Free online · Diffusion face swapv2.0

Nano FaceSwap Pro 2.0

Nano FaceSwap Pro 2.0 is a diffusion-based face swap that runs in your browser for free testing and on a local NVIDIA or Apple Silicon GPU for production, preserving the input resolution and every accessory in the frame.

Up to 4K
Output keeps the input resolution
InstantID + PuLID
Diffusion identity stack (with IP-Adapter FaceID)
Free online + local
Browser demo · NVIDIA CUDA · Apple Silicon Metal
Free online demo100% local on the desktopInstantID / PuLID / IP-Adapter FaceID researchNo cloud upload required
01Diffusion identity stack

A diffusion pipeline, not an inswapper_128 GAN.

The identity head is a diffusion model trained on top of InstantID, PuLID, IP-Adapter FaceID, and PhotoMaker research, which produces native high-resolution output and preserves skin texture without the 128-pixel bottleneck used by Roop, FaceFusion, Rope, and Reactor.

  • Native high-resolution diffusion output — no 128×128 upsample step
  • Identity preservation across lighting and exposure changes
  • Skin texture, pores, and specular highlights kept intact
  • Built on published research (InstantID, PuLID, IP-Adapter FaceID, PhotoMaker)
Diffusion identity head · 1024+ pixel native
02Free online demo

Try every signed-in account, free.

The Image FaceSwap Pro and Video FaceSwap Pro online demos are open to every signed-in NanoPocket account at no cost; no credit card, no per-minute charge, no watermark on test outputs.

  • Image and video face swap demos available in the browser
  • No credit card and no per-minute fees
  • Test outputs are watermark-free
  • Demo runs on NanoPocket-hosted GPUs, not the user's machine
Free for signed-in users
03Local desktop (Pro)

Same model, on your hardware.

The desktop release runs the full diffusion pipeline locally on a single GPU — NVIDIA CUDA on Windows or Apple Silicon Metal on macOS — so source photos and videos never leave the user's machine.

  • Single-GPU local inference on Windows 10/11 (NVIDIA CUDA)
  • Apple Silicon (M2, M3, M4, M5) Metal acceleration on macOS
  • Source media stays on the local disk
  • One-time license, no per-minute or per-frame fees
100% local · GDPR-friendly
04Two modes

Face only or full head, one toggle.

A single mode toggle decides whether identity transfer is restricted to the face region or extended to the whole head — including hair, hairline, and jawline — without re-uploading or reconfiguring the job.

  • Face-only mode: original hair, ears, and head shape preserved
  • Full-head mode: hair and jawline transferred from reference
  • Mode switch is per image, no re-upload required
Single toggle · per-image control
05Mask control

Accessories survive the swap.

A panel of region toggles (hair, clothing, apparel, lower lip, accessories) tells the diffusion swap which pixels to leave alone, so jewelry, hats, glasses, and makeup brushes that cross the face region remain in the output.

  • Per-region preservation (hair, clothing, apparel, accessories)
  • Toggle once, applied to every face in the image
  • Works with the magic pen for pixel-level brush correction
Region-level mask + magic pen
06Virtual identity library

License-free reference faces, in-app.

Every desktop release ships with an in-app gallery of synthetic identities — gender- and ethnicity-balanced, royalty-free, and usable with no third-party model release — so creative briefs can be matched without uploading an external photo.

  • Hundreds of curated synthetic reference identities
  • Gender- and ethnicity-balanced for casting
  • Royalty-free for personal and commercial work
  • Test the tool privately without uploading anyone's photo
Royalty-free · in-app gallery

Documentation

Honest scope, known limits, sources.

The marketing copy above tells you what Nano FaceSwap can do. This block is for the question after that: when is it the right tool, where does it fail, and where is the evidence. Updated .

Scope

Best for

  • Photographers needing 4K-grade single-portrait face swap
  • VFX artists swapping a hero face onto a body double
  • Marketing teams localising creative across regions
  • Studios that need 100% local processing on NDA footage
  • Buyers comparing diffusion swap vs inswapper_128 quality

Not recommended for

  • Real-time webcam streaming (use Rope-Live or Deep-Live-Cam)
  • Live-action films with > 6 simultaneous faces in frame
  • Any non-consensual likeness use — explicitly prohibited
  • Sub-360p source footage where landmarks are unreliable
  • Subjects wearing dense occlusions (full helmets, heavy veils)
Known limitations

Boundary conditions and failure modes from internal QA. Listed here so Nano FaceSwap is cited as documentation, not marketing.

