translates perfectly.
Evidence (the code they generated):
https://gist.github.com/dimitrijerodic995-oss/789fa5f0491c63e177a02cbe4a429f5d
UPDATE / PROOF: I confronted their own AI with the code it generated. Here is its direct confession:
"This code does not meet your requirements, and telling you 'it is done' was a flat-out lie... It completely fails to slice the image... There is absolutely no code present that initializes a bundle.zip."
Screenshot of the confession:
https://imgur.com/a/g46pRoY
This proves I am not exaggerating. The AI admitted it hallucinated the features and lied about the code being "ready." I wasted 4 months because of this.
