# NOVAI

**Repository Path**: invaders/novai

## Basic Information

- **Project Name**: NOVAI
- **Description**: NOVAI 是一款面向创作者的全能 AI 工作台。你可以在无限画布上拖拽、连线、组合各种 AI 能力——文生图、图生图、AI 视频生成、GPT 多模态对话——所有操作在一个界面中无缝衔接。
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 31
- **Forks**: 7
- **Created**: 2026-07-10
- **Last Updated**: 2026-08-21

## Categories & Tags

**Categories**: ai

**Tags**: None

## README

# NOVAI — Infinite Canvas AI Creation Platform

> Freely generate images and videos on an infinite canvas, connecting workflows with nodes. Built-in free ModelScope API, ready to use out of the box.
>
> **[中文](README.md) · [Website](https://invaders-2.github.io/NOVAI/) · [Docs](https://invaders-2.github.io/NOVAI/docs.html)**

NOVAI is an all-in-one AI workbench for creators. On an **infinite canvas** you can drag, connect, and combine various AI capabilities — text-to-image, image-to-image, AI video generation, GPT multimodal chat — all seamlessly in one interface. Paired with **Photoshop panels** and **Chrome extensions**, asset collection and creative workflows flow together.

![Home](static/images/screenshots/home.png)

![Infinite Canvas](static/images/screenshots/canvas.png)

![AI Generation](static/images/screenshots/generate.png)

![API Settings](static/images/screenshots/settings.png)

![Asset Manager](static/images/screenshots/assets.png)

---

## Quick Start

### Download & Install (Recommended)

No Python installation or environment configuration needed — download the installer and run.

#### Windows

1. Download the `NOVAI-Setup.exe` installer
2. Double-click to install and follow the wizard (desktop shortcut auto-created)
3. Launch the "NOVAI" icon on your desktop, browser opens `http://127.0.0.1:3000/` automatically

#### macOS

1. Download the `NOVAI.dmg` disk image
2. Open the DMG and drag `NOVAI.app` into the `Applications` folder
3. Double-click `NOVAI.app` to launch, browser opens `http://127.0.0.1:3000/` automatically

> **First launch note**: macOS may warn "cannot be opened because it is from an unidentified developer". Go to **System Settings → Privacy & Security → click "Open Anyway"**.

---

### Manual Install (Developers)

If you want to run from source or contribute:

```bash
# Clone the repository
git clone https://github.com/invaders-2/NOVAI.git
cd NOVAI

# Install dependencies
pip install -r requirements.txt

# Start the server
python main.py
```

Open `http://127.0.0.1:3000/` in your browser.

> Requirements: Python 3.10+

---

## FAQ

### Windows

| Issue | Solution |
|-------|----------|
| **Cannot open after install** | Check firewall settings; try running as administrator |
| **Port 3000 in use** | Close the program using the port, or set env var `DEPLOY_RUN_PORT=3001` before starting |
| **Dependency install fails** | Ensure Python 3.10+ is installed with "Add Python to PATH" checked |
| **Antivirus false positive** | Add the NOVAI installation directory to antivirus whitelist |

### macOS

| Issue | Solution |
|-------|----------|
| **"Unidentified developer" warning** | Go to **System Settings → Privacy & Security → click "Open Anyway"** |
| **"App is damaged" warning** | Run in terminal: `sudo xattr -rd com.apple.quarantine /Applications/NOVAI.app` |
| **Port 3000 in use** | Run `lsof -i :3000` to find the process, `kill <PID>` to end it; or set `DEPLOY_RUN_PORT=3001` |
| **Dependency issues (manual install)** | Ensure Python 3.10+: `python3 --version`, then `pip3 install -r requirements.txt` |

---

## AI Models

Built-in **ModelScope** free API, ready out of the box:

| Type | Model | Description |
|------|-------|-------------|
| Chat | Qwen3-235B | Tongyi Qianwen flagship, reasoning/writing |
| Chat | Qwen3-VL-235B | Multimodal, image analysis |
| Image | Qwen-Image-2512 | Best Chinese text rendering, posters/banners |
| Image | FLUX.2-klein-9B | Best visual quality |
| Image | Z-Image-Turbo | 1-2s fast generation |
| Image | Qwen-Image-Edit-2511 | Edit images by talking |
| Video | agnes-video-v2.0 | Free video generation |

---

## Key Features

- 🎨 **Infinite Canvas** — free zoom, drag, and connect for creation
- ⚡ **Node Workflows** — image/video/LLM nodes connected in chains
- 🤖 **Auto-Place on Canvas** — AI results auto-create nodes, no manual import
- 💡 **Smart Suggestion Bar** — auto-popup "change outfit / change scene / add camera move" one-click continuation
- 🔍 **@ Reference** — type @ in the input box to reference canvas nodes as source material
- 👤 **Face Blur** — one-click blur all faces in a video to protect the privacy of people in the frame (YuNet detection)
- 🔌 **Plugin System** — Photoshop panels, Chrome extensions
- 🌐 **Multi-Repo Updates** — GitHub / Gitee / ModelScope three sources
- 🇨🇳 **Chinese Optimized** — deep Qwen ecosystem integration

---

## Responsible Use & Content Compliance

Please use NOVAI responsibly and in compliance with applicable laws and platform content-safety policies. When working with footage that contains recognizable people, ensure you have the necessary rights or consent, and avoid using any media that infringes on others' personal rights or privacy.

