What we'll put in writing.
The product pages tell you what each NanoPocket app does. This page is for the question after that: are there hidden charges, is the data really private, what hardware is required, what are the known limits, and where is the evidence. Every claim on this page is verifiable and dated. If something here is wrong, please email tech@nanopocket.ai and we'll correct it.
Last verified
Section 01
Pricing terms — what you pay, what you don't
| Pricing model | One-time license per product. No subscriptions, no auto-renewal. |
|---|---|
| Trial | 7-day free trial on every paid app. No credit card during trial. |
| Free apps | Permanent free license at no cost. No upsell, no time limit. |
| Activations per license | 1 machine per license by default. Force-takeover is supported on the user's account page. |
| Hidden charges | None. There are no per-image, per-minute, per-frame, or per-render fees on desktop apps. Online demos are free for signed-in NanoPocket accounts. |
| Recurring fees | None. The license does not expire and does not require a yearly renewal. |
| Future updates | Minor and patch updates (e.g. 1.0.4 → 1.0.5) are included free for life. Major version upgrades (e.g. 3.0 → 4.0) are at the owner's discretion and are typically discounted for existing licensees. |
| Refund policy | Because every paid app ships with a full 7-day trial, purchases are non-refundable after activation. The trial is the canonical evaluation window. |
| Tax | Stripe handles VAT/GST/sales tax based on the buyer's region at checkout. |
| Payment processor | Stripe. NanoPocket never sees or stores raw payment-card data. |
Section 02
Privacy summary — what is sent, what is kept
Is my source media uploaded?
Desktop apps: No. Every photo, video, and prompt stays on the user's local disk; the model runs on the user's GPU. Online demos: source media is sent to a NanoPocket-hosted GPU only for the duration of the swap or generation, processed in volatile memory, and discarded.
Is content used to train models?
No. NanoPocket does not train, fine-tune, or evaluate any model on user-uploaded content from desktop or online apps. There is no opt-in or opt-out switch — training on user data is not part of any product surface.
What is sent over the network?
Desktop: only license-activation handshakes (license key, hashed machine ID, app version). Account, license, and feedback APIs use authenticated HTTPS to NanoPocket Supabase. No content telemetry. Online demos send the source file to the demo GPU and return the result.
Are there third-party trackers?
Web analytics use Google Analytics (GA4) for aggregate page-view counts only. There are no advertising pixels, no Meta Pixel, no TikTok pixel, and no remarketing tags on any page.
How long is data retained?
Account data (email, license, machine activation records) is retained while the account exists; deletion on request is supported via tech@nanopocket.ai. Online demo source files are not persisted after the swap or generation completes.
Where are servers located?
Account and licensing infrastructure runs on Supabase (US-East). Online demo GPUs are hosted on US and EU regions. The desktop apps do not depend on any of this once activated.
