Back to Trust
Verify

Verify NanoPocket independently

This page is the auditable counterpart to our marketing surfaces. Every claim listed here is something a third party can check without taking our word for it: SHA-256 checksums on installers, code-signing fingerprints, VirusTotal scan links, a step-by-step procedure to confirm offline execution at the network layer, and Hugging Face commit IDs for every model in the pipeline.

Last reviewed

Honest status — what is and isn't verifiable today

We deliberately separate claim from evidence. The table below is honest about both. We would rather understate verifiability than overstate it.

  • Open-weight model layer is independently reproducible today (see §4).
  • Offline-execution claim is reproducible today using free tools (see §3).
  • Code-signing posture is reproducible today (run signtool verify /pa /v on Windows or codesign -dv --verbose on macOS).
  • SHA-256 checksums and VirusTotal permalinks are scheduled, not yet historically published. Starting with the next release of every product, the value will be published in this page within 24 hours of release. Older releases will be backfilled where the original artifact is still archived.
  • We have no independent press coverage and no third-party benchmark report published yet. See §5 for the explicit gap list.

Build manifest

Every shipping product, with the version it was last reviewed against, the platform target, the signing model, and the policy under which checksums and VirusTotal permalinks are published.

AppVersionPlatformReleaseSigningSHA-256 / VirusTotal
Nano ImageEnh Pro3.0.0Windows x64 + macOS Apple Silicon2026-04Authenticode (Win) + Apple Notarisation (Mac)

Published per release on /verify and in /release-notes/nano-imageenh-pro

Each installer is uploaded to VirusTotal at release; permalink shipped in this page within 24h of upload.

Nano VideoEnhance1.0.5Windows x642026-03Authenticode (Win)

Published per release on /verify

VirusTotal permalink published within 24h of release.

Nano VideoGen1.0.xWindows x642026-Q2Authenticode (Win)

Published per release on /verify

VirusTotal permalink published within 24h of release.

Nano ImageEdit1.0.5Windows x642026-03Authenticode (Win)

Published per release on /verify

VirusTotal permalink published within 24h of release.

Nano FaceSwap1.0.xWindows x642026-Q1Authenticode (Win)

Published per release on /verify

VirusTotal permalink published within 24h of release.

Nano FacialEdit1.0.xWindows x642026-Q1Authenticode (Win)

Published per release on /verify

VirusTotal permalink published within 24h of release.

Nano ImageTryon1.0.xWindows x642026-Q1Authenticode (Win)

Published per release on /verify

VirusTotal permalink published within 24h of release.

Code-signing & VirusTotal

We commit to the following posture for every public release:

  • Windows. Every installer is Authenticode-signed. Verify with signtool verify /pa /v <installer.exe>; the certificate's subject must match CN=NanoPocket (we will publish the exact subject line + thumbprint with the next release; please treat any installer signed by a different subject as untrusted).
  • macOS. Every DMG is signed with the NanoPocket Apple Developer ID and submitted to Apple notarisation. Verify with codesign -dv --verbose=4 NanoApp.app and spctl -a -vv NanoApp.app.
  • SHA-256. Compute with certutil -hashfile installer.exe SHA256 (Windows) or shasum -a 256 installer.dmg (macOS). The expected value is published on this page within 24 hours of release. Mismatch = the artifact is not the one we shipped; do not run it and please email security@nanopocket.ai.
  • VirusTotal. Every installer is uploaded to virustotal.com (70+ AV engines). The permalink is published on this page within 24 hours of release. VirusTotal's verdict is third-party — we cannot influence it.

Verify the “offline” claim yourself

“100% offline” is meaningless if you can't check it. Use any of the following standard tools to confirm zero outbound traffic during local processing.

Windows — pktmon (built into Windows 10/11)

# In an elevated PowerShell:
pktmon start --etw -p 0
# 1) Open the Nano app and run a full image / video export
# 2) Stop capture:
pktmon stop
pktmon etl2pcap PktMon.etl
# Open PktMon.pcap in Wireshark; filter:
#   ip.dst != 192.168.0.0/16 and ip.dst != 10.0.0.0/8 and ip.dst != 127.0.0.0/8
# Expected: 0 packets to public internet during processing
# (license activation + update check are separate, opt-outable network calls)

macOS — Little Snitch / Lulu / pfctl

# Option A — Little Snitch / Lulu (graphical):
#   Run the app, perform a full export, observe network connections panel.
#   Expected: zero outbound during processing.

# Option B — built-in pfctl + tcpdump:
sudo tcpdump -i any -n 'not net 192.168.0.0/16 and not net 10.0.0.0/8 and not net 127.0.0.0/8'
# Then run an export. Expected: no packet output during processing.

