Stack Pix

Node-based AI photo studio that runs entirely on your PC

Node-based AI photo studio for Windows

Drag nodes, wire them up, run the result on thousands of photos. 53 node types and 81 ready-made workflows — all inference happens on your own machine.

Stack Pix Quick Start: 81 ready-made workflows grouped by category (ID & certificate, e-commerce, lifestyle, social media, avatar, poster, food, document, real estate, photography, creative 3D)
Quick Start — 81 ready-made workflows, grouped by the job you are doing.

Local processing only — your images are never uploaded, and no account is required.

WHY IT FEELS DIFFERENT

Pipelines, not presets

Chain any of the 53 node types into a graph that matches how you actually work. Save it once, run it forever.

Built for thousands of files

Feed a folder, group it any way you like, and walk away. Interrupted jobs resume where they stopped.

Recompute only what changed

Node-level caching keeps finished results and re-runs just the branch you edited. Deep models stay loaded.

Private by architecture

No cloud queue, no upload step, no per-image cost. The model runs next to your files.

CORE CAPABILITIES

/ 01 — CANVAS

Visual pipeline builder

A ComfyUI-style canvas, but built for photo work: every node has a form, a preview and a summary.

  • · Drag, connect, box-select, copy, undo/redo, minimap and grid
  • · Condition nodes branch on width, height or file size, with automatic dead-branch skipping
  • · Merge, switch, reroute, comment and subgraph nodes keep large graphs readable
  • · Live preview: change a parameter, see the result without re-running the whole graph
  • · Open as many workflow windows as you need — they share one engine
Canvas editor with two background-removal nodes wired together (BiRefNet Portrait and RMBG-2.0), a Start badge on the first node, GPU and parameter controls in the toolbar, and the node tip bar at the bottom
The canvas: two background-removal nodes wired into one graph — pick the model per node, watch the GPU readout, and the graph itself is the recipe.

/ 02 — BATCH

Batch, grouping and output control

The batch semantics are the point: an entire folder is one job, and grouping decides how results come out.

  • · Array semantics: one input set fans out through map nodes, then folds back with reduce nodes
  • · Group by folder, filename prefix, fixed count — or let AI cluster faces so every person in a group shot gets their own set
  • · 1→N fan-out (grid slicing, stereo views) and N→1 reduction (collage, best group photo, ID-photo layout)
  • · Resume by fingerprint (path + size + mtime): unchanged files are skipped, finished work is kept
  • · Pause, resume and stop between images and between nodes
  • · Name templates with {index}, {date}, {group}, {width}… and automatic subfolders
Running a template: import images on the left, the five-step pipeline on the right (background removal, condition, upscale, background set, compress), its parameters below and the Run button
One folder is one job: open a ready-made workflow and its five steps (cut out → condition → upscale → set background → compress) run over the whole batch.

/ 03 — AI

AI that removes the boring part

The unglamorous 80% of photo work — cutting out, cleaning up, sorting a shoot — done locally and in bulk.

  • · Background removal: RMBG-2.0, BiRefNet (general & portrait) and MODNet, plus object extraction and AI backgrounds
  • · AI culling: flags exposure problems, missed focus, closed eyes and near-duplicates, then splits keepers from rejects
  • · Best group photo: fuses a burst into one frame, picking each person’s best expression
  • · Watermark removal, smart erase, fingerprints and dust cleanup
  • · Portrait retouch, makeup, face restore, old-photo restore and document scanning
Settings — AI Models: each model listed with its size, Apache-2.0 licence, hardware compatibility and download state (inference engine, RMBG-2.0, RMBG-1.4, BiRefNet)
Models live on your own disk: sizes, licences and download state are all in the app — no cloud queue, no per-image cost.
Settings — AI Assistant: connect any OpenAI-compatible endpoint (including a local LLM such as LM Studio or Ollama) and see usage statistics
Optional AI Create: point it at any OpenAI-compatible endpoint — including a local LLM — and describe the workflow in words.

/ 04 — DEPTH

Depth-based effects

One depth estimate powers a whole family of effects that usually need a 3D pipeline.

  • · Depth haze, depth-of-field blur and re-lighting computed from real depth
  • · Stereo pair: 1→N views for 3D viewers, generated by forward-scatter reprojection
  • · Anaglyph red/cyan output for classic 3D glasses
  • · Wiggle GIF: a looping parallax animation with smooth easing
Creative 3D workflow library in Stack Pix: Depth Bokeh Portrait, Depth Wiggle GIF, Stereo View Pair, Red-Blue 3D Anaglyph, AI Depth Map / Fog Mask and Depth Lighting, all tagged Depth Pro
The same depth estimate, six ready-made outputs: bokeh, wiggle GIF, stereo pair, anaglyph, depth map and re-lighting.

