.svg)
I published a comparative test between Transformers 3 and its version recreated by AI. What was supposed to be a simple technical exploration turned into an explosive debate with hundreds of comments, mostly critical. A humble and instructive look back at a controversy that taught me much more than the test itself.
The test that started it all
My methodology: reproducing Transformers 3 with Kling AI
The idea seemed simple: take iconic scenes from Transformers 3, play them in Kling AI 2.5 (integrated into the Scenario platform) via an image-to-video function, and compare the results. My aim? Concretely assess whether AI can compete with traditional VFX on complex courses of action.
My process:
- Selection of scenes from the original movie
- Generation AI in “one-shot” without complex prompts
- Direct visual comparison: movie vs AI
My first results seemed impressive to me. In some ways, the quality impressed me. I thought it could be used before post-production. Looking back, maybe I was a bit too excited.
The hook that rocked everything
I asked the question: “Do we still need an army of VFX artists for shots that an AI can generate in a one-shot?”
Today, I admit that I have fallen into the trap of the LinkedIn algorithm. This provocative hook, designed to generate engagement, completely biased the perception of my intention. Mon Original post has accumulated hundreds of reactions - and many have hurt me, but above all made me think.
The paradox I hadn't seen
The circularity of my approach
The first blow of the sledgehammer came quickly: to create my AI version, I used the original movie. This evidence seemed secondary to me at the time, but it invalidates any claim to demonstrate that AI could replace the initial creative process.
As one expert pointed out to me: I'm not showing that AI can do VFX, I'm just showing that it can copy existing content - created precisely by these “armies of VFX artists” that my title seemed to make obsolete.
I confused reproduction and creation
It's probably the hardest lesson to learn. Reproducing is not creating. My test is the equivalent of using a photocopier and pretending that writers are no longer needed. The metaphor is brutal, but accurate.
AI has not replaced the creative process that involves:
- Art design and creative direction
- Storyboarding and previewing
- Choice of narrative techniques (focal lengths, camera movements)
- Iterative adjustments with the director
- Construction of a coherent visual universe
I tested a reproduction tool, not a creation tool. Fundamental nuance.
The technical limits that I had not seen
My lack of VFX expertise revealed
I am a DA and an illustrator, not a VFX expert. And it showed. A compositing artist reminded me of the obvious: how to ask the AI for a 28mm rendering rather than 50mm? How to obtain a specific anamorphic appearance?
The technical data that I was missing:
- Camera settings (focal length, aperture, depth of field)
- Sensor information and lenses used
- Point distance and tracking data
- Consistent light behaviors and reflections
- Precise motion blur management
These elements make up the visual narration of a movie. Without them, I produced images that were certainly impressive for the untrained eye, but narratively empty for a professional.
The quality gap that I minimized
Looking back and the feedback received, I must admit: the original images are vastly superior. Framing, contrast, patina, blur, grain... The AI version is consistently in the background, and not marginally.
I was impressed with what AI was producing because I did not expect this level. But comparing it to the excellence of a real Hollywood production is a different story.
What the professionals taught me
Professional pipeline: the reality on the ground
The feedback from production professionals opened my eyes to the real requirements:
- Broadcast-ready 4K/HDR resolution
- Total absence of tolerated artifacts
- Controlled iterations with versioning
- Integration into complex pipelines
- Multi-team collaborative management
My test was a “digital communication” exercise where a certain vagueness passes. In Netflix or movie production, these limitations are unacceptable. AI remains confined to creative tests and less demanding content.
The ethical error that I did not anticipate
Use Transformers 3 scenes without permission to demonstrate that we could do without their creators... Presented like that, it hurts. And yet that is exactly what I did, even without that intention.
This approach illustrates a lack of understanding of the creative value chain. In a professional context, I would have been exposed to intellectual property infringement lawsuits.
My mea culpa: the mistakes made
The LinkedIn algorithm trap
I fully admit it: “I fell into the LinkedIn algorithm trap. The catchphrase gives the impression that I am asking a legitimate question... even though I use these tools on a daily basis and I know full well that it is still a tool with huge limitations.”
