Fix the shine, the redness and the tired look in a portrait without wiping out the pores that make skin look like skin. Upload a photo, describe a light touch in one sentence, and compare the results at full size before you keep one.
Most skin retouching goes wrong in the same way: the blemishes disappear, and so does the face. Pores vanish, a cheek turns to wax, and anyone can tell the photo was edited. AI retouching can do the opposite job. You ask for a light touch in plain words, keep the texture that makes skin look like skin, and fix only the thing that bothered you in the first place.
Beauty filters and one-tap smoothing sliders work by blurring. They average neighboring pixels, so uneven tone gets softer, and pores, fine lines and tiny hairs go down with it. That is the plastic look. An image editing model works differently. It reads the whole photo, understands that this area is a cheek and that one is an eyelid, and redraws the part you point at according to your instruction. Because the instruction is a sentence, you can ask for the result you actually want: even out the redness, keep the pores.
Picasso IA runs 502 models behind one workspace, the toolkit, and the ones that matter here are image editors: they take your photo plus a short instruction and return a new version. One of them, Qwen Image Edit Plus LoRA Skin, is built for exactly this job, bringing out natural skin detail instead of removing it. General editors such as nano banana and Flux Kontext Pro handle targeted fixes like a shiny forehead or a tired under-eye. New accounts get free credits, so you can run one portrait through a couple of approaches before you spend anything, and the full catalog lists the rest.
Viewers do not judge a retouch by how clean the skin is. They judge it by whether it looks like a person. What reads as real is pores that are still visible up close, small shifts in tone between the forehead and the cheek, a few asymmetries, and highlights that sit on the skin instead of a haze laid over it. What reads as fake is a flat, even surface with no variation at all.
So the useful split is between temporary and permanent. Temporary things are fair game: a pimple, a scratch, a red patch after a workout, a shiny nose, a shadow under the eyes after a bad night. Permanent things are part of the face: freckles, moles, laugh lines, scars. Leave those unless their owner asks, and say so in the prompt. A good test is simple. If you would have to point out the retouch for someone to notice it, it worked. If the first thing a friend says is that the photo looks filtered, ask for less.
Wording does most of the steering. Describe the change and the limit in the same sentence, for example: reduce the shine on the forehead and nose, even out the redness on the cheeks, keep visible pores and freckles, natural look. Skip words like flawless, perfect or porcelain, because models take them literally and push toward exactly the wax finish you are trying to avoid.
No single model fits every portrait. This is how the options differ, and why many retouches end up using two of them.
| Approach | Best for | Watch out for |
|---|---|---|
| Texture-restoring editor (Qwen Image Edit Plus LoRA Skin) | Skin already over-smoothed by a phone beauty mode or an earlier edit | It restores overall texture, it does not hunt single blemishes |
| General editor with a precise prompt (nano banana, Flux Kontext Pro) | One-off fixes: a blemish, shine, redness, flyaway hairs | A vague prompt changes more than you asked for |
| Upscaler after the edit (Clarity Pro Upscaler, Crystal Upscaler) | Recovering sharpness once the retouch is finished | Sharpening can exaggerate texture you wanted subtle |
| Two passes, editor then texture model | Portraits with both a blemish problem and a plastic look | Every pass drifts a little from the original, so compare against the untouched file |
The workflow below keeps you in control of how much changes. It assumes one photo and a short working session.
Texture-focused models usually expose a strength setting, so if the first result feels too heavy, lower it and run again instead of piling on a second instruction. Keeping the seed fixed while you change one setting at a time tells you what each change did.
Skin retouching touches real faces, so a few rules come before any technique. Use only photos you own or have the rights to: your own, or someone else's with their clear agreement. Do not run a stranger's picture or a photo you found online through an editor. Faces are people, not material.
The second rule is about what the picture claims. Do not misrepresent people in ads or profiles. If the skin in a skincare or cosmetics ad is the product's promise, the image must not show a result the product did not produce. On a dating, ID or professional profile, the person who turns up should recognize themselves in the photo. Platforms and countries have their own advertising and identity rules, so check the current requirements of the platform you post on instead of assuming.
Retouching removes distractions. The moment an edit changes what a viewer would believe about a product's results or about how a real person looks, it has stopped being retouching.
The third rule is about body image. Heavy smoothing sets a standard that nobody has, and it adds up across thousands of images. When you retouch someone else, ask what they want fixed, show them the before and after, and lean toward less. For related work, see AI headshots and AI dating profile photos, where the same honesty rules apply.
This is where the approach stops being useful, and what to do about it:
A retouch also cannot fix a bad photo. Harsh flash, mixed light and heavy compression leave problems that no instruction removes cleanly.
Yes, if you ask for the right thing. Filters blur, which is what causes the plastic look. Image editing models redraw the area from your instruction, and a model like Qwen Image Edit Plus LoRA Skin is aimed at keeping pores and tonal variation instead of erasing them. You still have to check the result at full size, because a prompt that asks for flawless skin can push any model back toward wax.
It depends on the problem. For skin that is already too smooth, Qwen Image Edit Plus LoRA Skin is the one built for restoring natural detail. For a specific fix such as shine, redness or a blemish, general editors like nano banana and Flux Kontext Pro follow a pointed instruction well. Try the same photo with two of them and keep whichever result still looks like the person.
Say so in the instruction, in the same sentence as the change: reduce the shine on the forehead, keep visible pores and freckles. Avoid words like flawless or porcelain. If the first result is too heavy, lower the strength setting where the model offers one, or make a smaller second edit on the best version instead of starting over with a longer prompt.
Light retouching of a photo you have the rights to is ordinary practice, but the picture must not claim more than the truth. A skincare ad should not show results the product did not deliver, and a profile photo should still look like the person who shows up. Rules differ by platform and country, so check the current requirements where you publish before you post.
Only with their clear agreement, or if you hold the rights to the photo and the person has agreed to the edit. Do not upload a stranger's image, a photo you found online or a picture of a public figure. If you retouch a client or friend, show them the before and after, and let them decide how much change is too much.
Usually close, but not guaranteed. Editing models can shift small things like the shape of a nose or the age of a face, especially with long or vague prompts. Keep the instruction narrow, name what must stay, generate more than one version and compare each to the original at full size. If the likeness has drifted, pick another result or edit with less.
Fix the source first. If pores and fine detail were never captured, a retouching model has nothing real to restore and will invent texture instead, which can look off. Run the photo through an upscaler, or follow the blur repair route, then retouch the sharper version. An upscaler after the edit can also bring back crispness that the edit softened.
The models described on this page are image editors, so they work one photo at a time. The catalog also includes video editing models, but whether any of them handles skin the way you want has to be tested on a short clip, and results can vary from frame to frame. For a portrait that matters, retouching a still frame is the more predictable route.
When the person would not recognize themselves, or when the edit makes the skin look like nobody has ever looked. A practical limit: fix temporary distractions, keep permanent features, and stop when the photo looks natural at full size. Be especially careful with photos of other people, because heavy smoothing feeds an unrealistic standard of how skin should look.
Each generation costs credits, and the amount depends on the model and the settings you choose, so a number written here would go stale. New accounts start with free credits, enough to test a portrait with a couple of models. Current plans live on the pricing page, which is the only place to trust for numbers.
A portrait deserves a retouch you would be comfortable explaining. Open the Picasso IA toolkit, upload one photo you own, and compare a light touch against the original.
Choose your language
