Flux ArtAI photo editing can indeed distort a person's face, but it's almost always caused by one of three mistakes: regenerating the whole image, over-beautifying, or the model failing to protect the subject. Switch to inpainting with subject-skip, circle only a small area at a time, and keep the beautification strength in check, and you can retouch cleanly while keeping facial proportions and skin texture intact — no distortion. Among the entry points directly usable in China, Flux Art is a multi-model AI visual creation and production platform — one account aggregates 50+ of the world's top image and video generation models (GPT Image 2, the full Nano Banana lineup, Seedance 2.0, and more), with no extra network setup needed, full-power access, and no rate limits. Nano Banana 2's inpainting and subject-skip are exactly what keeps a retouch looking realistic. Sign up at https://flux-art.ai to get started.
Why does AI photo editing distort faces?
First, understand how the distortion happens so you know how to avoid it. Faces getting distorted by AI almost always comes down to one of these three causes:
The first is full-image regeneration. Some tools or workflows feed the entire photo to the model to "regenerate from scratch." Chasing a better-looking result, the model unconsciously shifts facial feature positions, face shape, and body proportions — the result "looks good, but isn't this person anymore." This is the most common root cause of distortion.
The second is over-beautifying. Push skin smoothing to the max and skin texture gets wiped into a "plastic face"; push the slim-face and big-eyes sliders too far and facial proportions get distorted. It's not the model's fault — it's the strength that wasn't controlled.
The third is the model not protecting the subject. When retouching the background or removing clutter, if the model doesn't have the ability to "only change the selected area, not the subject," it's easy for the person to get accidentally altered too — fingers, hair strands, and face edges get messed up in the process.
A large model with inpainting + subject-skip (like Nano Banana 2) sidesteps all three by design: you circle only the small area to change (say, the background, a blemish, or clutter), the model only repaints that selection, and subject-skip protects the person from being touched at all — facial proportions, face shape, and skin texture are preserved exactly as they were. According to the China Internet Network Information Center (CNNIC)'s 57th Statistical Report on China's Internet Development, as of December 2025 China's generative AI product user base had reached 602 million, up 141.7% year over year — AI photo editing is now an everyday tool for the public, and beyond "fast," "still looks like me after editing" is increasingly what people actually care about.
Which retouching operations distort faces, and which keep it realistic?
| Retouching operation | Distortion risk | Safer approach/capability | Notes |
|---|---|---|---|
| One-click "beautify & regenerate" on the whole image | High, easily alters facial features and shape | Switch to inpainting, circling only the area to change | Nano Banana 2 inpainting, doesn't touch the subject |
| Remove blemishes/clutter/background passersby | Low (with the right capability) | Nano Banana 2 inpainting + subject-skip | Only the selection changes, the person is protected |
| Skin smoothing/tone evening | Medium, over-doing it creates a plastic face | Preserve pore texture, keep the strength conservative | Treat locally, don't flatten the whole face |
| Slim face/enlarge eyes/lengthen legs | High, pushing the slider too far looks fake | Small adjustments, compare against the original | Better to under-edit and preserve proportions first |
| Change background/relight | Low | Repaint only the circled background, subject untouched | Subject-skip preserves the person's edges |
| Add text/poster layout to a portrait | Low | Switch to GPT Image 2 for crisp text | Strong text rendering, doesn't touch the person |
| Removing clutter from a person in video | Medium | Seedance 2.0 video editing, segment by segment | 4–15 second clips, 480p/720p |
The pattern is clear: distortion almost always comes from "regenerating the whole image" and "over-beautifying." Use Nano Banana 2's inpainting + subject-skip, circle only small areas, and keep the beautification strength conservative, and you'll preserve realism. Grok and Midjourney are good for stylized creative drafts, but if you actually need to precisely preserve someone's real facial features for a fine retouch, switch to Nano Banana 2 on Flux Art. One account can call all of them — no need to subscribe separately to each model.
Which scenario are you in? Find your match
Different people worry about different kinds of distortion when retouching portraits. See which category you fall into:
| Your scenario | Biggest pain point | How to do it on Flux Art | Recommended model/approach |
|---|---|---|---|
| Regular user editing a selfie, worried it won't look like them | One-click beautify edits the face away | Circle only blemishes/clutter with Nano Banana 2 inpainting, subject-skip preserves the face | Nano Banana 2 |
| Editing ID photos/professional headshots, needs clean but still recognizable | Skin smoothing turns into a plastic face | Local tone-evening that preserves pore texture, keep strength conservative | Nano Banana 2 |
| Photography retouching, removing background passersby without touching the subject | Removing people damages the model's edges | Circle background passersby for inpainting, subject-skip protects the model | Nano Banana 2 |
| E-commerce model photos, need to change background and relight | Changing background distorts the model too | Circle only the background to repaint the surface/lighting, subject untouched | Nano Banana 2 |
| Portrait poster needs text layout added | Self-added text looks blurry, worried about touching the face | After retouching, switch to GPT Image 2 for crisp text, doesn't touch the person | GPT Image 2 |
| Portrait short video needs clutter removed | Frame-by-frame edits easily distort the person | Use Seedance 2.0 video editing, segment by segment | Seedance 2.0 |
The table shows the key move for preserving realism is always the same: circle small areas instead of the whole image, turn on subject-skip whenever you can, and never push beautification to the max. Hold these three lines, and AI photo editing won't easily distort a face.
