Layers
A transform is a stack. The prompt is only the top layer.
It is tempting to think of an AI transform as one button. It is closer to a stack: a source, a region, a thing being held constant, a quantity of change, and a description of the result. Naming which layer went wrong is most of the skill.
Reading the stack
Bottom to top, what each layer decides
The lower the layer, the more expensive it is to change your mind about — which is the opposite of the order most people set them in.
Source — what you started from
Resolution, aspect ratio and compression all limit everything above. A small or heavily compressed source constrains the result no matter what the rest of the stack does, and no later layer recovers it.
Mask — where the change is allowed
A whole-image transform and a masked one are different operations with different risk. Masking is the difference between “change the sky” and “change the picture, and hopefully mostly the sky”.
Constraint — what must survive
Edges, depth or pose can be held while everything else is redrawn. Without a constraint, composition is only preserved by accident and proportionally to how little you changed.
Strength — how much change
The dial that trades fidelity for freedom. It is blunt: it gives up everything at once, which is why the constraint layer beneath it exists.
Prompt — what you asked for
The top layer, and the one people reach for first when a result disappoints. If strength is low the prompt has almost no room to act, so rewriting it changes nothing and the real problem is two layers down.
When a transform comes back wrong, the instinct is to rewrite the prompt. It is usually the least effective change available. Work from the bottom: was the source good enough, was the region defined, was anything holding the structure, and only then was the description right. Most disappointing results are a missing mask or a missing constraint wearing the costume of a prompt problem.
Non-destructive
What the layer metaphor gets right, and where it breaks
Right: they are separable
Each decision can be changed independently, and thinking about them one at a time is how you debug a result rather than reroll it.
Wrong: it is not reversible
An editor's layers can be turned off. A generated transform is baked the moment it renders — there is no stack to go back to unless you kept the source and the settings.
So: keep the settings
Seed, strength and prompt are the only record of how a result happened. Tools that discard them make a good output unrepeatable, which matters the first time a client asks for a small change.
House rules
How this will be written
No invented numbers
No scores out of ten, no star ratings, no tier boards. Where something is better, it gets said in words and the reason is shown.
Only claims we can stand behind
Nothing describes a test that did not happen. Pricing and feature claims are checked against the vendor's own pages.
No paid placements
Nothing here is sponsored, and outbound links carry no affiliate tracking and earn no commission. No company pays to be included or to be written about more kindly.
This site is new and there are no articles on it yet. The panel above is an illustration of the metaphor — the blend modes and percentages in it are furniture, not settings from a real transform. When articles land they will state the settings behind any image they show.