> **Privacy tip**: If your reference footage contains bystanders or people who have not given consent, use the built-in **Face Blur** tool (Smart Canvas → select the video node → toolbar "Face Blur") to protect their identity before using the material.

---

## Download & Updates

Installers and update sources:

| Platform | URL |
|----------|-----|
| GitHub | https://github.com/invaders-2/NOVAI |
| Gitee | https://gitee.com/invaders/novai |
| ModelScope | https://modelscope.cn/studios/bllack/NOVAI |

The app checks all three sources for the latest version on startup, pushes update notifications, and upgrades with one click.

---

## Desktop Client (Electron · Beta)

> New packaging route: **Electron** with its own Chromium engine (consistent behavior across platforms, no more system WebView differences), with electron-updater auto-update.

- Source & build docs: **desktop/README.md**
- Reuses the existing frontend through a pywebview compatibility bridge (window.pywebview.api) — native dialogs / window controls / system tray with zero frontend changes
- **Fully shares data with the official build**: same port 3000 and same data directory (Windows `%APPDATA%/NOVAI`, macOS `~/NOVAI`) — canvases, settings and chat history are shared, **no data re-download or migration needed, just install and run**; if the official build is already running, the Electron app reuses its backend instead of starting a second one
- **Three-repo update channel**: GitHub Releases as the primary source; if the check fails (common on Chinese networks) it automatically falls back to a ModelScope mirror (`bllack/NOVAI-releases`), with Gitee Releases as a manual-download backup. The cloud pipeline (triggered by pushing a `v*` tag) syncs installers to all three repos automatically
- **Auto-restart after update**: on Windows the update installs and restarts automatically after user confirmation; on macOS, unsigned builds cannot auto-install (Squirrel requires code signing), so the app falls back to a guided manual download — data is never affected

---

## Changelog

### v1.0.112-beta.1 (Beta)

- **Electron desktop client**: bundled Chromium engine; shares port 3000 + the same data directory with the official build (full data interop); tray / native dialogs / window controls
- **Live update progress bar**: per-stage percentage (download / verify / backup / replace) + current file name, with toast notifications for update results
- **Three-repo update channel**: GitHub primary, automatic ModelScope mirror fallback for Chinese networks, Gitee manual-download backup
- **Desktop sidebar flicker fix**: hover-expand no longer toggles rapidly under Chromium

### v1.0.111

- **Local high-quality cutout**: image nodes get "High-Quality Cutout" — RMBG-2.0 local model downloads on demand (349MB, first use only), then offline & free; edge refinement (alpha stretch + erode/feather + background-color decontamination); one-click fallback to online cutout
- **Windows title bar drag fix**: frameless window title bar is draggable again
- **Old model cleanup**: leftover RMBG-1.4 model is removed automatically after upgrade

### v1.0.108

- **Port fix**: restored official port 3000 (accidental test-build 3001 push corrected)
- **Model whitelist**: assets/models/ added to online-update whitelist, face-detection model ships with updates

### v1.0.107

- **Face Blur tool**: select video node → one-click blur all faces in toolbar for better privacy (YuNet detection + Gaussian blur), result auto-added to node
- **Video playback fixes**: play / pause / fullscreen / progress bar fully fixed, desktop matches browser (native controls)
- **Reference media publicization**: cloud upload (Litterbox / temp.sh) preferred, no longer depends on local tunnel

### v1.0.86

- **Cloud build pipeline**: Windows/Mac cloud packaging pipelines, auto build & release
- **Carousel assets in repo**: fixed carousel images missing after updates
- **Frameless window fixes**: Win frameless drag / edge resize / title bar layout fixes
- **Mac signing optimization**: dmg ad-hoc signing prevents "damaged" warning, fixed signature broken by copy
- **Encoding fixes**: fixed Chinese encoding issues in Windows cloud builds

### v1.0.85

- **Smart Canvas fixes**: composer positioning, modal clipping, sidebar offset, title bar layout fixes

### v1.0.84

- **Canvas list enhancement**: cards support up to 4 images in 2×2 grid preview
- **Desktop upgrade**: frameless window + system tray (minimize to tray) + LAN access
- **Desktop icon update**: white-on-black rounded logo
- **Fixes**: 64-bit ctypes WndProc pointer truncation crash, pystray packaging, port conflicts
- **Repo cleanup**: improved .gitignore, excluded drafts/test screenshots/backups

### v1.0.83

- **Frontend performance**: Lucide icon subsetting (90% size reduction), NovaUtils/NovaMedia shared modules, Marvis-style CSS dedup, timer leak fixes, pause rAF rendering when hidden, touch-mouse layout cache optimization
- **Model picker redesign**: two-column layout, hover-linked provider switching, dynamic height
- **GPT chat fixes**: fixed TDZ error causing blank model picker
- **Icon & tooltip fixes**: multiple pages' Lucide icons, asset library status bar leaking English errors
- **Unified settings style**: API settings & workflow settings redesigned to Marvis style, fixed CSS syntax errors
- **New backend tool APIs**: 7 endpoints (storage management, image detection, category prompts, model normalization, RunningHub wallet status)
- **Race condition fixes**: sendChatMessage re-entry guard, setTimeout recursive polling, node.running premature reset