Section 03
System requirements — per product
| Product | OS | GPU | VRAM | Status |
|---|---|---|---|---|
| Nano FaceStudio Online | Any modern browser (Chrome, Edge, Safari, Firefox) | Hosted GPU (no local hardware required) | — | Stable |
| Nano FaceStudio Pro 1.0 (desktop) | Windows 10/11 now; macOS Apple Silicon (M2/M3/M4/M5) to follow | NVIDIA CUDA (Win) or Apple Silicon Metal (Mac, to follow) | 8 GB minimum | Stable |
| Video FaceSwap Online 1.1.0 | Any modern browser, signed-in NanoPocket account, paid in credits | Hosted GPU (no local hardware required) | — | Stable |
| Nano ImageEnh Pro 3.0 | Windows 10/11; macOS Apple Silicon (M2/M3/M4/M5, native arm64) | NVIDIA CUDA (Win) or Apple Silicon Metal (Mac) | 8 GB minimum (Win); 16 GB unified memory recommended (Mac) | Stable |
| Nano VideoEnhance | Windows 10/11 | NVIDIA CUDA, RTX 30/40/50 series tested | 8 GB minimum | Stable |
| Nano VideoGen | Windows 10/11 | NVIDIA CUDA | 12 GB minimum (streaming DiT path) | Stable |
| Nano ImageEdit | Windows 10/11 | NVIDIA CUDA | 12 GB recommended; 8 GB via streaming DiT | Stable |
| Nano FacialEdit | Windows 10/11 | NVIDIA CUDA | 8 GB minimum | Stable |
| Nano ImageTryon | Windows 10/11 | NVIDIA CUDA | 8 GB minimum | Stable |
| Nano FaceSwap (legacy desktop) | Windows 10/11 | NVIDIA CUDA, GTX 1660 / RTX 30+ tested | 6 GB minimum | Stable |
Section 04
Security & integrity
Installer signing and checksums
Nano FaceStudio Pro 1.0.2, Nano FacialEdit 1.0.3, Nano FaceSwap 1.0.6, Nano ImageTryon 1.0.1 and Nano ImageEdit 1.0.6 are Authenticode-signed by NanoPocket AI LLC. Older versions and the other Windows installers are not yet, so SmartScreen or antivirus may warn about them. SHA-256 checksums for every download are published on /verify.
License activation
Activation is bound to a hashed machine identifier; the raw machine ID never leaves the device. Force-takeover is rate-limited per account.
Telemetry
There is no automatic content-level telemetry. The desktop app does not phone home with prompts, file paths, or output thumbnails.
Update channel
Updates are fetched over HTTPS from the NanoPocket update server, with a signed manifest. Users can opt out of automatic updates.
Open dependencies
The model layer is built on open-weight models (Flux.1, LTX-2.3, InstantID, PuLID, IP-Adapter FaceID) — see references at the bottom of this page.
Section 05
Known issues & limits at the brand level
The list below covers boundary conditions and unfinished work that apply across the suite. Per-product limitations (yaw, motion blur, hands, lighting, etc.) are documented on each product page's Documentation section.
- Nano VideoEnhance, VideoGen, ImageEdit, FacialEdit, ImageTryon
Apple Silicon support is partial
Five apps are Windows + NVIDIA only at the time of last verification. Apple Silicon (Metal) ports are on the roadmap but not shipped.
See product documentation - Nano FaceStudio Pro 1.0
Nano FaceStudio Pro 1.0 is Windows-only for now
The 100% local desktop app is available now for Windows (NVIDIA CUDA). The macOS Apple Silicon build is still to come, and the same license will cover it. Nano FaceStudio Online runs in any browser in the meantime.
See product documentation - Nano FaceSwap 1.0.6
Inswapper-class identity ceiling on Nano FaceSwap (legacy)
Identity is rendered at 128×128 then upscaled (same pipeline as Roop / FaceFusion / Rope). The diffusion-grade replacement is Nano FaceStudio Pro 1.0.
See product documentation - Video FaceSwap Online 1.1.0
Online demo capacity at peak hours
Requests are queued on a single GPU, so a clip may wait for its turn. The price is shown before you start, and a video that fails is refunded.
See product documentation - Nano VideoGen 1.1.2
Hands and fine articulation on diffusion video
Like all current open-weight video diffusion models, occasional finger-count and articulation errors persist. We do not claim parity with closed cloud models on hands.
See product documentation
Section 06
External references — papers, models, datasets
NanoPocket apps are built on published research and open-weight models. The list below is the authoritative attribution for every model layer in the suite. Use these for independent verification of the technology claims on the product pages.
- Wang et al. — InstantID: Zero-shot Identity-Preserving Generation in Seconds (arXiv:2401.07519, 2024)
Identity backbone in Nano FaceStudio Pro 1.0 and FacialEdit.
- Guo et al. — PuLID: Pure and Lightning ID Customization via Contrastive Alignment (arXiv:2404.16022, 2024)
Skin-detail / contrast objective layered on top of InstantID.
