💡 What I Learned (And Why It Matters)
After this intense week, here are my results:
1. Human skills are irreplaceable
AI is a tool.
Extraordinary revolutionary.
But it is only a tool.
What matters is the intention behind it.
If you let an AI decide your aesthetic, you’ll get an AI aesthetic.
If you direct the AI with your skills, you will find your aesthetic.
2. Open source = artistic freedom (not just “free”)
Here’s what I learned:
Open source is free and transparent
You control your data – I know exactly which photos trained my style. There is no hidden dataset. No undue bias. If a client asks tomorrow, “Where does this style come from?” , so I can show every image.
You know how to create your style – The process is not a black box. I’m watching every step. I can adjust. Improve repetition.
Zero surprises, zero hidden bias – When I describe my photos, I consciously choose my words. I don’t let an algorithm decide for me and reproduce hidden stereotypes.
🌍 We learn together, we move forward together
Open source tools are collaborative.
Thousands of creators, researchers and enthusiasts share their workflows, discoveries, mistakes.
Result:
- Best ideas circulate (someone in Korea finds a trick → 2 days later it’s on GitHub)
- We enrich features every week (no need to wait for a company’s roadmap)
- Anyone can adapt this tool to their needs (weird workflow? You code it.)
For my film, I used techniques shared by the community. And now, I share my workflow instead.
It’s open source: a rising tide lifts all boats.
🎨 Customization at the heart of the process
With these tools, I don’t just use AI.
I shape it.
150 carefully crafted images. Each image is described with exact words that translate my intent.
Result:
- Your style doesn’t resemble anyone else’s (because it’s built on your choices).
- You protect your visual identity (guaranteed intellectual property).
- No dependence on a provider (Opnai closes tomorrow? You don’t care.)
✅ And it is ethical
Copyright Respect: My styles are trained on my own work, royalty-free images, or client datasets that I have permission to use. Zero theft
GDPR Compliance: I know exactly what data goes into my model. I can trace every image. If a client requests deletion: I can.
Consciously trained models: When I describe my images, I make conscious choices. I don’t reinforce stereotypes. I do not hold prejudice.
3. Quality > Quantity (Always)
150 intentional images > 1000 random images.
In any field
Careful curation trumps blind accumulation.
4. The risk of AI slope is real
Do you know the term “slope”?
A tsunami of auto-generated content polluting the internet.
Articles written by Chat GPT without any supervision.
Midjourney Images Clone Unlimited.
Mass produced videos without any intention.
The result?
Aesthetic uniform.
The race to the bottom.
Loss of homogeneity
My advice?
Automation is attractive.
It’s fast. It’s easy. It is seductive.
But if you want to create something with value,
something that is truly you,
Something that stands…
You can’t just “generate” and hope.
you need:
- Curate with care
- Describe skillfully
- Supervise training
- Test and Iterate
Is it long? yes
Is it more complicated? yes
But there is a difference between:
- A fast food joint (instant, standard, forgettable)
- A Michelin star restaurant (intentional, unique, memorable)