Retouching People Responsibly: Skin, Hair, Teeth

Retouching a portrait used to be a purely aesthetic call: smooth this, brighten that, ship it. In at least one market it hasn't been a purely aesthetic call for a few years now — it's a compliance decision, with a specific visual trigger, a specific label, and a specific place that label has to sit. Most teams building a retouching preset have never read the actual rule, so they either over-comply (label everything, including edits that don't need it) or under-comply (skip the label on exactly the kind of change that does). Both are avoidable once you know where the line actually sits, and it's a narrower line than "any retouching" — it's specific to three things, and it splits skin, hair and teeth into three different answers.

The rule, precisely

Norway has required labelling of retouched advertising since 1 July 2022, under guidance from Forbrukertilsynet, the Norwegian Consumer Authority. The trigger, per the regulator's own guidance, is specific: an ad must be labelled when a body's shape, size or skin has been altered through retouching or other manipulation — and that includes a pre-capture filter or lens, not just post-processing. The rule doesn't apply retroactively; material published before the effective date doesn't need relabelling unless it's relaunched after it.

That's a narrower target than "any edited photo of a person," and the guidance is unusually concrete about where the boundary falls for each of the three things a retouching preset actually touches.

Skin: the category that's covered almost completely

Skin is where the rule is broadest. The regulator's examples read like a working list for a retouching preset's "will trigger a label" column: removing or adding wrinkles, freckles, moles or birthmarks; removing or adding scars, individual pimples, acne, cellulite, bruises, wounds or stretch marks. If a preset's job is "even out the skin," it is, almost by definition, doing something on this list. There isn't a meaningful skin-smoothing edit that sits outside the trigger — softness, blemish removal and texture changes are the core of what the rule was written for.

Hair: shape-changing edits trigger it, most color work doesn't get a pass by default either

Hair gets a specific, slightly counterintuitive example in the guidance: changing hair shape — making it fuller, or hiding a receding hairline or hair loss — is named directly as a triggering edit. That's a shape-of-the-body-adjacent change, not a color swatch, which is why it's called out on its own rather than folded into general skin retouching. A preset that adds volume or coverage to hair is doing the same category of thing as reshaping a jawline: changing what's actually there, not just how it's lit.

Teeth: the one place color and shape get different answers

Teeth are the cleanest illustration of how specific this rule is, because the guidance draws the color/shape line explicitly and the two sides land differently. Retouching tooth color — whitening — does not require a label. Retouching tooth shape or size does. Same body part, same preset category, opposite compliance answer, depending on which slider you actually moved. A preset that only ever whitens is on one side of the line; a preset that also straightens, lengthens or reshapes has crossed onto the other, whether or not the change was subtle.

What the label itself has to look like

Once an edit triggers the rule, the label isn't a free-text caption — it has specific, checkable dimensions. It must be sized to roughly 7% of the image area, use a version (light or dark) that contrasts against the background so it's actually visible rather than nominal, and sit in the upper-left corner beneath any filter or username marks already there (moving to another corner only if something else already occupies that spot). On video, it has to stay visible for the entire piece, not flash once at the start. Because the mark itself contains the word for advertisement, it also satisfies Norway's separate ad-identification requirement in the same stroke — one label doing two jobs.

A different platform, a more permissive default

It's worth knowing that this is a national advertising rule, not a platform policy, because the two don't line up. YouTube's synthetic-content disclosure guidance explicitly carves out the aesthetic end of this same territory: applying a beauty filter, or a color/lighting adjustment, is listed as something creators do not need to disclose, on the reasoning that "minor edits are ones that are primarily aesthetic, and don't alter the content in a way that could mislead the viewer about what actually happened." A beauty filter that smooths skin in a YouTube video needs no disclosure under YouTube's own rule. The same skin-smoothing effect, run on a paid ad served to a Norwegian audience, needs the label under Forbrukertilsynet's rule. Nothing about the pixels changed between those two cases — the market and the context did. "Responsible" retouching in 2026 means knowing which regime a given piece of content is actually operating under, not applying one global instinct everywhere.

For the broader landscape of what counts as synthetic or altered content and when it needs to be labelled at all, see synthetic media disclosure — the Norway rule is a narrow, retouching-specific instance of a much wider set of obligations that's growing rather than shrinking.

Versely walkthrough: a preset that knows the difference

The practical fix isn't a policy document nobody rereads before a shoot — it's a preset built around instruction, not adjustment sliders, because an instruction preset can say exactly what it does and doesn't touch. Versely's AI photo editor runs several instruction-editing models from the catalog on the same file, through generate_image_from_image, which takes a plain-language prompt describing the change alongside the source image_url. Two things about how it's built make it a reasonable base for a compliant preset rather than a liability:

1. It edits by instruction, so the instruction is the audit trail. A preset written as "even out skin tone and reduce visible blemishes; do not change face shape, body shape, or tooth shape or size" is legible — anyone can read it and know which side of the Norway line it's designed to sit on, before a single image runs through it. A stack of unlabeled slider values in an old desktop tool can't tell you that.
2. It supports a mask parameter for edits scoped to one region. Rather than an instruction that touches the whole frame, a mask lets a preset target only the area a brand has decided is in scope — the under-eye area for shadow reduction, say, without the instruction drifting into jaw or cheek shape the way a whole-frame prompt sometimes does.

The workflow: run the source photo through the instruction preset, then compare the result against the original side by side — Versely keeps both the source and the edited output rather than overwriting one with the other, which is the only reliable way to actually see what moved. If what changed is color, tone or blemish removal within the skin category, treat it as a Norway-labelling case for any ad served there. If what changed is shape or size — face, body or tooth — it's also a labelling case, full stop, regardless of how subtle it looks in the compare. If the only change was tooth color, it's the one specific case in this list that the guidance says doesn't need the mark.

Building three separate instruction presets — one for skin tone/blemish work, one for hair volume/coverage, one for teeth — rather than one do-everything retouch prompt makes this classification trivial after the fact: you already know which preset ran, and each preset maps directly to one row in the compliance table above. If a specific model in the roster keeps ignoring the "don't touch shape" clause of an instruction, the current best image editing model roundup is the fastest way to find one that holds masked edits more reliably before that model becomes the default in the preset.

The takeaway

"Responsible retouching" isn't a vibe, at least not in every market anymore — in Norway it's a specific rule with a specific label, and the rule cares about what kind of change was made more than how much. Skin edits are almost always in scope. Hair edits are in scope when they change shape, not just tone. Teeth are in scope for shape and size, explicitly not for color. Build the preset around that distinction — plain-language instructions, masked to the region actually in scope, checked against the original rather than trusted on sight — and the compliance question stops being a judgment call made per-image and becomes a property of which preset you reached for.