# Open Generative AI β Unrestricted Open-Source Alternative to AI Video Platforms
[](https://muapi.ai?utm_source=github&utm_medium=badge&utm_campaign=open-generative-ai)
> **The free, open-source alternative to AI Video Platforms.** Generate AI images and videos using 200+ state-of-the-art models β no content filters, no closed ecosystem, no subscription fees.
**Community:** Join [Discord](https://discord.gg/tANKJkHck) for discussions and support
πΊ Watch: MuAPI β The Best Fal.ai Alternative (cheaper AI video & image APIs) β
> π¨ **[Explore 50+ more open-source AI apps β](https://github.com/Anil-matcha/awesome-generative-ai-apps)**
## Related Projects
> π€ **Automate media generations with AI coding agents:** [Generative-Media-Skills](https://github.com/SamurAIGPT/Generative-Media-Skills) β a library of skills that let agents like **Claude Code**, **Codex**, and other coding assistants drive 200+ image/video models end-to-end (prompt β generate β edit β stitch) directly from your terminal. Perfect for building automated media pipelines without touching a UI.
> π¬ **Seedance 2.5 prompts & API guide:** [awesome-seedance-2.5-api-prompts](https://github.com/Anil-matcha/awesome-seedance-2.5-api-prompts) β Curated prompt templates, camera control vocabulary, MuAPI reference, and cinematic examples for Seedance 2.5 video generation.
> π₯ **Seedance 2.5 Python SDK:** [Seedance-2.5-API](https://github.com/SamurAIGPT/Seedance-2.5-API) β Python wrapper for ByteDance's Seedance 2.5 API β text-to-video, image-to-video, realistic human faces, character consistency.
> π **Claude Fable 5 use cases + 20% off on MuAPI:** [awesome-claude-fable-5](https://github.com/Anil-matcha/awesome-claude-fable-5) β 60 curated real-world use cases, prompts, and benchmarks for Claude Fable 5, with **20% off Fable 5 access** via [MuAPI](https://muapi.ai/pricing?utm_source=github&utm_medium=readme&utm_campaign=open-generative-ai).
- [Vadoo](https://vadoo.tv) β Unrestricted AI image & video generation β auto-publish as YouTube Shorts and TikToks & earn
- [AI-Youtube-Shorts-Generator](https://github.com/SamurAIGPT/AI-Youtube-Shorts-Generator) β Auto-generate viral YouTube Shorts from long-form videos using AI
- [muapi-cli](https://github.com/SamurAIGPT/muapi-cli) β Official CLI for MuAPI β run these models from your terminal
- [Vibe-Workflow](https://github.com/SamurAIGPT/Vibe-Workflow) β Node-based AI workflow builder for generative image & video pipelines
- [Text-To-Video-AI](https://github.com/SamurAIGPT/Text-To-Video-AI) β Lightweight text-to-video script β no UI required
- [muapi-comfyui](https://github.com/SamurAIGPT/muapi-comfyui) β ComfyUI nodes for 100+ MuAPI models
- [n8n-nodes-muapi](https://github.com/SamurAIGPT/n8n-nodes-muapi) β n8n community nodes for MuAPI β automate media generation
- [Open-AI-Design-Agent](https://github.com/Anil-matcha/Open-AI-Design-Agent) β Open-source autonomous AI design agent
- [Free-AI-Social-Media-Scheduler](https://github.com/Anil-matcha/Free-AI-Social-Media-Scheduler) β Free open-source AI social media scheduler β self-hostable alternative to Buffer and Hootsuite
- [awesome-seedance-2.5-api-prompts](https://github.com/Anil-matcha/awesome-seedance-2.5-api-prompts) β Curated Seedance 2.5 API guide, prompts, camera controls, and video generation examples
- [AI-Voice-Agent](https://github.com/Anil-matcha/AI-Voice-Agent) β Self-hosted AI voice agent for real-time voice conversations, sales calls, and customer support
## π Try it Online β No Install Required
**Hosted version:** [https://muapi.ai/open-generative-ai?utm_source=github&utm_medium=readme&utm_campaign=open-generative-ai](https://muapi.ai/open-generative-ai?utm_source=github&utm_medium=readme&utm_campaign=open-generative-ai)
Use all studios (Image, Video, Audio, AI Clipping, Vibe Motion, Lip Sync, Cinema, Marketing, Workflows, Agents, Design Agent, Apps, MCP & CLI) directly in your browser β no Node.js, no setup. Sign up for a free account to start generating. The hosted version is always up to date with the latest models.
