Cull where your photos live.

"Early Access" is about the product's maturity: the app is functional and we use it daily on real shoots, but it's young and moving fast — expect frequent updates, the occasional rough edge, and a responsive team if something breaks.

As an early buyer you get full Pro features for $99/yr, and that price is locked for good — it stays $99/yr even after regular pricing begins at release. What you're buying is an Annual License with Perpetual Use Rights. At release, pricing becomes two paid tiers:

• Standard — $99/year: the full culling & organizing toolkit.
• Pro — $179/year: everything, plus on-device AI, cloud / network drives, and the pro workflow.
• Loyalty renewal pricing — renewals are 40% off ($59/yr Standard, $107/yr Pro), and that rate is locked for as long as you keep renewing. If you let your license lapse for more than 30 days, renewing later starts again at full price.
• Updates during your active term — every release we ship while your license is active is included.
• Perpetual use rights — if you choose not to renew, you keep your last licensed version, working, indefinitely; you only stop receiving new updates.
• 14-day refund window, no questions asked, plus 3 device activations (desktop + laptop + backup).

Prefer to try before you buy? The 30-day free trial unlocks the full Pro feature set — no credit card, no account.

Photo Mechanic is the gold standard for fast metadata-driven culling — particularly for sports and news photographers with deep IPTC workflows. Cullstack focuses on a different problem: helping photographers cull across multiple storage systems without consolidating to local first. We work with NAS, SFTP, Dropbox, OneDrive, Google Photos, iCloud, and your phone in one window.

Yes — via Live Ingest. Point your camera's tether software, a Wi-Fi/FTP card, or an SD auto-import at a watch folder, and Cullstack imports each new frame the moment it finishes writing — auto-renamed from a template, stamped with your IPTC creator and copyright, and jumped to in the grid so you cull as you shoot. It's the same hot-folder live-ingest workflow Photo Mechanic pioneered, with write-completion detection so a half-transferred frame is never imported truncated.

Cullstack doesn't drive the camera itself (firing the shutter or changing ISO/aperture from the computer) — that's full tethered capture, which Capture One and Lightroom handle. For the cull-as-you-shoot workflow most event, studio, and news photographers actually use, Live Ingest covers it.

Yes — Cullstack ships on-device AI helpers: semantic image search ("sunset over water"), face detection, closed-eye flagging, NSFW flagging, and AI auto-stacking by visual similarity, capture-time bursts, or face identity (with optional ArcFace / MobileFaceNet models).

What Cullstack doesn't do is pick your keepers for you. Aftershoot's lane is "give me the AI's picks and I'll refine"; Cullstack's lane is "give me AI hints + the fastest possible cockpit to drive my own cull." If you want the machine to decide, Aftershoot's a better fit. If you want machine-assisted speed but keep the final eye on every frame, that's us.

All models run locally — no photos ever leave your machine, no Cullstack server in the loop. Compatible with the strictest client NDAs.

Yes — there's an AI filter in the grid filter bar with three modes: Any, AI only, and Non-AI only. Photos detected as AI also get a magenta ✦ AI badge in the top-right of their grid cell. Useful for separating client deliverables from moodboard inspiration, auditing a delivery folder before sending, or tagging stock-license-clean photos.

Detection runs in six layers, fastest first, and stops on the first match:

1. Embedded provenance metadata — A1111, Forge, Fooocus, ComfyUI PNGs; A1111 JPEGs/WebPs; Midjourney JPEGs all leave their own tags, and Cullstack reads them.
2. Filename pattern match — DALL-E, ChatGPT, Stable Diffusion, Midjourney, Flux, Ideogram, OpenArt, Leonardo, Playground AI, Adobe Firefly, Bing Create, Runway, Imagen, StarryAI, NightCafe + more.
3. Parent folder name — folders called AI, Generated, Renders, etc.
4. C2PA Content Credentials — catches ChatGPT, DALL·E 3, Adobe Firefly, Bing Create, and anything else signed with the C2PA industry standard.
5. EXIF camera-field shape — real photos carry camera Make, Model, ExposureTime, ISO; AI photos almost never do. Combined with format/dimension heuristics (PNG, or sides divisible by 64 ≤ 2048).
6. ONNX visual classifier (optional, described below).

Tiers 1–5 are built in and run automatically. In testing across mixed libraries they catch around 90 % of AI photos on their own.

Cullstack ships without bundled AI models — each feature is opt-in, and you choose the model that fits your workflow (size, accuracy, licence). All models live in ~/.cullstack/models/. Drop the right .onnx file in with the exact filename below and the feature lights up next launch.

Important for commercial users: Cullstack itself is commercial software. Match each model's licence before installing — permissive licences (Apache 2.0, MIT, CC-BY-4.0) allow commercial use with attribution. NonCommercial (CC-BY-NC) and strong copyleft (AGPL) licences can be problematic for paid users running professional photography businesses. When in doubt, check with the model author or a lawyer.

All inference runs on your machine via ONNX Runtime — no photos or model outputs leave the computer. Each feature no-ops gracefully if its model isn't installed, so you can try one at a time without disrupting the rest of the app.