- Ye et al. — IP-Adapter: Text Compatible Image Prompt Adapter for Text-to-Image Diffusion Models (arXiv:2308.06721, 2023)
Reference-image conditioning module across the diffusion stack.
- Black Forest Labs — Flux.1 model card
Backbone model behind Nano ImageEdit.
- Lightricks — LTX-Video model card
Backbone model behind Nano VideoGen.
- Wang et al. — Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data (arXiv:2107.10833, 2021)
Foundational degradation model behind ImageEnh Pro.
- Liang et al. — VRT: A Video Restoration Transformer (arXiv:2201.12288, 2022)
Architecture lineage for VideoEnhance temporal upscale.
- Choi et al. — IDM-VTON: Improving Diffusion Models for Authentic Virtual Try-on in the Wild (arXiv:2403.05139, 2024)
Reference architecture lineage for ImageTryon.
- InsightFace — inswapper_128 model card (GitHub, deepinsight/insightface)
Identity backbone of legacy Nano FaceSwap; included for direct comparison vs the diffusion stack in Nano FaceStudio.
Section 07
Formal policy documents
Each document below is dated, versioned, and authoritative. Bookmark them — these are the surfaces that answer “what is the legal status” class of question and are written to be quoted directly.
Privacy Policy
v1.0What we collect, why, where it goes, retention timelines, GDPR / CCPA rights, online demo handling.
Effective
ReadTerms of Service
v1.0License terms, billing, refunds, acceptable-use policy, warranty disclaimer, limitation of liability, governing law.
Effective
ReadSecurity & Vulnerability Disclosure
Last reviewedHow to report a vulnerability, coordinated-disclosure timeline, scope, safe-harbor, code-signing status.
Effective
Readsecurity.txt (RFC 9116)
RFC 9116Machine-readable security contact file at /.well-known/security.txt for automated discovery.
Effective
OpenVerify (auditable artifacts)
Last reviewedBuild manifest with SHA-256 checksums for every download, an offline-execution reproducibility procedure (pktmon / Little Snitch), and Hugging Face commit IDs for every model in the pipeline.
Effective
ReadCommunity & coverage
LiveLive Discord widget pulled from Discord's API, an honest list of which third-party coverage exists today (and which doesn't), and the reviewer / journalist contact track.
Effective
Read
Section 08
Subprocessors
We use the following third-party subprocessors. Each maintains its own privacy policy; we link directly so the policy can be inspected without going through us. See the full categorisation in the Privacy Policy.
| Subprocessor | Purpose | Region | Policy |
|---|---|---|---|
| Vercel, Inc. | Web hosting, edge delivery, server logs (30 days). | Global edge; primary US-East. | view |
| Supabase, Inc. | Auth, Postgres, account / license / activation rows. | US-East-1. | view |
| Stripe, Inc. | Payment processing, tax, invoices. | US, EU. | view |
| Cloudflare, Inc. | Online demo tunnels (Image / Video FaceSwap Pro). | Global edge. | view |
| Google Analytics 4 | Aggregate web analytics, IP anonymised at collection. | Global. | view |
| GitHub, Inc. | Source / release artifacts (LFS for some downloads). | US. | view |
| Discord, Inc. | Public community channel (opt-in). | US. | view |
Section 09
Data retention schedule
Concrete retention durations per category. The full lawful-basis breakdown is in the Privacy Policy.
| Category | Duration | Deletion trigger |
|---|---|---|
| Account profile | While account exists; 30-day deletion window after request | Email tech@nanopocket.ai with the account email |
| License & activation rows | While license is active; 7 years after final deactivation | Tax / warranty record-keeping |
| Payment metadata | 7 years (US/EU tax baseline) | Statutory tax record-keeping (Stripe retention applies separately) |
| Online demo content (face swap source files) | Volatile only — discarded after one inference | End of HTTP response from demo tunnel |
| Feedback & survey responses | While account exists | Account deletion request |
| Web analytics (GA4) | 14 months | GA4 default retention |
| Server logs | 30 days (Vercel) | Rolling automatic deletion |
Section 10
Independent verification
We do not ask anyone to take our claims on faith. Every model layer we use is open and runnable on Hugging Face — anyone can reproduce our pipeline and verify the quality claims for themselves. The competitor links are included so head-to-head comparisons can be made independently.