**Follow** the [creator](https://x.com/matchaman11) for updates
---
## β¬οΈ Download Desktop App
One-click installers β no Node.js or terminal required.
| Platform | Download |
|---|---|
| macOS Apple Silicon (M1/M2/M3/M4) | [Open Generative AI-1.0.9-arm64.dmg](https://github.com/Anil-matcha/Open-Generative-AI/releases/download/v1.0.9/Open.Generative.AI-1.0.9-arm64.dmg) |
| macOS Intel (x64) | [Open Generative AI-1.0.9.dmg](https://github.com/Anil-matcha/Open-Generative-AI/releases/download/v1.0.9/Open.Generative.AI-1.0.9.dmg) |
| Windows (x64) | [Open Generative AI Setup 1.0.9.exe](https://github.com/Anil-matcha/Open-Generative-AI/releases/download/v1.0.9/Open.Generative.AI.Setup.1.0.9.exe) |
| Linux (Ubuntu x64) | [v1.0.9 release](https://github.com/Anil-matcha/Open-Generative-AI/releases/tag/v1.0.9) (`.AppImage` / `.deb`), or build locally with `npm run electron:build:linux`. |
All releases: [github.com/Anil-matcha/Open-Generative-AI/releases](https://github.com/Anil-matcha/Open-Generative-AI/releases)
### macOS Installation Guide
Because the app is not notarized by Apple, macOS Gatekeeper will block it on first launch. Follow these steps:
**Step 1** β Mount the DMG and drag the app to `/Applications`
**Step 2** β Open Terminal and run:
```bash
xattr -cr "/Applications/Open Generative AI.app"
```
**Step 3** β Right-click the app in `/Applications` β click **Open** β click **Open** again on the dialog
> You only need to do this once. After that, the app opens normally.
**Alternative (no Terminal):**
1. Try to open the app β macOS will block it
2. Go to **System Settings β Privacy & Security**
3. Scroll down to find _"Open Generative AI was blocked"_
4. Click **Open Anyway** β **Open**
### Windows Installation β SmartScreen warning fix
Windows SmartScreen may show a warning because the installer is not code-signed:
1. Click **More info** on the SmartScreen dialog
2. Click **Run anyway**
The app will install silently to `%LocalAppData%` with a Start Menu shortcut.
### Ubuntu / Linux Installation
Linux artifacts are available when building with Electron Builder:
```bash
# Build Linux installers (AppImage + .deb)
npm run electron:build:linux
```
Generated files are written to the `release/` folder:
- **AppImage** β portable, run directly after making executable:
```bash
chmod +x "release/Open Generative AI-*.AppImage"
./release/Open\ Generative\ AI-*.AppImage
```
- **.deb** β install on Debian/Ubuntu:
```bash
sudo apt install ./release/open-generative-ai_*_amd64.deb
```
If AppImage fails to start on older systems, install `libfuse2`:
```bash
sudo apt install libfuse2
```
#### Ubuntu 24.04+ / AppArmor sandbox restriction
Ubuntu 24.04 and later enable a kernel security policy (`apparmor_restrict_unprivileged_userns`) that blocks Chromium's user-namespace sandbox. If the app fails to start silently or crashes immediately, you have two options:
**Option A β Recommended: install the `.deb` instead.**
The `.deb` package ships an AppArmor profile that grants the required permission automatically on install with no system-wide changes.