Network assertions per product

ProductExpected during processingDocumented exceptions
Nano ImageEnh Pro 3.0 (local)Zero outbound traffic during image processing.License activation (one-time HTTPS POST to license.nanopocket.ai) and update check (HTTPS GET, opt-out via Settings → Updates).
Nano VideoEnhance (local)Zero outbound traffic during video processing.Same activation + update check exceptions as above.
Nano VideoGen (local)Zero outbound traffic during generation.Same activation + update check exceptions as above.
Nano ImageEdit (local)Zero outbound traffic during edit.Same activation + update check exceptions as above.
Nano FaceSwap (legacy desktop)Zero outbound traffic during swap.Same activation + update check exceptions as above.
Image / Video FaceSwap Pro online demosOutbound HTTPS to the Cloudflare demo tunnel is required by design — these are explicitly online services, not local ones.n/a — this is the contract.

Model provenance — every weight, every upstream

Every model in the NanoPocket pipeline is open-weight. Anyone can clone the upstream repository, run the official inference script on the same input, and compare the output to ours. This is the strongest form of independent verification available today: we cannot fake which model produced an output if you have the original weights.

ComponentUsed byUpstreamLicenseHow to reproduce
Identity (face)
ip-adapter.bin + ControlNetModel		Nano FaceSwap Pro 2.0 (Image / Video)InstantX / InstantID Apache-2.0 (model card terms)git lfs clone the upstream repo + run the diffusers pipeline on the same input — output identity embedding distance should match within numerical noise.
Identity (face)
pulid_v1.x.bin		Nano FaceSwap Pro 2.0ToTheBeginning / PuLID Apache-2.0Clone PuLID and run the inference notebook on the same source / target pair.
Identity (face) — legacy
inswapper_128.onnx		Nano FaceSwap (desktop)InsightFace / inswapper_128 Non-commercial research (inswapper_128)Run the InsightFace face_swap example on the same crop and compare outputs.
Image generation
flux1-dev.safetensors		Nano ImageEditBlack Forest Labs / FLUX.1-dev FLUX.1-dev Non-Commercial License (community), FLUX.1-pro for commercialUse diffusers FluxPipeline with the same prompt + seed and compare to our output.
Video generation
ltx-video-2b.safetensors		Nano VideoGenLightricks / LTX-Video LTX-Video Open License (research / commercial-with-terms)Run the official LTX inference script with the same seed / fps / resolution and compare frame-level PSNR.
Image super-resolution
RealESRGAN_x4plus.pth		Nano ImageEnh Pro 3.0ai-forever / Real-ESRGAN BSD-3-ClauseRun the upstream Real-ESRGAN CLI on the same low-res input and compare PSNR/SSIM.
Image restoration
diffbir_v2.x.ckpt		Nano ImageEnh Pro 3.0Xinntao / DiffBIR Apache-2.0Use DiffBIR's reference inference script on the same degraded input.
Background matting
u2net.pth		Nano ImageEnh Pro 3.0xuebinqin / U²-Net Apache-2.0Run the reference U²-Net colab on the same image and compare alpha mattes.
Video super-resolution
vrt_x4.pth, basicvsrpp_x4.pth		Nano VideoEnhanceJingyunLiang / VRT + open-mmlab / BasicVSR++ Apache-2.0Run the VRT inference script on the same input clip and compare per-frame PSNR.
Optical flow (video stability)
raft-things.pth		Nano VideoEnhanceprinceton-vl / RAFT BSD-3-ClauseRun RAFT on the same frame pair and compare flow magnitude EPE.

What we do not yet have — explicit gap list

We list these openly because pretending they exist would be worse than acknowledging they don't.

  • Independent press coverage: none yet. No major outlet (The Verge, Engadget, Tom's Hardware, PCMag, Wirecutter) has reviewed NanoPocket as of the date at the top of this page.
  • Independent benchmark report: none yet. We have not commissioned or appeared in a third-party benchmark study (e.g. a VBench / GenEval / DAVIS-evaluation paper that names NanoPocket as a system).
  • SOC 2 / ISO 27001 audit: not in scope for our current operating size. We follow OWASP ASVS Level 1 and document our posture on /security, but we do not claim a formal certification we don't hold.
  • App-store presence: NanoPocket apps are distributed directly from nanopocket.ai. We are not currently in the Mac App Store or Microsoft Store; therefore there are no app-store privacy nutrition labels to cite. Apple notarisation is the closest equivalent we ship today.
  • User-volume disclosure: we do not publish download counts or active-user counts as marketing. The closest public signal is the Discord member count on /community, which is fetched live from Discord's widget API.

For reviewers, journalists, and academic researchers

If you are writing a review, a benchmark study, or a security analysis: we will cooperate fully. Specifically:

  • Free reviewer license keys for any of the paid apps — email press@nanopocket.ai with the publication and the apps you'd like to cover.
  • Raw demo footage / sample inputs / sample outputs on request, no NDA.
  • On-the-record interviews with the engineering team for technical follow-ups.
  • For security researchers specifically: see the Security & Vulnerability Disclosure policy (researcher safe-harbor included).