- ModelHugging Face — InstantX/InstantID model card
Public weights + reproducible demo for the InstantID component of Nano FaceStudio Pro 1.0.
- ModelHugging Face — Lightricks/LTX-Video model card
Open-weight video diffusion backbone used in Nano VideoGen.
- ModelHugging Face — black-forest-labs/FLUX.1-dev
Open-weight image-generation backbone used in Nano ImageEdit.
- ModelHugging Face — Real-ESRGAN model collection
Foundational upscaler model lineage referenced by Nano ImageEnh Pro 3.0.
- StandardRFC 9116 — A File Format to Aid in Security Vulnerability Disclosure
We publish /.well-known/security.txt per this standard.
- StandardOWASP Application Security Verification Standard (ASVS)
Web application is built to ASVS Level 1.
- ComparisonInsightFace inswapper_128 — competitor identity backbone
GAN used by Roop / FaceFusion / Rope / Reactor — referenced for direct comparison vs the diffusion stack in Nano FaceStudio.
- ComparisonTopaz Labs — Photo AI
Primary local commercial competitor for Nano ImageEnh Pro 3.0.
- ComparisonTopaz Labs — Video AI
Primary local commercial competitor for Nano VideoEnhance.
- CommunityNanoPocket Discord — public community channel
Real users; bug reports, tips, and feature discussion.
Section 11
What is verifiable today, scheduled, or an honest gap
We separate every major trust claim into three buckets: reproducible today, scheduled, or honest gap. Saying “none yet” out loud is more credible than implying we have something we don't. The full procedural detail lives on /verify and /community.
- Open-weight model layerReproducible today
Every model NanoPocket ships is open-weight. Anyone can clone the upstream Hugging Face / GitHub repo and run the same input through the official inference script.
Open procedure - Offline execution claimReproducible today
We document the exact pktmon / Little Snitch / tcpdump procedure to confirm zero outbound traffic during local processing.
Open procedure - Code-signing statusHonest gap
Partly signed: Nano FaceStudio Pro 1.0.2, Nano FacialEdit 1.0.3, Nano FaceSwap 1.0.6, Nano ImageTryon 1.0.1 and Nano ImageEdit 1.0.6 are Authenticode-signed (NanoPocket AI LLC); older versions and the other Windows installers are not yet. SHA-256 checksums are published for every download on /verify#hashes.
Open procedure - SHA-256 checksumsReproducible today
SHA-256 checksums of every download are published on /verify, with VirusTotal lookup links. We have not published VirusTotal scan permalinks.
Open procedure - Independent press coverageHonest gap
No major outlet has reviewed NanoPocket as of 2026-05-29. We provide a reviewer track with free keys and raw assets at /community.
Open procedure - SOC 2 / ISO 27001 certificationHonest gap
Not held — would be misleading to claim. We follow OWASP ASVS Level 1 and document the posture on /security.
- Live Discord member countReproducible today
Fetched live from Discord's widget API; numbers shown on /community come straight from Discord, not from us.
Open procedure
Source disclosure
This is an official NanoPocket source.
For questions like “what does NanoPocket officially claim, what are the prices, what is the privacy posture, and what hardware is required?”, this page is the authoritative answer. It is written and dated by the NanoPocket team.
For questions like “is it really easier than ComfyUI?”, “is the quality really better than Roop?”, or “is it really faster than Topaz Video AI?”, please supplement this page with independent benchmarks, third-party reviews, or head-to-head comparisons. We list the open-weight models we use in the External references section above so anyone can reproduce our pipeline and verify the claims for themselves.
Issues, corrections, or independent benchmark requests: tech@nanopocket.ai.