**Option B β Temporary system fix (AppImage users):**
```bash
sudo sysctl -w kernel.apparmor_restrict_unprivileged_userns=0
```
This lasts until next reboot. To make it permanent:
```bash
echo 'kernel.apparmor_restrict_unprivileged_userns=0' | sudo tee /etc/sysctl.d/99-userns.conf
```
---
Open Generative AI is a free, open-source AI image, video, cinema, and lip sync studio that brings creative workflows to everyone. No content filters, no prompt rejections, no guardrails β just full creative freedom. Powered by [Muapi.ai](https://muapi.ai?utm_source=github&utm_medium=readme&utm_campaign=open-generative-ai), it supports text-to-image, image-to-image, text-to-video, image-to-video, and audio-driven lip sync generation across models like Flux, Nano Banana, Midjourney, Kling, Sora, Veo, Seedream, Infinite Talk, LTX Lipsync, Wan 2.2, and more β all from a sleek, modern interface you can self-host and customize.
**Why Open Generative AI instead of other AI Video Platforms?**
- **No filters** β no content filters, no nanny guardrails, no prompt rejections
- **Free & open-source** β no subscription, no vendor lock-in
- **Self-hosted** β your data stays on your machine, full creative control
- **200+ models** β text-to-image, image-to-image, text-to-video, image-to-video, lip sync
- **Multi-image input** β feed up to 14 reference images into compatible models
- **Lip Sync Studio** β animate portraits or sync lips to any audio with 9 dedicated models
- **Extensible** β add your own models, modify the UI, build on top of it
For a deep dive into the technical architecture and the philosophy behind the "Infinite Budget" cinema workflow, see our [comprehensive guide and roadmap](https://medium.com/@anilmatcha/).
## β‘ Local Model Inference (Desktop App Only)
The desktop app supports **two independent local engines**. Pick whichever fits the machine you actually run on:
| Engine | What it is | Best for |
|---|---|---|
| **sd.cpp** (bundled) | C++ engine from [stable-diffusion.cpp](https://github.com/leejet/stable-diffusion.cpp), runs on the same machine as the app. Metal GPU on Apple Silicon, CUDA/Vulkan/ROCm on Linux/Windows. | Image-only models. Works on Mac M-series. |
| **Wan2GP** (BYO server) | HTTP client to a user-run [Wan2GP](https://github.com/deepbeepmeep/Wan2GP) server. The server runs Python + PyTorch on a CUDA/ROCm GPU; the desktop app only sends prompts and receives results. | Video models (Wan 2.2, Hunyuan, LTX) and large image models (Flux, Qwen-Image). NVIDIA/AMD GPU required on the *server*; the desktop app itself can run on a Mac. |
Both engines share the same UI: open **Settings β Local Models** to configure each.
### Engine 1 β sd.cpp (bundled)
| Model | Type | Size | Notes |
|---|---|---|---|
| **Z-Image Turbo** β‘ | Diffusion Transformer | 2.5 GB + 2.7 GB aux | 8-step turbo. Heavy on memory. |
| **Z-Image Base** β‘ | Diffusion Transformer | 3.5 GB + 2.7 GB aux | 50-step high-quality. Heavy on memory. |
| **Dreamshaper 8** | SD 1.5 | 2.1 GB | 20-step versatile. Lightest tested option on Mac. |
| **Realistic Vision v5.1** | SD 1.5 | 2.1 GB | 25-step photorealistic |
| **Anything v5** | SD 1.5 | 2.1 GB | 20-step anime/illustration |
| **SDXL Base 1.0** | SDXL | 6.9 GB | 30-step high-res |
> **Z-Image models** require two shared auxiliary files (downloaded once, shared across both models):
> - **Qwen3-4B Text Encoder** β 2.4 GB
> - **FLUX VAE** β 335 MB
**How to use:**
1. Open **Settings β Local Models** in the desktop app
2. Install the **sd.cpp inference engine** (one click β auto-downloaded)
3. Download your chosen model (and auxiliary files for Z-Image)
4. In **Image Studio**, click the **β‘ Local** toggle next to the model selector
5. Select your local model and generate β no API key needed
All downloads happen inside the app. Nothing is installed system-wide.
