About Platform Resources Articles Contact
AI Tools  ·  IMA AI

Generation or Editing?
I Was Asking the Wrong Question.

Every time I opened an AI image tool, I lost a few seconds to the same small decision: am I generating or editing? It sounds trivial. It wasn't. I went in circles on this for longer than I'd like to admit. Then it clicked — and the answer is almost annoyingly simple.

Published  June 2026
By  Chin Qi Yong, CEO — IMA AI
© 2026 Chin Qi Yong
Read time  ~4 min

It's all generation

At the level that matters — what the model is actually doing — there is only one thing happening: the model generates pixels from whatever you give it. That is the whole engine. It never does anything else.

"Generation" and "editing" are not two different machines. They are two workflows running on the same machine. The word "edit" describes what you are doing — changing something that already exists. It does not describe what the model is doing, which is always the same: generating.

Once I saw that, the confusion dissolved. Brushing over a region in an image isn't a different mode. It's just one more instruction — you're telling the model where it's allowed to generate. Still generation. Just pointed at a smaller area.

What actually changes: the dial

If the engine is always generation, then the real difference between "generating" and "editing" is something else entirely: how much of an existing image you feed back in as a constraint.

That's not a switch with two positions. It's a dial.

The constraint dial
Free
Give it only words — nothing to preserve — and it generates freely. We call this "generation."
Give it words plus a reference image, and it generates while leaning on that image for structure, style, or composition.
Give it words plus an image plus a brushed region, and it generates only inside that region and leaves everything else alone.
Locked
Give it words plus an image plus a tiny instruction — "make the shirt red" — and it changes almost nothing and preserves almost everything. We call this "editing."

Same engine, the whole way down. The only thing moving is how tightly you're pinning it to something that already exists. Low constraint — we call it generation. High constraint — we call it editing. Everything useful happens on the dial in between.

The question that replaced the old one

So I stopped asking "is this generation or editing?" — because the honest answer is always "both, it's generation." That question was never going to resolve, which is exactly why I kept getting stuck on it.

The question that actually pays off is: how much of an existing image do I need to keep?

Nothing to keep? You're at the free end. Describe it and let it run. A real product, a real face, a real layout that has to survive untouched? You're at the constrained end. Give the model the image and pin down everything you're not changing.

The most expensive mistake I see people make: trying to generate something that had to stay real. If you "generate" your own product, you don't get your product — you get a confident lookalike with the wrong logo and invented details. The model was never preserving anything. You just hadn't given it anything to preserve.

That single question — how much do I need to keep? — routes almost every decision I used to fumble. You are no longer choosing a mode. You are choosing a position on the dial.

One catch — the tool still matters

Here's the part that keeps this from being a neat "it's all the same, relax" story.

Mechanically, yes — it's all generation. But models are trained for different ends of the dial. A pure text-to-image model, pushed to edit, will quietly redraw things you wanted left alone — because it was built to invent, not to preserve. A model trained specifically for editing is taught to hold the rest of the image still while it changes the one thing you asked for.

The sharper takeaway
It's all generation — so match the model to how much you're trying to keep. Reaching for a generator when you needed an editor is most of the frustration people experience with AI image tools. It looks like the tool failing. Really it was the wrong tool for that point on the dial.

The short version

The industry sells these as two features with two buttons. Under the hood it is one engine and a constraint dial. The labels "generate" and "edit" describe your workflow, not the model's mechanism.

Once you stop asking "which mode is this?" and start asking "how much do I need to preserve?", the guessing stops — and you start picking the right tool on purpose instead of by trial and error.

That reframe cost me a few weeks of low-grade confusion. Hopefully it costs you about five minutes.

CQ
Chin Qi Yong
CEO, IMA AI
Chin Qi Yong is the CEO of IMA AI. IMA AI builds AI-powered infrastructure for commerce and content operations in Malaysia. Currently building Ultra Studio — an internal automation platform for script-to-content production.
Follow on LinkedIn

Published by IMA AI — June 2026.