What deterministic means here
Give Pacyra the same photograph, the same settings and the same engine version, and it returns exactly the same guide. The same regions, the same boundaries, the same palette, in the same order, on any machine. Painterly mode adds variation on purpose and takes a seed, so even the varied version repeats when the seed does.
That is a design decision rather than a limitation. The engine is a sequence of measurements over the pixels you gave it, so there is nothing in it that could produce a different answer on a Tuesday.
Why that matters at the easel
The first time you generate a guide, determinism is invisible. It starts mattering at the point where something goes wrong, which is always.
You spill tea on the printed guide halfway through a painting. You reprint it, and the reprint matches the twenty colours already mixed on your palette. You decide the picture wants to be bigger, regenerate at the new size, and the palette is the same palette. A friend asks for the settings you used, types them in, and gets your guide rather than a cousin of it.
None of that is possible if the tool answers differently each time it is asked. A generative model produces something plausible on each run, which is what makes it good at proposing and unsuitable for referring back to.
The test is simple. Generate a guide, close the tab, generate it again with the same settings, and put the two side by side. If they differ, it is not a reference.
What generative tools are genuinely better at
It would be dishonest to frame this as one approach beating another everywhere. A generative model does things this engine cannot and never will.
It can invent what is behind an object you want removed. It can change a grey sky to a dramatic one. It can produce ten variations of a composition to choose between. Those are real capabilities and if that is what you want, a deterministic segmentation engine is the wrong tool and will disappoint you.
What it cannot do is promise you the same answer twice, or tell you why it made a decision, or work with no connection on a photograph it never receives. The two approaches are solving different problems that happen to produce a superficially similar picture.
The other consequence
There is a second thing that follows from being a measurement rather than a model, and it is the one this whole site is built around.
A model of any size has to run somewhere, and for most tools that means the photograph goes to a server. A sequence of measurements over pixels runs wherever the pixels already are, which is your own browser. The determinism and the privacy come from the same architectural choice, which is why neither of them is a feature that could be added to the other kind of tool later.
Your picture stays on your device because the thing analysing it is small enough to come to the picture, rather than large enough that the picture has to go to it.