/ 05 — OUTPUT

Everyday editing and delivery

The unexciting controls you need on every job, in the same graph as the AI steps.

  • · 17-parameter adjustments, filters, curves, colour and HSL tools
  • · Crop, perspective correction, transform and smart aspect presets
  • · Grid slicing, collages, multi-copy ID-photo layouts
  • · Borders, sticker outlines, blur, halftone, selective recolour, privacy mosaic
  • · Text, QR-code and image overlays
  • · Compression to MozJPEG / WebP / AVIF / JXL — animated GIFs are written through untouched
E-commerce workflow library in Stack Pix: Product Color Swap, Main Image Tile Split, Anti-Theft Watermark Overlay, Etsy Lifestyle, Amazon / Taobao / Pinduoduo / eBay / Shopify / Shopee main-image presets, Transparent Product PNG and DSLR Macro Depth of Field
Output is part of the pipeline: marketplace main-image presets (Amazon, Taobao, Pinduoduo, eBay, Shopify, Shopee, Etsy) live in the same graph as the AI steps.

/ 06 — APPEARANCE

Make the workbench yours

39 curated colour palettes ship with the app, grouped as Geek, Monochrome, Dark, Light, Festive and Creative & Mood. Switching one re-skins the whole workbench — your images and results are untouched either way.

  • · Dark sets: Midnight Blue, Abyss Ocean, Cyberpunk Neon, Nord Polar, Graphite Gray, Darkroom Gold
  • · Light sets: Pure Mono Light, Press Newspaper, Rice Paper, Sakura Pink, Sky Blue, Cream Warm, Fresh Mint
  • · Geek & retro: Matrix Code, Terminal Cyan, Amber CRT, Linux Bash, Neon Synth
  • · Festive: Spring Festival, Christmas Green, Halloween Night, Mid-Autumn Gold
  • · Every palette is shown as a preview swatch in Settings → Appearance before you switch
Settings → Appearance in Stack Pix: a grid of 39 colour palette previews with Midnight Blue selected, filtered by All, Geek, Monochrome, Dark, Light, Festive and Creative & Mood
Settings → Appearance: 39 curated palettes with preview swatches — Midnight Blue, Abyss Ocean, Terminal Cyan, Press Newspaper, Sakura Pink and more.

ALL NODES

Every node you can build with

Custom workflows are built from 53 node types. These are all of them, grouped the way the canvas sidebar groups them — name and one line each, so you can see the whole toolbox before you download.

MATTE / BACKGROUND

Background Remove
Cut the subject out with RMBG-2.0, RMBG-1.4 or BiRefNet, with optional auto-crop to the subject.
Object Extract
Pull out just the part you describe in a text prompt, as its own layer.
Background Set
Put something else behind the subject: transparent, white, a solid colour, a gradient or your own image.
Smart Erase
Remove an unwanted object, watermark or text by brush mask, text box or prompt.
Auto Text Erase
Detect text regions automatically and erase them, with mask dilation control.
AI Dust & Scratch Removal
Find and heal dust spots and scratches on scanned or old photos (sensitivity, max area, scratch length).

TRANSFORM / CROP

Transform
Rotate by any angle, flip horizontally or vertically.
Crop
Crop to a ratio such as 1:1, manually or smart, as a rectangle or an ellipse.
Upscale
Enlarge 2× or 4×, or resize to exact pixel dimensions.
Perspective
Correct perspective distortion — auto-detected, or manual four-point.
Grid Collage / Split
Split one image into a 3×3 grid, or stitch several images into one long strip.
True Grid Split
Cut an image into a true rows × columns grid (or a smart blank-cut) and output the tiles.
Doc Scanner
Flatten a photographed document: binarise (Sauvola / Gaussian), remove shadow and glare.
AI Privacy Mosaic
Automatically mosaic whatever you name — licence plates, faces — before an image leaves your hands.

AI GENERATION

Outpaint
Extend the canvas in any direction and let the model fill the new area from a prompt.
AI Background/Sky
Generate a new background from a prompt, or swap in a sky preset (dawn, noon, sunset, night, cloudy).
Fabric Transfer
Re-dress a garment in the fabric you describe.
Fog Depth Mask
Produce a depth map or an atmosphere/fog mask from the image.
Depth Bokeh
Blur by real depth: set the focus depth, the depth range and the bokeh shape (circle, hexagon, octagon).
Depth Lighting
Re-light the scene using depth — spotlight, ambient or layered depth lighting.
Stereo Anaglyph
Build red-cyan 3D-glasses images, plus stereo-pair and wiggle modes (eye separation, convergence).
Stereo Views
Generate N parallax views from a single photo (1→N) for stereo and multi-view output.
Depth Wiggle GIF
Turn a depth pass into an N-frame wiggle GIF (N→1 — a real animated GIF, not a video).

