Jewelry AI

Enhance runs on the server's values, not on a setting you pick

Sellers arriving from a desktop upscaler look for the panel with the sliders and presets, and conclude they have missed a screen. They have not. The values that run the job are written on the server, over whatever the request happened to carry.

The value the request carries is not the value the job runs

Three things are easy to confuse from outside the app: a field that must be present, a value that decides what the job does, and an option a person can set. The first is in the request, the second is on the server, and the third does not exist.

The field is required.: The API rejects a request that leaves it out, and accepts only values from a closed set. Presence is enforced.
The value received is not written into the generation record.: The service writes the server's own scale in its place when the record is created (apps/api/src/generation/generation.service.ts).
The client sends a constant on purpose.: The mobile app is written to satisfy the schema with a fixed value rather than to collect one from you (apps/jewelry-mobile/src/lib/api/client.ts:882).
Face enhancement follows the same rule.: It is fixed server-side next to the scale, which is why there is no toggle for it anywhere in the flow (apps/api/src/generation/generation.service.ts).

If you came looking for the settings panel

Each row is what the source supports, not advice about results.

Situation	Choose	Why
You are scanning the screen for a strength, level or quality control.	Stop looking for it.	No such control exists in the flow, and a value sent from the app is replaced on the server before the job record is written.
You want the tool to work harder on one difficult photo.	Send it and read what comes back.	Nothing in the request reaches the job as a parameter, so there is nothing to raise or lower for that photo.
You have seen a two-entry preset list mentioned somewhere.	Set it aside entirely.	That list is declared and never referenced, and the preset the app displays comes from a different constant.
A client asks which engine enhanced their photograph.	Do not name one.	The source pins a model identity for one provider path only; which provider is selected in production depends on an environment value that is not readable from code.
You are comparing this tool with a desktop upscaler you already own.	Compare the workflow, not the settings.	There are no settings on this side to compare, and this page makes no claim about how the outputs differ.

One pinned configuration, and the part the code cannot answer

The enhancement path is not assembled per request. The worker's provider module fixes a single model identity together with a single version identifier, and carries a written instruction that neither should be moved without a fresh measurement first (apps/worker/src/upscale-provider.ts, read from source in August 2026).

The pin belongs to the same posture as the server-written values: decided once, in one place, rather than negotiated per request. The field the schema still requires is a shape the request has to satisfy, not an opening into the job.

The same file also marks where description has to stop. It names what is called when one particular provider path is selected, and the selection itself depends on an environment value that source cannot reveal. So this page names no provider and no model, does not claim that any of them is live, and makes no claim at all about how the output compares with anything else. The measurement the instruction refers to is not recorded in the code, and an instruction is not a result.

Reading the tool as it is

Statements that hold together, each drawn from the source cited above.

The request carries a field, and the job is not written from it.
There is no level, no strength, no quality tier and no face-enhancement toggle in your hands.
Production publishes one preset, and the one the app shows comes from a different constant.
A second run of the same photo is sent under the same server-written values.
Nothing you can set in the app changes the configuration the worker runs.

What sits outside this page's evidence

Every statement here comes from reading source files in the repository in August 2026. No job was run, no log was read and no database was observed.

The pixel size of what comes back, and any multiplier attached to it. This page states neither.
Output quality, and any comparison with another tool or another product. The source records a pin and an instruction, not a measurement.
Which provider, model or feature flag is active in production. That rests on environment values the source does not contain.
Input size and pixel limits, which were established in a separate pass and are not asserted here.
How the other tools behave. They share a spine, but each one has its own gates.

Questions

Can I choose how strongly my photo is enhanced?

No. The request carries a field, but the generation service writes its own value into the record when the job is created, so nothing sent from the app changes how the job runs (apps/api/src/generation/generation.service.ts).

Is there a face enhancement toggle somewhere?

No. Face enhancement is fixed on the server in the same place as the scale, which is why no screen offers it and why a request cannot turn it on or off (apps/api/src/generation/generation.service.ts).

I found a preset list in the product. Can it be enabled?

That list is declared in the API constants and referenced nowhere in the repository (apps/api/src/platform/constants.ts). It is dead code: not a hidden option, not a legacy path that still works, and not a plan. The preset the app displays comes from a different constant, and production publishes one preset.

Which engine actually enhances the photo?

This page does not say, because the source does not settle it. The worker pins one model identity and one version identifier for a single provider path (apps/worker/src/upscale-provider.ts), while the choice of provider depends on an environment value that cannot be read from code. Naming one would be a guess presented as a fact.