# Vision Studio·X

> Vision Studio·X is the marketing site for Vision Studio - a free, open-source (MIT-licensed) desktop application for AI image and video generation that runs on the user's own GPU. Local by default, with optional cloud routing only through the user's own API key. No subscriptions, no usage caps, no account. Version 3.4.1, released 2026-09-16.

The site is the public face of an instrument-grade local AI generation platform built for creators with capable GPUs. The product is an Electron + React desktop app with a bundled Python engine (FastAPI, PyTorch 2.5.1, Hugging Face Diffusers) supporting FLUX.1, Stable Diffusion 3.5, SDXL and SD 1.5 (image) and LTX Video, Stable Video Diffusion and AnimateDiff (video). The website mirrors the product's aesthetic: professional gear, not a SaaS landing page.

## What the product does
The complete, source-linked list is at https://vision-studio-x.com/features/ (and in llms-full.txt). In short:
- Image generation: FLUX.1 [dev] and [schnell], Stable Diffusion 3.5 Large and Medium, SDXL, SD 1.5
- Video generation: LTX Video and AnimateDiff from a prompt, Stable Video Diffusion from a still; MP4 output
- LoRA stacking with per-LoRA weights; ControlNet on SD 1.5, SDXL, FLUX.1 [dev] and SD 3.5 Large; IP-Adapter reference images on SD 1.5, SDXL and FLUX.1 [dev]
- Guided edit tools: background removal, AI upscale, face restoration, style transfer, generative fill, object removal, AI expand, background replace; inpaint and outpaint on the selected checkpoint (FLUX.1 Fill for FLUX)
- Storyboard with onion skin, and a timeline that exports MP4 with transitions and audio
- Asset library with search, favourites and export; an iteration tree to fork or re-roll any render; prompt-derived tags
- Model Foundry: search Hugging Face and CivitAI, install into the local library, consent gates for pickle and trust_remote_code models, hardware fit checks, a one-click starter set of 33 models (about 137 GB)
- ComfyUI: import API-format graphs and run them on the built-in engine; plain image and video jobs go to a ComfyUI server if one is running at launch
- Performance panel: Auto / On / Off for SDPA, channels-last, torch.compile, quantization, attention slicing and TensorRT; automatic CPU offload when a model only fits that way
- Privacy: local by default, no telemetry; optional OpenRouter / Hugging Face routing with the user's own key sends prompt text, never images

## Free & open source
- 100% free, no subscription, no usage caps, no account
- MIT licensed - source: https://github.com/Git-Rocky-Stack/vision-studio
- Release notes and tags: https://github.com/Git-Rocky-Stack/vision-studio/releases
- Every package carries the engine; model weights are downloaded separately, only the ones the user chooses
- Installers are served from the project's own delivery host, not GitHub (release assets are capped at 2 GB). The builds are not code-signed yet:
  - Windows x64: https://updates.vision-studio-x.com/win/Vision-Studio-3.4.1-Setup.exe (0.91 MB web installer, pulls 2.56 GB during install)
  - Windows x64, portable: https://updates.vision-studio-x.com/win/Vision%20Studio-3.4.1-win.zip (2.60 GB)
  - macOS Apple Silicon: https://updates.vision-studio-x.com/mac/Vision-Studio-3.4.1-arm64.dmg (464 MB)
  - Linux x86_64: https://updates.vision-studio-x.com/linux/Vision-Studio-3.4.1-x86_64.AppImage (3.10 GB)

## Latest release - 3.4.1 (2026-09-16)
A security patch on 3.4.0; additive, no known breaking changes.
- Security: the js-yaml override that cleared GHSA-2883-xcg3-v3hh is narrowed from the 5.x line to ^4.3.2 - the line electron-updater actually declares, and the line the advisory was fixed on. The tree resolves a single js-yaml 4.3.2; npm audit reports zero both with and without dev dependencies