ADJUST / EFFECT

Text Overlay
Add one or more text layers — content, font, size, colour and position.
Attributes
Adjust brightness, contrast, saturation, sharpness, exposure, temperature, tint and gamma, with auto contrast / levels / white balance.
Color Replace
Swap one colour for another with a tolerance control — the classic product colour variant.
Color Transfer
Copy the colour mood of a reference photo onto your image at a chosen strength.
Watermark
Add a tiled text watermark or an invisible watermark, with font size, colour, opacity, angle and density.
QR / Barcode
Overlay a QR code or barcode with its content, size, position and opacity.
Selective Recolor
Recolour only the object you describe (for example "shirt"), by target colour or hue shift.

PORTRAIT

Beauty
Skin smoothing, brightening, whitening, blemish removal, face slimming and eye enlarging, with person type (auto / female / male / child / elder).
Face Restore
Restore detail in faces, with strength and optional 1× / 2× / 4× upscale.
Photo Restore
Repair old photos — auto or manual, denoise / dehaze / sharpness, optional face enhancement and upscale.
Makeup
Apply a makeup preset, or tune foundation, brows, eyeshadow, eyeliner, blush and lipstick individually.
Detail Enhancer
Local contrast and edge enhancement (CLAHE + guided filter) for texture that survives compression.
Face Align & Crop
Centre and crop to a target size on the face, optionally aligning the eyes — the ID-photo node.

EFFECT / FILTER

Border & Shadow
Add a coloured border with corner radius and an optional drop shadow.
Blur
Gaussian or motion blur, over the whole image or just the background.
Shadow & Reflection
Add a contact shadow, a projected shadow or a reflection, with angle, blur, offset and opacity.
Filter
Apply one of the built-in filters at a chosen intensity.
Halftone & Duotone
Stylise as half-tone dots or as a two-tone duotone.
Subject Bokeh Blur
Keep the subject sharp and blur everything else — background bokeh driven by the cut-out.
Sticker & Glow Stroke
Turn a cut-out into a sticker with a coloured stroke and optional glow.

FLOW CONTROL

Condition
Branch on image size or area (long edge, short edge, width, height, area) — only the matching true / false branch runs.
Merge
Combine several images into one output — first, concat or stack.
Switch
Pass on one of several inputs: the first, or a chosen index.
Comment
A sticky note on the canvas for documenting what a graph does.
Reroute
A passthrough point that keeps long connections tidy.
Subflow
Reuse a saved workflow as a single node, with its own inputs and output.
AI Culling
Quality-check every image — exposure, focus, closed eyes, duplicates — and route keep / reject down two branches.
Best Group Photo
Fuse a burst into one photo, taking the best expression per person (and optionally adding a person).

OUTPUT / OPTIMIZE

Compress
Encode as MozJPEG, WebP or PNG with a quality target, optional maximum dimensions and progressive output.
Metadata
Write title, description, author, copyright and keywords — or strip EXIF before the image leaves your machine.

All 53 ship with the app. Nodes are registry-driven, so a new one arrives with its ports, form, validation and summary already wired.

SPOTLIGHT

From one flat photo to a depth-driven set

Stack Pix estimates depth once, then reuses it across six node types. No separate 3D tool, no manual masking.

  • — Depth haze (fog_drm)
  • — Depth-of-field bokeh
  • — Depth-based relighting
  • — Stereo pair (N views)
  • — Anaglyph red/cyan
  • — Wiggle GIF

HOW IT COMPARES

How it compares

Four ways people solve the same problem. The honest answer is that they are not interchangeable — pick by who is doing the work and how many images there are.

 Stack PixComfyUIPhotoshop + batch actionsOnline AI image sites
Where it runsLocal Windows desktop appLocal, self-hosted (Python + GPU knowledge)Desktop appVendor cloud
Getting startedInstall and open — 81 ready-made workflows includedSet up the environment and download models firstFamiliar if you already know the editorOpen the site and upload
Where your photos goNever leave the machineStay localDepends on the feature — AI features are cloud-basedUploaded to the provider
BatchesBuilt in: array batches, grouping, resume, naming templatesPossible, but you build itRecorded actions — limited branchingQueued, usually priced per image
How you payOne-time purchase (free tier included)Free and open sourceMonthly or annual subscriptionPer image or subscription
Best forAnyone with a folder of their own photos or product shotsPeople who enjoy building model pipelinesProfessional retouching and compositingOne or two images, once in a while

ComfyUI

The node-canvas workflow tool this category is built on — free, open source, and endlessly extensible.

Pros

  • + Free and open source, with a huge community of shared workflows
  • + Almost any model can be wired in
  • + No per-image cost

Cons

  • − You install Python, manage models and maintain the environment yourself
  • − No finished workflows out of the box — you assemble them
  • − Steep for anyone who does not want to think about nodes and model files

Photoshop + batch actions

The professional default: a powerful editor with recorded actions for repetitive work.