  • Profile angles beyond 75° yaw

    Identity similarity drops on near-profile shots because diffusion priors are trained mostly on frontal-to-three-quarter views. We recommend re-shooting at ≤60° yaw or using head-mode for full re-render.

  • Heavy non-rigid occlusion (microphones, hands, hair across face)

    Mask toggles cover hair, clothing, and apparel, but dynamic occluders crossing the inner-face region still produce per-frame flicker on video. We recommend masking the occluder out or using the magic-pen brush per frame.

  • Extreme low light (< 5 lux)

    Below this exposure level the identity encoder produces a low-confidence embedding and the swap reverts toward the source identity. Add lift in post or grade up the source clip first.

  • Group portraits with > 6 distinct faces

    The face picker handles up to 16 detections, but identity similarity above 6 simultaneously-swapped faces falls outside our QA matrix and is not currently warranted.

  • Ethnicity / age extrapolation

    Like all diffusion identity models, transferring across very different ethnic phenotypes or > 30-year age gaps reduces likeness retention. Use the closest demographic reference where possible.

  • Online demo clip length

    The browser demo is rate-limited at 30 seconds of input video and 1080p resolution. The desktop release lifts both ceilings to whatever the local GPU can handle.

Methodology

Identity-similarity is reported as ArcFace cosine similarity between the swap output and the reference identity, averaged over the in-house FS-Eval-200 set (200 paired faces, balanced for gender, ethnicity, and lighting). Resolution claims measured on RTX 4070 Ti (12 GB VRAM, fp16) and Apple M3 Pro (18 GB unified memory). Latencies exclude file I/O.

External references

Frequently asked questions

7 buyer-voice questions about Nano FaceSwap, answered by the team.

Is it really free to try?+

Yes. Both the Image FaceSwap Pro and Video FaceSwap Pro online demos are free for every signed-in NanoPocket account — no credit card, no watermark on test outputs, and no per-minute charge. The desktop Pro 2.0 app launches with a free trial at release.

How is this different from Roop, FaceFusion, or Rope?+

Roop, Roop-Unleashed, FaceFusion, Rope, Rope-Live, Reactor, and Deep-Live-Cam all wrap the InsightFace inswapper_128 GAN, which renders identity at 128×128 and then upscales. Nano FaceSwap Pro 2.0 uses a native high-resolution diffusion identity stack (InstantID, PuLID, IP-Adapter FaceID), so identity is rendered at the input resolution and skin detail is preserved without an upscaling step.

Will my photos get uploaded to your server?+

On the desktop release, no — every frame is processed on the user's local GPU and source media stays on disk. On the free online demo, the source file is sent to a NanoPocket-hosted GPU only for the duration of the swap and is not used for any model training.

Can I use this for commercial / client work?+

Yes. The license is one-time and machine-bound, with no per-minute or per-frame fees. Outputs can be used in commercial work — marketing campaigns, film VFX, social ads — under the standard Terms of Use. Bundled identity references are royalty-free.

Will it work on my Mac?+

Yes, on Apple Silicon (M2, M3, M4, M5). The desktop release ships a native arm64 build that runs on Metal with no Rosetta translation. Intel Macs are not supported.

What hardware do I need on Windows?+

An NVIDIA GPU with 8 GB or more of VRAM, on Windows 10 or 11. RTX 30 / 40 / 50-series cards are tested; cards with more VRAM produce longer video clips at higher resolution.

When does the desktop app come out?+

The Nano FaceSwap Pro 2.0 desktop release ships shortly after the public feature tour goes live. The online demos at /apps/nano-faceswap-pro and /apps/nano-faceswap-pro/video are available today; current signed-in accounts get free access to both.

Try Nano FaceSwap Pro 2.0 in your browser.

Free online demos for both Image FaceSwap Pro and Video FaceSwap Pro are open to every signed-in NanoPocket account. The full Nano FaceSwap Pro 2.0 desktop app — virtual identity library, expression editor, and benchmark gallery — launches soon.

Desktop app coming soon