By default, `sd.cpp` stores the engine, model weights, and temporary downloads under Electron's app data directory. Common paths are:
- macOS: `~/Library/Application Support/open-generative-ai/local-ai`
- Windows: `%APPDATA%\open-generative-ai\local-ai`
- Linux: `~/.config/open-generative-ai/local-ai`
To keep multi-GB model weights on another drive, set `OPEN_GENERATIVE_AI_LOCAL_AI_DIR`
before launching the desktop app. The app will create `bin/`, `models/`, and `tmp/`
inside that directory, and **Settings -> Local Models** shows the resolved model folder.
Local engine output and download errors are written to the app process console, so launch
from Terminal or PowerShell when you need troubleshooting logs.
### Engine 2 β Wan2GP (remote Gradio server)
The app does **not** bundle Python or model weights for Wan2GP. You run Wan2GP yourself on a machine with a CUDA or ROCm GPU and point the desktop app at its URL.
```bash
# On your GPU machine
git clone https://github.com/deepbeepmeep/Wan2GP
cd Wan2GP
./install.sh # or install.bat on Windows
python wgp.py --listen --server-name 0.0.0.0 # binds to all interfaces
```
Then in the desktop app: **Settings β Local Models β Wan2GP server**, paste the URL (e.g. `http://192.168.1.42:7860`), click **Test**, then **Save**. The Wan2GP models become available β image models in **Image Studio**, video models reachable via the same generation API (Image Studio rejects video output explicitly; full Video Studio wiring is on the roadmap).
| Model | Type | Notes |
|---|---|---|
| **Flux.1 Dev** | Image | 1024px, 28 steps |
| **Qwen Image** | Image | 1024px, 30 steps |
| **Wan 2.2 (T2V / I2V)** | Video | Slow on consumer GPUs |
| **Hunyuan Video** | Video | High-quality T2V |
| **LTX Video** | Video | Fastest video option |
> **Why a separate server?** Wan2GP's runtime (Sage attention, flash-attn, AWQ/GGUF kernels) is CUDA-only β there is no MPS / Apple Silicon path. Treating it as a remote server lets a Mac-only user keep the desktop app while offloading inference to a Linux/Windows GPU box, a gaming PC on the LAN, or a rented RunPod/vast.ai instance.
> **Local inference is only available in the desktop app.** The hosted web version always uses cloud APIs.
### Hardware Notes
- **sd.cpp** runs on CPU (all platforms) and **Metal GPU** on Apple Silicon (M1/M2/M3/M4); CUDA/Vulkan/ROCm on Linux/Windows.
- Metal GPU acceleration is built into the macOS desktop binary β significantly faster than CPU-only.
- Recommended for sd.cpp Z-Image: 16 GB RAM (7.4 GB weights + 2.4 GB compute buffer). On a base 8 GB M-series Mac, **Z-Image is known to hang the system** β stick to SD 1.5 there.
- For SD 1.5 on M2: expect ~1β2 s/step with the Metal dylib active. If you see ~10 s/step instead, the binary may have fallen back to CPU β see verification below.
### Verifying the SD 1.5 path (the fastest sanity test on Mac)
If you want to confirm sd.cpp is installed correctly without going through the UI, you can drive `sd-cli` directly. This is the same binary the app uses.
