Why nothing here reshapes a face or a body

The controls a competing tool would list first are absent from this one on purpose. Not postponed, not reserved for a paid tier, not waiting on an engine — decided against, with a reason worth writing down.

A liquify mesh is not hard to build. Warping a triangulated grid under a brush is well understood, it runs fast, and dropping one into this tool would take a fraction of the effort the frequency separation took. Every commercial competitor in this category ships one, most of them put it on the front page, and it is reliably the control people reach for first.

It is not here, and it is not coming. No jaw slimming. No eye enlargement. No nose narrowing. No chin, no cheekbones, no waist, no height, no legs. At any setting, in any release.

The line, and why it is drawable

The distinction this tool runs on is between correcting the photograph and altering the person. It is not a hard philosophical line, but in practice it is far clearer than the debate around it suggests, and it survives almost every real case.

A red pupil is an artefact of the flash. A patch of shine is an artefact of a warm room. A spot is a state of that week rather than a property of that person. Sensor noise belongs to the camera. Every one of those is something the photograph reports incorrectly about the moment, and removing it makes the image more faithful to the person, not less. That is the entire set of things this tool does.

The width of a jaw is not an artefact of anything. It is the person. A control that narrows it is not correcting a photograph; it is producing an image of somebody who does not exist and circulating it as a picture of somebody who does. Every technical argument for shipping it — it is only a few percent, it is subtle, the user chooses the amount — applies exactly as well to any amount, which is the tell that there is no principled setting.

The harm is not hypothetical

The evidence on this is not new and it is not thin. Exposure to idealised and digitally altered images is associated with body dissatisfaction, and the association is measured most strongly in adolescents. Several jurisdictions now require altered commercial imagery of a body to be labelled as such — France since 2017, Norway since 2022, with similar proposals repeatedly before other legislatures — which is a legislative judgement that the alteration itself is the thing worth disclosing. Platform policies have moved in the same direction for the same reason.

We are not the authors of that research and are not going to pretend to a precision it does not have at the level of an individual photograph. What it supports is the modest claim that a free, frictionless body-reshaping tool with no disclosure attached is on the wrong side of a known problem. That is enough. The tool is not there.

There is a second consequence that is easier to state. Building a reshaping control means building a model of what a face should be shaped like. There is no neutral version of that model. Whatever direction the slider defaults toward is a claim about what is better, and those claims have historically travelled in one direction across features associated with ethnicity. The clean way not to encode it is not to build the control.

What we built instead

The effort that would have gone into a warp mesh went into things that are harder and less impressive in a before-and-after pair. A guided filter rather than a Gaussian, so the smoothing does not halo every lash and lip. A patch-match heal that penalises a candidate’s own texture, so a healed spot does not repeat a pore cluster two centimetres away. A skin gate that samples the median of the region it is in rather than comparing against a published table, because those tables were derived from light skin and a tool built on them quietly declines to include most of a dark face in its own mask. A chroma lock with a build gate behind it.

None of that will win a side-by-side against a one-press beauty filter judged at thumbnail size. All of it is what the difference between a retouched photograph and a good one actually consists of.

The honest objections

Lens distortion is a real thing and this rules out correcting it. True, and it is the strongest objection. A wide lens at close range genuinely enlarges a nose and narrows a jaw, and undoing that is correcting the photograph by the same test used above. The reason it is still not here is that the control required is indistinguishable from the one we declined to build — same mesh, same brush, same result if you keep going — and no interface separates the two reliably. A perspective correction derived from lens parameters rather than from a brush would be a different proposal, and it is not in this release.

People will use another tool. Some will, and that is fine. This site is not attempting to prevent anything; it has no way to and no standing to. It is deciding what it is willing to be the supplier of. Those are different questions and only the second one is ours.

You allow skin smoothing, which is also idealisation. At a high setting, yes — which is why it stops at 80, why it opens at 25, why a note appears above 60, and why there is a figure beside the picture the whole time. The distinction is that smoothing has a defensible range and a ceiling below the harm. Reshaping does not have a defensible range; it has a direction.