Pick a photo of someone — select it in the grid, or choose a file — and Tools → AI → Find this person filters the grid to their other shots. It matches on facial identity (the geometry of the face), not clothing, pose, or background, so it follows a person across outfit and scene changes. It does need a visible, detectable face: a shot from behind, or with the face turned fully away, won't match. (Requires the two face models listed in the table above.)

The Match slider runs from Loose to Strict. It mostly controls how many true shots of the person you keep, not how many wrong faces sneak in — different people score far below the same person, so mistaken matches are rare across the whole range:

• Strict (right). Only the most confident matches. Cleanest results, but you may miss harder shots (sharp profiles, poor light, motion blur).
• Loose (left). Casts a wider net and pulls in more of their shots, including tough angles — with little risk of look-alikes, since strangers sit well outside the range.

A good workflow: leave it at the balanced default, then nudge it toward Loose if a few shots you expected are missing, or toward Strict if you want only the surest matches. Hit Clear to drop the filter.

The visual classifier is opt-in because Cullstack doesn't ship with a model bundled — model licenses vary, and we'd rather you pick the one that fits your workflow than make that decision for you. Two minutes to install:

Most AI-detection models on HuggingFace ship as PyTorch only — you'll convert to ONNX once with one command, then install the resulting .onnx file in Cullstack. Five-minute setup, no Python needed afterwards.

1. Pick a model. Recommended for commercial use:
   • umm-maybe/AI-image-detector — Vision Transformer (ViT), trained on a mix of AI generators including SDXL, Midjourney v5+, and DALL·E variants. Licence CC-BY-4.0: commercial use allowed with attribution. ~348 MB before ONNX conversion shrinks it.
   Both dima806/ai_vs_real_image_detection and Organika/sdxl-detector look attractive but check their Files tab before investing time — at time of writing the first has no pre-built ONNX (PyTorch only) and the second is CC-BY-NC-3.0 (non-commercial only, not compatible with paid Cullstack use).
2. Convert to ONNX with HuggingFace's optimum toolkit. One-time Python setup, then one command per model:

   # One-time
   pip install "optimum[onnxruntime]" transformers
   
   # Convert
   optimum-cli export onnx \
       --model umm-maybe/AI-image-detector \
       --task image-classification \
       ~/Downloads/ai-detector-export/
   
   # Move into Cullstack's models folder
   mkdir -p ~/.cullstack/models
   mv ~/Downloads/ai-detector-export/model.onnx \
      ~/.cullstack/models/ai-detector.onnx

   optimum-cli downloads the PyTorch weights from HuggingFace, traces them through a representative input, and emits an equivalent ONNX. The resulting file works with the same AIImageDetector code in Cullstack.
3. (Optional) Verify in Cullstack. Open Settings → AI and confirm "AI image detector model: Installed". If you'd rather use the file picker than the command line in step 2, run optimum-cli with a different output folder and then in Cullstack click Install ONNX model… to copy the file in.
4. Each scanned file is classified once and the verdict is cached, so the new AI badges appear on previously-uncaught photos immediately and stay there across restarts.

Compatibility: any 3×224×224 ONNX image classifier with 1-output sigmoid, 2-class softmax (real / AI), or N-class softmax (multiple AI generators + real) will work — Cullstack auto-detects the output shape.

License compliance: the model's licence is between you and the model author when you download from HuggingFace. Check the licence on the model's HuggingFace page before installing — Cullstack is commercial software ($99/yr Standard, $179/yr Pro), so models with a NonCommercial clause (e.g. CC-BY-NC) are off-limits for paid users running professional photography businesses. Permissive licences (Apache 2.0, MIT, CC-BY-4.0) allow commercial use with attribution and are the safe pick. Inference runs entirely on your machine — no photos or detection results leave your computer.

Yes. Cullstack's AI image detector is model-agnostic — whatever .onnx file lives at ~/.cullstack/models/ai-detector.onnx is what Cullstack uses. The shipped install flow (Settings → AI → "Install ONNX model…") copies any compatible model into that slot, so swapping in a different detector is a 5-second operation. No code change in Cullstack, no support ticket, no reinstall.

Compatibility envelope: any ONNX image classifier that accepts a 3×224×224 RGB tensor and outputs 1, 2, or N classes is supported automatically — Cullstack inspects the model's input/output shapes at load time and adapts. Most modern transformer + CNN classifiers fit this envelope out of the box (ViT, Swin, ConvNeXt, ResNet, EfficientNet, MobileNet, etc.).