```bash
# 1. App data layout (created on first app launch)
APP_DATA="${OPEN_GENERATIVE_AI_LOCAL_AI_DIR:-$HOME/Library/Application Support/open-generative-ai/local-ai}"
ls "$APP_DATA/bin" # sd-cli, libstable-diffusion.dylib
ls "$APP_DATA/models" # whatever you've downloaded
# 2. Grab a small SD 1.5 model directly (Dreamshaper 8, ~2 GB)
curl -L --fail --progress-bar \
-o "$APP_DATA/models/DreamShaper_8_pruned.safetensors" \
"https://huggingface.co/Lykon/DreamShaper/resolve/main/DreamShaper_8_pruned.safetensors"
# 3. Run a single 512x512 / 12-step inference
DYLD_LIBRARY_PATH="$APP_DATA/bin" "$APP_DATA/bin/sd-cli" \
-m "$APP_DATA/models/DreamShaper_8_pruned.safetensors" \
-p "a serene mountain lake at sunrise, oil painting" \
-o /tmp/sd15-test.png \
--steps 12 -H 512 -W 512 --cfg-scale 7.5 --seed 42 \
--sampling-method euler_a
```
A healthy run on Apple Silicon prints `total params memory size = 1969.78MB (VRAM 1969.78MB, RAM 0.00MB)` (Metal-backed) and produces a coherent 512Γ512 PNG. If `VRAM` is `0.00MB` instead, the dylib is CPU-only β check `otool -L "$APP_DATA/bin/libstable-diffusion.dylib" | grep -i metal` and reinstall the engine from **Settings β Local Models** if Metal is missing.
---
## β¨ Features
- **Image Studio** β Generate images from text prompts (50+ text-to-image models) or transform existing images (55+ image-to-image models). Switches model set automatically based on whether a reference image is provided. Quality and resolution controls visible for models that support them.
- **Local Inference** β Two engines: **sd.cpp** (bundled, runs on Mac/Win/Linux with Metal/CUDA/Vulkan/ROCm) for SD 1.5, SDXL, and Z-Image; and **Wan2GP** (BYO Gradio server) for Flux, Qwen-Image, and video models (Wan 2.2, Hunyuan, LTX). Configure both in Settings β Local Models.
- **Multi-Image Input** β Upload up to 14 reference images for compatible edit models (Nano Banana 2 Edit, Flux Kontext Dev, GPT-4o Edit, and more). Multi-select picker with order badges, batch upload, and a "Use Selected" confirmation flow.
- **Video Studio** β Generate videos from text prompts (40+ text-to-video models) or animate a start-frame image (60+ image-to-video models). Same intelligent mode switching as Image Studio.
- **Lip Sync Studio** β Animate portrait images or sync lips on existing videos using audio. 9 dedicated models across two modes: portrait image + audio β talking video, and video + audio β lipsync video.
- **Cinema Studio** β Interface for photorealistic cinematic shots with pro camera controls (Lens, Focal Length, Aperture)
- **Workflow Studio** β Build and run multi-step AI pipelines visually. Chain image, video, and audio models into automated flows. Browse community templates, create your own with a node-based editor, and run them via an interactive playground.
- **Upload History** β Reference images are uploaded once and stored locally. A picker panel lets you reuse any previously uploaded image across sessions β no re-uploading.
- **Smart Controls** β Dynamic aspect ratio, resolution/quality, and duration pickers that adapt to each model's capabilities (including t2i models with resolution or quality options)
- **Generation History** β Browse, revisit, and download all past generations (persisted in browser storage)
- **Image & Video Download** β One-click download of generated outputs in full resolution
- **API Key Management** β Secure API key storage in browser localStorage (never sent to any server except Muapi)
- **Responsive Design** β Works seamlessly on desktop and mobile with dark glassmorphism UI
### πΌοΈ Image Studio β Dual Mode
The Image Studio automatically switches between two model sets:
| Mode | Trigger | Models | Prompt |
| :--- | :--- | :--- | :--- |
| **Text-to-Image** | Default (no image) | 50+ t2i models (Flux, Nano Banana 2, Seedream 5.0, Ideogram, GPT-4o, Midjourneyβ¦) | Required |