### Also shipped in 3.4.1, carried from 3.4.0 (2026-09-13)
- Security: GHSA-2883-xcg3-v3hh cleared from the shipped auto-update path (js-yaml reached the app through electron-updater's update-feed parse; resolved with an override scoped to that package)
- Fixed: the PyTorch stack is pinned to torch 2.5.1 / torchvision 0.20.1 / torchaudio 2.5.1, so macOS stops resolving its own version and shipping a bundle that dies on launch
- Fixed: region-lock masks can be drawn on the Canvas tab (the toolbar had no drawing surface behind it)
- Fixed: Settings and the workflow workbench no longer crash when the renderer runs outside Electron
- Performance: the Konva canvas and the edit inspector no longer re-render on project writes that do not concern them
- Added: a test-enforced gate on unguarded mount-path uses of the preload bridge
- Full changelog: https://github.com/Git-Rocky-Stack/vision-studio/blob/main/CHANGELOG.md

## Pages
- [Home](https://vision-studio-x.com/): Hero, pipeline rack, Performance-panel optimizations, model spec sheet, GPU ladder, framing pads
- [Features](https://vision-studio-x.com/features/): The complete feature reference; every feature links to the source files that implement it
- [Download](https://vision-studio-x.com/download/): Installers for Windows x64 (web installer + portable ZIP), macOS 14+ on Apple Silicon (DMG), Linux x64 (AppImage); release log; system requirements; first-launch notes; build from source
- [Pricing](https://vision-studio-x.com/pricing/): Free forever, MIT-licensed; $0 versus cloud AI plans checked in September 2026, with sources
- [Compare](https://vision-studio-x.com/compare/): Vision Studio vs ComfyUI, Automatic1111/Forge, and cloud tools (Midjourney, Runway, Pika)
- [How it works](https://vision-studio-x.com/how-it-works/): The user guide - install, first launch, pick a model, dial parameters, generate on your GPU, edit and export - plus troubleshooting
- [FAQ](https://vision-studio-x.com/faq/): Cost, source, installs, offline use, GPUs, models, Hugging Face tokens, privacy, cloud routing, ComfyUI, platforms, unsigned-build warnings, storage, updates
- [Blog](https://vision-studio-x.com/blog/): Field notes on local-first AI generation, build and release engineering, GPU performance, and the economics of open models versus cloud AI subscriptions. The blog contains 2 articles as of September 16, 2026. Full archive with one entry per post: https://vision-studio-x.com/long-llms.txt
- [Comprehensive reference](https://vision-studio-x.com/llms-full.txt): Feature reference with sources, how the app runs, distribution and licensing, release history, design language, tech stack, market position
- [Complete blog archive](https://vision-studio-x.com/long-llms.txt): Every published post, one entry each, with category, date and summary

## System requirements
- Minimum: Windows 10 x64 / macOS 14 (Apple Silicon) / Ubuntu 22.04 x64, 8 GB RAM, 10 GB disk
- Recommended: NVIDIA GPU with 8 GB+ VRAM (or Apple M-series), 16 GB RAM, 50 GB disk for models
- One GPU is used: the first NVIDIA card through CUDA, or Apple Silicon through Metal (MPS). With no GPU the engine falls back to the CPU, which works but is slow. FLUX.1 lists about 24 GB of VRAM; SD 1.5 about 4 GB.

## Parent organization & ecosystem

Vision Studio·X is one spoke in the Strategia-X family. The site footer carries a "Strategia-X Ecosystem" switch panel that links every product in the network. Blog posts mirror to strategia-x.com/blog.

- [Strategia-X](https://strategia-x.com/): Parent brand and operations HQ
- [Team-X](https://www.team-x.app): AI native workforce OS
- [Strategia-X AI](https://strategia-x.ai): AI app idea validator
- [Android Architect](https://www.android-x.com): AI Android app engineering
- [Lumina Studio](https://www.lumina-os.com): Creative subscription suite (Vision-X design-language progenitor)
- [ClipForge](https://clip-forge.io): AI video clipping
- [ElementForge](https://www.element-forge.app): Windows creative suite
- [ResumeForge](https://resume-forge.app): Career intelligence desktop app
- [STX.1](https://stx-1.com): Windows system monitor
- [Dynasty-X](https://www.dynasty-x.com): Android app suite
- [WealthWise OS](https://www.wealthwiseos.com): AI personal finance OS