Pros

  • + Everyone in a design team already knows it
  • + Best-in-class manual retouching
  • + Mature plugin ecosystem

Cons

  • − Subscription pricing, forever
  • − AI features run in the vendor cloud
  • − Actions are rigid — branching, grouping and per-image logic are painful

Online AI image sites

Upload a photo to a website, get a background removed or an upscale back.

Pros

  • + Nothing to install
  • + Fine for one image now and then
  • + Often free at very low volume

Cons

  • − Your images go to somebody else’s server
  • − Per-image pricing or daily caps once you have real volume
  • − Batches are slow, and you cannot chain steps into one pipeline

If the job is “a batch of my own photos, processed locally, without learning a tool”, Stack Pix is the one that fits. If you want to experiment with models for its own sake, ComfyUI is the better hobby. For one image a month, a website is fine — just know where the photo goes.

NOTES FROM USE

Notes from using it

Short write-ups of jobs we actually run through Stack Pix. These are workflows, not feature lists — each one is a canvas you can open and change.

  1. ID photos: one batch, printed sheet included

    Sep 10, 2026

    The ID-photo workflow does cut-out, background swap and the repeat_nup layout in one pass, so a folder of portraits comes out as print-ready sheets. The naming template puts each person in their own subfolder, which matters more than it sounds when you are handing files back to other people.

  2. Product shots: cut out, re-background, compress

    Sep 5, 2026

    Chain cut-out → background → compression and run it over a whole catalogue. The node cache is the part you notice: changing one parameter only recomputes that node and its children, so tuning the result on a big batch does not mean reprocessing everything from scratch.

  3. Culling a trip before editing it

    Aug 28, 2026

    Run ai_cull first and let it split the set into keep/reject on exposure, focus, closed eyes and duplicates. Colour work goes faster on the survivors, and the reject branch tells you why each frame was dropped instead of silently deleting it.

  4. Old scans: upscale first, restore second

    Aug 20, 2026

    Order matters. A 4× upscale before face restoration gives the restoration model far more to work with; doing it the other way round just enlarges the blur. Large scans are slow on CPU, so this is the one job where a CUDA GPU changes the experience.

Guides and articles

System requirements

Operating system
Windows 10 / 11 (64-bit)
Installer
MSI package, unbranded window shell
GPU
Optional — NVIDIA CUDA speeds up AI nodes, CPU fallback works without it
First-run download
≈700 MB engine pack, then models on demand
Model storage
Any drive — pick a folder in Settings, not tied to C:
App language
English, 简体中文
Interface themes
39 built-in colour palettes (Settings → Appearance)
Network
Only needed to download models and to activate
Account
None — a licence key is enough

AI models are large by nature (hundreds of MB to a few GB each). They download on demand from ModelScope, with Hugging Face as fallback, and stay on your disk.

Pricing

One-time payment, no subscription. Taxes are calculated at checkout — the payment provider is the merchant of record.

Single licence

$29.00

  • One-time payment
  • Up to 1 devices
  • Free updates within 1.x

Try it free first

The free version is not a demo with locked buttons: every node type works.

  • · All 53 node types available
  • · Up to 3 nodes per workflow
  • · Unlimited workflows, unlimited photos
  • · No account, no watermark, no expiry

QUESTIONS

Questions

Do my photos get uploaded?

No. All image processing runs locally on your machine. The only network calls are model downloads and licence activation.

Do I need a GPU?

No. Every node works on CPU. An NVIDIA GPU with CUDA makes the AI nodes significantly faster, but it is optional.

How many computers can I install it on?

That depends on the plan you buy — the number of devices is shown next to the price. Each machine is activated separately, and you can release a device yourself from the “My licences” page when you switch computers.

I replaced my computer. What now?

Release the old device in “My licences” and activate the new one — the seat is freed immediately. If the old machine is already gone, contact support with your order number.

What do I get in the free version?

Every node type, up to 3 nodes per workflow, unlimited workflows and photos. Buying a licence removes the node limit.

Which formats are supported?

JPEG, PNG, WebP, AVIF, JXL, GIF (animated GIFs are preserved frame-for-frame), TIFF and BMP. Output format is decided by the export node.

The model download failed. Can I retry?

Yes. Models are fetched on demand, and the download can be resumed. If ModelScope is unreachable, Stack Pix falls back to Hugging Face automatically, and you can point the model folder at any drive in Settings.

Do updates cost extra?

Updates within the major version you bought are free. If a future major version ships, you keep using the version you own — upgrading is optional.

Download

The installer is code-signed and published with a SHA-256 checksum.

Version 1.0.0 · released 2026-09-16