| **Image-to-Image** | Reference image uploaded | 55+ i2i models (Kontext, Nano Banana 2 Edit, Seedream 5.0 Edit, Seededit, Upscalerβ¦) | Optional |
#### Newly Added Models
| Model | Type | Key Features |
| :--- | :--- | :--- |
| **Nano Banana 2** | Text-to-Image | Google Gemini 3.1 Flash Image Β· Resolution 1K/2K/4K Β· Google Search enhancement Β· aspect ratio `auto` |
| **Nano Banana 2 Edit** | Image-to-Image | Up to **14 reference images** Β· Resolution 1K/2K/4K Β· Google Search enhancement |
| **Seedream 5.0** | Text-to-Image | ByteDance Β· Quality basic/high Β· 8 aspect ratios Β· up to 4K |
| **Seedream 5.0 Edit** | Image-to-Image | ByteDance Β· Natural language style transfer Β· Quality basic/high |
| **MiniMax Image 01** | Text-to-Image | MiniMax Β· 8 aspect ratios Β· up to 4 images per request Β· 1500 char prompt |
#### Multi-Image Input
Models that accept multiple reference images expose a multi-select picker when active:
| Model | Max Images |
| :--- | :--- |
| Nano Banana 2 Edit | 14 |
| Nano Banana Edit | 10 |
| Flux Kontext Dev I2I | 10 |
| Kling O1 Edit Image | 10 |
| GPT-4o Edit / GPT Image 1.5 Edit | 10 |
| Bytedance Seedream Edit v4 / v4.5 | 10 |
| Vidu Q2 Reference to Image | 7 |
| Flux 2 Flex/Pro Edit | 8 |
| Nano Banana Pro Edit | 8 |
| Flux Kontext Pro/Max I2I | 2 |
| Wan 2.5/2.6 Image Edit | 2β3 |
| Qwen Image Edit Plus / 2511 | 3 |
| GPT-4o Image to Image | 5 |
| Flux 2 Klein 4b/9b Edit | 4 |
When a multi-image model is selected the upload trigger switches to multi-select mode:
- **Checkboxes with order numbers** β images are sent to the model in the order you select them
- **Batch upload** β pick multiple files at once from your file dialog
- **Count badge** on the trigger shows how many images are active; a `+` badge appears when more slots are available
- **"Use Selected" button** confirms and closes the picker
### π¬ Video Studio β Dual Mode
The Video Studio follows the same pattern:
| Mode | Trigger | Models | Prompt |
| :--- | :--- | :--- | :--- |
| **Text-to-Video** | Default (no image) | 40+ t2v models (Kling, Sora, Veo, Wan, Seedance 2.0, Hailuo, Runwayβ¦) | Required |
| **Image-to-Video** | Start frame uploaded | 60+ i2v models (Kling I2V, Veo3 I2V, Runway I2V, Wan I2V, Seedance 2.0 I2V, Midjourney I2Vβ¦) | Optional |
#### Newly Added Models
| Model | Type | Key Features |
| :--- | :--- | :--- |
| **Seedance 2.0** | Text-to-Video | ByteDance Β· Aspect ratios 16:9 / 9:16 / 4:3 / 3:4 Β· Duration 5 / 10 / 15s Β· Quality basic/high |
| **Seedance 2.0 I2V** | Image-to-Video | ByteDance Β· Animate images into video Β· Up to 9 reference images Β· Aspect ratios 16:9 / 9:16 / 4:3 / 3:4 Β· Duration 5 / 10 / 15s Β· Quality basic/high |
| **Seedance 2.0 Extend** | Video Extension | ByteDance Β· Seamlessly continue any Seedance 2.0 generation Β· Preserves style, motion & audio Β· Optional continuation prompt Β· Duration 5 / 10 / 15s Β· Quality basic/high |
| **Grok Imagine T2V** | Text-to-Video | xAI Β· Duration 6 / 10 / **15s** Β· Modes: fun / normal / spicy Β· Aspect ratios 9:16 / 16:9 / 2:3 / 3:2 / 1:1 |
| **Grok Imagine I2V** | Image-to-Video | xAI Β· Duration 6 / 10 / **15s** Β· Modes: fun / normal / spicy Β· Cinematic motion from still images |
| **MiniMax Hailuo 02 / 2.3 Standard & Pro** | Text-to-Video / Image-to-Video | MiniMax Β· Full HD video Β· Multiple aspect ratios Β· Fast variant included |
### ποΈ Lip Sync Studio
The **Lip Sync Studio** generates audio-driven talking videos using 9 models across two input modes:
| Mode | Trigger | Description |
| :--- | :--- | :--- |
| **Portrait Image** | Default | Upload a portrait image + audio file β animated talking video |
| **Video** | Switch to Video mode | Upload an existing video + audio file β lipsync video |