I sacrificed nuance for reach. The hook worked - hundreds of comments, lots of visibility. But at what price? By distorting my real thinking and by offending a community that I deeply respect.
What I should have said
With hindsight, here's how I should have worded my post:
“AI test: I wanted to see how far Kling AI can replicate scenes from Transformers 3. Interesting results but obvious limitations. Spoiler: we are very far from being able to do without VFX pro teams. Feedback and technical limitations →”
Less sexy for the algo, infinitely more honest intellectually.
The lessons I learned
On the methodology: Starting from existing images fundamentally biases the test. Next time I'll create original content from start to finish.
On communication: Algorithmic performance should never take precedence over the rigor of the subject. Toxic reach brings nothing.
On humility: When you are dealing with an area that you do not fully understand, it is better to ask questions than to affirm conclusions.
The hybrid approach: what I really think
AI as a support tool, not a replacement
My deep conviction, which my post misexpressed: AI will not replace VFX artists. It can assist them with specific tasks. In fact, this is already the case in many professional workflows.
Legitimate use cases I see:
- Filling in missing post-production shots
- Temporal extension of existing sequences
- Quick preview for creative validation
- Upstream environment and concept tests
- Upscaling and resolution improvement
But always in a logic of assistance, never a complete substitution.
Free up time on the essentials
The vision that drives me: to relieve teams of repetitive and exhausting tasks to allow them to focus on complex sequences with high creative value. Have a “fitter and less exhausted army to focus on the things that are more valuable.”
This pragmatic approach preserves human expertise where it is irreplaceable: artistic vision, creative judgment, and narrative coherence.
What this controversy taught me
The importance of a professional perspective
The harshest reviews often came from real VFX professionals. And that's normal: they saw immediately what I didn't see. The artifacts, the inconsistencies, the lack of technical mastery.
A senior composer pointed out to me that in the collective unconscious, this type of post creates a dangerous perception: that a “random guy” could replace 20 specialized artists. That was not my intention, but that is what my title suggested.
The responsibility of AI content creators
We who work with and communicate with AI have a responsibility: not to oversell current capabilities, not to distort reality for the sake of reach, not to devalue human creative work.
This controversy reminded me that every post has an impact. Decision-makers who read us can form their opinions on these bases. If you distort reality, you are helping to create unrealistic expectations.
My outlook today: more nuanced, more humble
What AI can really do
After this experience and all these exchanges, here is my realistic vision:
Video generative AI is impressive for rapid creative testing, previewing, and less demanding digital content.
She is beginning to integrate into hybrid workflows where she assists professionals who maintain creative control.
It is not ready to replace VFX teams on demanding productions, and may never fully be.
The future I see: collaboration, not disruption
Contrary to what my title suggested, I don't believe in a sudden replacement. I believe in a collaborative evolution where:
- AI offloads on simple and repetitive tasks
- Humans maintain creative and artistic control
- Workflows are hybridized intelligently
- Creative excellence remains the priority objective
This is less spectacular for LinkedIn, but infinitely more realistic and respectful of creative work.
Conclusion: thanks for this lesson
This controversy was painful at times. Some of the comments were harsh, sometimes even aggressive. But the majority were fair and constructive.
What I remember:
I confused the performance of a tool with the replacement of a profession. Fundamental error.
My hook betrayed my real thinking to perform algorithmically. Communication error.
I lacked humility in a field that I don't know enough about. Ego error.
But I also learned a lot. About the real limits of AI, about the professional requirements that I didn't know about, about my responsibility as a content creator.
I am leaving this post online as a reminder. So as not to fall back into the trap of sensationalism. To maintain this newfound humility. To continue to explore AI, but with greater rigor and respect.
AI is a fascinating tool. But it will never replace the talent, experience, and vision of real VFX artists. I should have started there.
And you, what do you think? How do you integrate AI into creative workflows without devaluing human work? I am listening to your opinions, and this time I will really read before replying.
.webp)