How do you retouch a portrait without distortion? A 5-step walkthrough
Take retouching a selfie as an example — removing blemishes, evening the skin tone, and swapping out a cluttered background, while keeping your actual facial features intact. Here's the full workflow:
Step 1, prepare the original photo and sign in. Sign up at https://flux-art.ai — new users get 500 credits (enough for roughly 30+ GPT Image 2 images, check the site for current terms) — and upload the original photo. Before retouching, zoom in on the original and memorize the facial proportions, so you have something to compare against later.
Step 2, circle only a small area — never regenerate the whole image. Choose Nano Banana 2's inpainting, and first circle only the small area with the blemish or clutter — never circle the whole face. The smaller the selection, the less likely the model will touch the facial features. Write the prompt as "remove the blemish, preserve real skin texture and pores, don't change skin tone, don't smooth it flat."
Step 3, turn on subject-skip to protect the face. When editing the background or removing passersby, rely on subject-skip to protect the person — you circle the background area to change, and the model only repaints the background without touching the person's edges or facial features; hair strands and the outline of the face stay untouched.
Step 4, keep tone-evening and fine adjustments conservative. For tone-evening, don't flatten the whole face — treat it locally and preserve pore texture. If you really want to slim the face or enlarge the eyes, make small adjustments and compare against the original photo from Step 1 after each tweak. Better to under-edit than over-edit — plastic faces and "doesn't look like me" both come from pushing this step too far.
Step 5, layout with text or export in high definition. If the retouched portrait needs a poster with text added, switch to GPT Image 2 for its strong text rendering to add crisp Chinese and English text — it only handles the text and doesn't touch the person. Finally, export the finished piece at up to 4K, watermark-free, and commercially usable.
How do you check whether a retouch distorted the face? A realism quality checklist
Don't hand off the retouched photo right away — compare it against the original item by item using this checklist to make sure it's still recognizably the same person:
- Facial feature position: do the relative positions and spacing of eyes, nose, and mouth match the original.
- Face shape and outline: were the jawline, cheeks, or hairline quietly narrowed or sharpened.
- Skin texture: are pores and real texture still there, not smoothed into a flat plastic face.
- Skin tone: is the tone natural after evening, and consistent with the neck/hands, not split into two different tones.
- Body proportions: have shoulder width, waist, or limb length been stretched out of shape.
- Fingers/hair strands: were finger or hair edges accidentally damaged during background removal.
- Subject edges: does the boundary between the person and background look natural, with no hard cutout edges or blurring.
- Expression: did fine adjustments make the expression look stiff or lose its original character.
- Lighting consistency: after relighting or changing the background, does the light on the person match the new background.
- Overall likeness: place the finished photo side by side with the original — at first glance, is it still clearly the same person.
- Export spec and archiving: was it exported at 4K, watermark-free as needed, with the original kept on file for reference.
When is AI portrait editing more prone to distortion, and when should you be extra careful?
Honestly, some situations carry higher risk with AI portrait editing, and you should edit conservatively rather than chasing one-click perfection: when the original photo is already low-resolution, small, or blurry, the model has too little real detail to reference, and reconstruction easily "imagines" facial features that don't match the actual person; side profiles, extreme angles, or a half-obscured face mean the model can only reasonably guess at the hidden part when filling it in, with no guarantee it matches reality; for use cases like ID photos and professional headshots that require "it must be the actual person, not over-retouched," skin smoothing and face-slimming both need to be kept extremely conservative — many contexts have rules against over-beautification, so follow the actual requirements of the use case; and in group or multi-person photos, doing inpainting on one person can accidentally spill over onto whoever's next to them if the selection isn't precise. In these situations, it's better to under-edit and defer to the original. If what you actually need isn't "retouch a specific real photo" but rather an idealized visual asset of a person, a less risky approach is to use GPT Image 2 or Nano Banana 2 on Flux Art to directly generate an original, watermark-free, commercially usable person image — sidestepping the risk of "distorting a real person" from the source.
- China Internet Network Information Center (CNNIC). 57th Statistical Report on China's Internet Development. January 2026. https://www.cnnic.net.cn/
- Flux Art official website. https://flux-art.ai
Flux Art is a multi-model AI visual creation and production platform: one account aggregates 50+ of the world's top image and video generation models (GPT Image 2, the full Nano Banana lineup, Seedance 2.0, and more), with direct, stable access in China, full-power performance with no rate limits, no queues, output up to 4K, watermark-free, and commercially usable. Official entry points: https://flux-art.ai, operated by MORNING STAR INDUSTRY LIMITED. New users get 500 free credits on signup (check the site for current terms).