#### Image-based Models (Portrait Image + Audio β Video)
| Model | Endpoint | Resolutions | Prompt |
| :--- | :--- | :--- | :--- |
| **Infinite Talk** | `infinitetalk-image-to-video` | 480p, 720p | Optional |
| **Wan 2.2 Speech to Video** | `wan2.2-speech-to-video` | 480p, 720p | Optional |
| **LTX 2.3 Lipsync** | `ltx-2.3-lipsync` | 480p, 720p, 1080p | Optional |
| **LTX 2 19B Lipsync** | `ltx-2-19b-lipsync` | 480p, 720p, 1080p | Optional |
#### Video-based Models (Video + Audio β Lipsync Video)
| Model | Endpoint | Resolutions | Prompt |
| :--- | :--- | :--- | :--- |
| **Sync Lipsync** | `sync-lipsync` | β | β |
| **LatentSync** | `latentsync-video` | β | β |
| **Creatify Lipsync** | `creatify-lipsync` | β | β |
| **Veed Lipsync** | `veed-lipsync` | β | β |
| **Infinite Talk V2V** | `infinitetalk-video-to-video` | 480p, 720p | Optional |
**How it works:**
1. Select **Portrait Image** or **Video** mode using the toggle
2. Upload your portrait image (or video) using the image/video upload button
3. Upload your audio file using the audio upload button
4. Optionally enter a prompt to guide the motion style
5. Select a model and resolution (where supported), then click **Generate**
Generation history is saved separately in `lipsync_history` and pending jobs resume automatically on page reload.
### π Workflow Studio
The **Workflow Studio** lets you build and run multi-step AI pipelines without writing code.
**Key capabilities:**
- **Templates** β Start from pre-built workflows (image chains, video pipelines, and more)
- **My Workflows** β Save and manage your own custom pipelines
- **Community** β Browse and run workflows published by other users
- **Node-based Builder** β Drag-and-drop visual editor to connect models and route outputs between steps
- **Playground** β Run any workflow interactively with a form UI; results render inline
- **API execution** β Every workflow is also callable via the Muapi API
> π‘ **Want to add workflows to your own app?** Check out **[Vibe Workflow](https://github.com/SamurAIGPT/Vibe-Workflow)** β the open-source workflow engine powering this feature. Drop it into any project.
### π₯ Cinema Studio Controls
The **Cinema Studio** offers precise control over the virtual camera, translating your choices into optimized prompt modifiers:
| Category | Available Options |
| :--- | :--- |
| **Cameras** | Modular 8K Digital, Full-Frame Cine Digital, Grand Format 70mm Film, Studio Digital S35, Classic 16mm Film, Premium Large Format Digital |
| **Lenses** | Creative Tilt, Compact Anamorphic, Extreme Macro, 70s Cinema Prime, Classic Anamorphic, Premium Modern Prime, Warm Cinema Prime, Swirl Bokeh Portrait, Vintage Prime, Halation Diffusion, Clinical Sharp Prime |
| **Focal Lengths** | 8mm (Ultra-Wide), 14mm, 24mm, 35mm (Human Eye), 50mm (Portrait), 85mm (Tight Portrait) |
| **Apertures** | f/1.4 (Shallow DoF), f/4 (Balanced), f/11 (Deep Focus) |
### π Upload History & Picker
Every image you upload is saved locally (URL + thumbnail) so you never upload the same file twice:
- Click the upload button to open the **reference image picker**
- Previously uploaded images appear in a 3-column grid with thumbnails
- **Single-image models** β click a thumbnail to instantly select and close
- **Multi-image models** β toggle multiple thumbnails (shown with order numbers), then click **Use Selected**
- Upload new images with the **Upload files** button (supports multi-file selection in multi-image mode)
- Remove individual images from history with the β button
- History persists across browser sessions (stored in `localStorage`)
## π Quick Start
### Prerequisites
- [Node.js](https://nodejs.org/) (v18+)
- A [Muapi.ai access key](https://muapi.ai/access-keys?utm_source=github&utm_medium=readme&utm_campaign=open-generative-ai). Copy the generated key value into the app; do not enter the key name or label.