Common substitution paths and who they're for:

• Free HuggingFace models — covered in the "How do I enable the visual AI classifier" FAQ above. Best starting point for most users.
• Commercial detectors with budget — services like Hive AI, Reality Defender, Sensity, etc. sell paid AI- detection models. If they offer an ONNX or PyTorch export (ask their sales team — many will export for self-hosting customers), Cullstack runs the inference locally with the same install flow. Useful for studios that need higher accuracy than open-source models provide, or that need contractual SLAs on detection quality.
• Generator-specific detectors — fine-tuned for one tool (Midjourney-only, SDXL-only, deepfake-specific face detectors). Useful when your library contains a predictable AI source — e.g. concept-art studios reviewing junior submissions against a known Midjourney pipeline. Search HuggingFace for keywords like midjourney detector, sdxl detector, deepfake detection.
• Custom-trained models — ML engineers and large studios sometimes train their own detectors on proprietary data (their own historical archive vs. their AI tool's output, for example). Export from PyTorch / TensorFlow with torch.onnx.export or tf2onnx, drop into Cullstack. The model never touches Cullstack's servers (we don't have any AI-related servers).
• Multiple models, one workflow — Cullstack only loads one detector at a time, but you can swap them in and out instantly. A wedding photographer might run a general classifier on import day, then swap to a Midjourney- specific one when auditing a concept-art folder. The per-file verdict cache is keyed to file path + mtime, so the old verdicts are kept; only files that haven't been scanned yet by the new model get re-classified.

Power-user override: the visual classifier's confidence threshold is fixed at 0.85 in the UI to prevent casual false-positive floods, but the underlying preference (ai.threshold.aiImage, integer percent 0–100) is readable via the standard Java Preferences API. Power users on macOS can set it via:

defaults write com.apple.java.util.prefs \
    /com/aviastone/cullstack/util/AppPreferences/ai.threshold.aiImage 75

Confidence floors below ~0.70 substantially increase false- positive risk on real photography — recommended only for researchers comparing classifier behaviour, not for production photo culling.

What Cullstack doesn't do: validate model accuracy, vouch for vendors, or take responsibility for misclassification consequences. The model is your choice, the license is between you and the model author, and the downstream decisions (delivering, flagging, deleting) are always your call. Cullstack is the tool; you are the photographer.

All connections live in one place: open the Connection → Manage Connections… menu. Click + Add, pick the backend, fill in the fields, save, then Connect. Each profile becomes its own branch in the sidebar tree (☁ WebDAV, 🔒 SFTP, 📦 Dropbox, ☁ OneDrive) and you can browse, copy, move, and stack across them the same as a local folder.

• Dropbox — one-click OAuth. Choose Dropbox, give the profile a label, click Authorize, approve in your browser. Tokens stay on your machine; we never see them.
• OneDrive — same OAuth flow as Dropbox. Choose OneDrive → Authorize → sign in with Microsoft. Works for personal accounts and Microsoft 365.
• WebDAV — server URL (https://… or http://…), username, password. Toggle Bypass SSL check only if you're connecting to a self-signed-cert NAS on your own LAN. Works with Nextcloud, ownCloud, Synology WebDAV, Box (with WebDAV enabled), and any standards-compliant server.
• SFTP — host, port (default 22), username, and either a password or a private key file. Key paths support both OpenSSH (~/.ssh/id_ed25519) and PuTTY (.ppk) formats. Optional passphrase unlocks encrypted keys.
• iCloud — no setup needed. iCloud Drive appears as a normal folder under your macOS home directory (~/Library/Mobile Documents/com~apple~CloudDocs) — just navigate to it in the sidebar tree like any other local folder. iCloud Photos surfaces automatically on macOS as the iCloud Photos sidebar entry once you grant Photos library access on first launch.

Credentials are kept locally in the macOS Keychain (or the OS-equivalent secure store on other platforms). Cullstack never relays them through any Aviastone server.

Plain drag-and-drop in Cullstack copies by default — the safer choice for photo workflows, since the original always stays put. Hold Shift (or ⌘ on macOS / Ctrl on Windows) at the start of the drag to move instead. This applies to every backend: local folders, network drives (WebDAV/SFTP), and cloud (Dropbox/OneDrive). External drags in from Finder or Explorer follow the OS's own copy-vs-move conventions.

Cullstack reads your photos from wherever they already live — we never copy them to a Cullstack server. Ratings, tags, and collections are stored locally in a SQLite sidecar database next to your photos. No telemetry, no cloud sync (unless you opt into syncing the sidecar yourself via Dropbox/iCloud Drive).

No. Cullstack reads and writes your photos where they already live — local drives, your NAS, your own SFTP / WebDAV server, or the cloud accounts you authenticate. Nothing is uploaded to a Cullstack-controlled server. This is a contractual win for wedding and portrait photographers whose NDAs prohibit transmitting images to third-party services.

Cullstack is an annual license with perpetual use rights — the version you've licensed keeps working forever, even if you stop renewing. Activation is verified at purchase and re-checked occasionally for refund/fraud protection, but the app continues to function with a long offline grace window. We will publish a final unlock build if we ever wind down the product.

Yes. Each license includes 3 device activations. You can deactivate from a machine you're retiring (Help → Manage License → Deactivate) and free up a slot for a new install. If you've run out of slots due to lost devices, email support@aviastone.com and we'll reset them.

Yes — 30% off for verified students, educators, and registered non-profits. Email sales@aviastone.com with proof of status.