High-precision face reconstruction
Built on an improved StyleGAN2/3 architecture trained on large-scale face data, modelling skin tone, texture and facial structure. Stays stable in low light, motion blur and occlusion.
On-device · GAN beauty · Real-time · Cross-platform
StyleGAN-based AI beautification with 106 / 240 facial landmarks. Retouching, reshaping, micro-plasty, AR makeup, LUT filters, generative AI effects, animated stickers and virtual background in one SDK for iOS, Android, Web, Windows, macOS and Flutter. 30 FPS at 1080p on mid-range devices.

Face detection, skin retouching, reshaping, micro-plasty, makeup, filters, generative AI effects, animated stickers and virtual background all share one face engine and render pipeline. Enable modules per product, stack them freely, and expose every parameter to your own team.
Unlike filter-based retouching, Meishe GAN beauty uses generative adversarial networks to enhance detail while preserving the user's own facial features — high fidelity, high naturalness.

Built on an improved StyleGAN2/3 architecture trained on large-scale face data, modelling skin tone, texture and facial structure. Stays stable in low light, motion blur and occlusion.
Built-in semantic segmentation treats forehead, cheeks, nose and eye area differently: control shine on the T-zone while keeping lip texture, soften eye lines without blurring them.
Tracks lighting changes frame by frame so the look stays consistent across environments, with no flicker or jumping during long streaming or call sessions.
GAN beauty is available from SDK 3.15.0 and can be layered on the classic smoothing algorithms or used on its own. Validate the result on your own target devices.
Four smoothing algorithms to pick from per audience — from texture-preserving to silky — plus multi-scale texture retention and layered blemish removal. Also whitening, skin tone (cool, warm, pink, tan), blush, sharpening and shine removal limited to areas outside the facial features.
Eleven reshaping controls: slim face, narrow face, small face, forehead, chin, nose width, nose length, mouth, mouth corner, eye size, eyelid. Plus twelve micro-plasty items: head size, nose and eye refinement, bright eyes, nasolabial folds, dark circles, teeth, cheekbone, jaw and temples.
Nine makeup dimensions — lipstick, eyeliner, eyeshadow, contour, brows, blush, highlight, lashes, contact lenses — as single items or full looks. Every shade is adjustable, lips come in natural or glossy finish, and occlusion handles hand-to-face interaction.
Over 200 portrait filters plus custom 3DLUT and pnglut import with intensity control, and 14 adjustment dimensions including exposure, highlights, shadows, contrast, saturation, colour temperature, clarity, vignette and noise reduction. Authoring tools and training included.
Generative effects such as stylisation and AI portraits, on the same render pipeline as beautification, makeup and filters, processed on device.
Hair-level portrait segmentation for background replacement, blur and keying, running on device with no green screen and stackable with beautification.
2D and 3D stickers locked to facial landmarks, with mouth-open, blink and nod triggers plus gesture interaction; assets can be authored in-house. Avatars support 18 facial attributes, real-time expression driving and custom sculpting.
2D and 3D props, trigger effects, physics, 3D scenes and portrait segmentation, with a visual authoring tool so your team ships new effects without waiting on us.
Beauty templates bundle retouching, reshaping, micro-plasty and makeup into one-tap looks such as classic, natural, fair or male. A global skin-tone solution covers eight regions across nine levels, and smart beauty recommends settings by detected gender and face shape.
Measured on device, not in a slide. We can share a per-model benchmark report for your target devices before you commit to anything.
Runs fully on device with no cloud dependency, so no camera frame leaves the device and effects keep working offline. See the RTC section below for integration coverage.
Meishe sits as a render layer between capture and encoding, so beautification and effects finish before the frame is pushed. No need to replace your RTC provider or rework the pipeline.
Listed stacks are either pre-integrated or already running in customer projects. Anything not listed can be assessed against your existing capture and streaming pipeline.
These are the key points buyers repeatedly confirm during evaluation; final details depend on each vendor's actual offering.
| Comparison point | Meishe AI Beauty SDK | Typical alternative |
|---|---|---|
| Engine and output | Own face engine and render pipeline; region-aware processing keeps skin texture | Assembled third-party parts; flat global smoothing, limited control |
| Generative AI effects | Generative AI effects built in, so stylised looks ship with a trend instead of a roadmap | Fixed effect packs only; new looks wait on the vendor |
| Product range | Beyond the beauty SDK, Meishe also ships audio and video SDKs, so capture, editing and beauty can come from one vendor | Single-point beauty only; audio and video means another vendor |
| Stability and track record | Founded in 2014, 12 years in audio and video, with long-running deployments across platforms | Short company history; long-term maintenance unproven |
| Assets and roadmap | Authoring tools for effects and AR props, bring your own assets and brand props | Assets produced by the vendor; long lead times and unpredictable cost |
| Support and licensing | Named technical contact, fast response; license by app package or device | Ticket queues, slow turnaround, complex contract terms |
Different products care about different things: streaming cares about sustained frame rate, meetings care about privacy. Presets ship tuned per scenario.

Skin tone stays stable under mixed lighting, and frame rate holds through long sessions with no thermal throttling.

Shoot and export in one pass. Retouch, filter and sticker layers stack together without extra round trips.

Runs on device, so nothing is uploaded. Cleaner image with no lighting setup.

Hosts stay camera-ready while product colour, texture and detail pass through untouched.

Tutors look clear and natural, with background replacement that keeps a home setup looking professional.

Runs fully offline, so kiosks and fixed-location capture devices need no connection.
Most beauty SDK evaluations stall in security review. On-device processing removes the part reviewers push back on.
No camera frame is sent to a server, so there is nothing to store, transfer or leak.
Effects keep running with no connection, including kiosks and fixed-location devices.
Local-only processing shortens security review and DPA work under GDPR, CCPA and PIPL.
The risk in choosing a beauty SDK is not seeing it on your own hardware. We get the trial running first and talk commercial terms after.
Share your platforms and target devices. We send a trial SDK and license so you can measure on real hardware.
Per-platform demos, integration docs and preset parameter bundles for streaming, short video and calling.
Our engineers join the integration, tune performance on your device list and help with pre-launch stress tests.
Yes. Beautification, reshaping, makeup, filters, stickers and background separation can be combined module by module. Exact packaging is confirmed with sales.
On the device. Detection, beautification and effects run locally, no camera frame is uploaded, and the SDK keeps working without a connection.
Yes. Meishe drops into your existing capture and streaming pipeline as a render layer, with pre-integration samples for common RTC stacks.
There is no permanently free commercial version. You receive a free trial SDK and license to validate on your own devices; commercial licensing is per app package or per device, quoted per scenario.
Yes. Every parameter is exposed, you can supply your own LUTs and sticker assets, and custom looks or device-specific tuning are scoped as custom work.