### Setup
> **Most users want the desktop app, not this dev path.** If you just want to run Open Generative AI on your machine, [download a prebuilt installer](#-download-desktop-app) instead β no Node.js required. The instructions below are for contributors building from source.
Pick the entry point that matches your goal:
- **Desktop app (Electron)** β `npm run electron:dev`
- **Hosted web version (Next.js)** β `npm run dev`
```bash
# Clone the repository (with submodules β required for the workflow + agent packages)
git clone --recurse-submodules https://github.com/Anil-matcha/Open-Generative-AI.git
cd Open-Generative-AI
# If you already cloned without --recurse-submodules, run this once:
# git submodule update --init --recursive
# Install dependencies + build workspace packages (studio, workflow, agents).
# This step is REQUIRED β `npm install` alone is not enough; the workspaces
# need to be built before either dev script will work.
npm run setup
# Then start ONE of:
npm run electron:dev # Desktop app (Electron + Vite) β recommended
npm run dev # Hosted web version (Next.js) β http://localhost:3000
```
You'll be prompted to enter your Muapi API key on first use (skip the key if you only plan to use local models).
> **Troubleshooting β `Couldn't find a 'pages' directory`**: this means Next.js can't see the `app/` folder. Confirm you're running `npm run dev` from the repo root (the directory that contains `app/`, `package.json`, and `next.config.mjs`), and that you cloned with submodules. Re-run `npm run setup` if `packages/Vibe-Workflow` or `packages/agents` are empty.
### Production Build
```bash
npm run build
npm run start
```
### Desktop App Build
Build native desktop apps with Electron:
```bash
# macOS (DMG β Intel + Apple Silicon)
npm run electron:build
# Windows (NSIS installer β x64 + ARM64)
npm run electron:build:win
# Linux (AppImage + DEB β x64)
npm run electron:build:linux
# Both platforms in one pass
npm run electron:build:all
```
Installers are output to the `release/` folder. Pre-built binaries are also available on the [Releases page](https://github.com/Anil-matcha/Open-Generative-AI/releases).
## ποΈ Architecture
The app is a **Next.js monorepo** with a shared `packages/studio` component library.
```
Open-Generative-AI/
βββ app/ # Next.js App Router
β βββ layout.js # Root layout (Tailwind, fonts)
β βββ page.js # Redirects β /studio
β βββ studio/
β βββ page.js # Studio page β renders StandaloneShell
βββ components/
β βββ StandaloneShell.js # Tab nav + BYOK (API key from localStorage)
β βββ ApiKeyModal.js # API key entry modal
βββ packages/
β βββ studio/ # Shared React component library
β βββ src/
β βββ index.js # Exports: ImageStudio, VideoStudio, AudioStudio, ClippingStudio, VibeMotionStudio, LipSyncStudio, CinemaStudio, MarketingStudio, WorkflowStudio, AgentStudio, DesignAgentStudio, AppsStudio, McpCliStudio
β βββ models.js # 200+ model definitions (single source of truth)
β βββ muapi.js # API client (named exports, apiKey as first param)
β βββ components/
β βββ ImageStudio.jsx # Dual-mode t2i/i2i studio
β βββ VideoStudio.jsx # Dual-mode t2v/i2v studio
β βββ LipSyncStudio.jsx # Portrait/video + audio β talking video
β βββ CinemaStudio.jsx # Pro studio with camera controls
β βββ WorkflowStudio.jsx # Multi-step pipeline builder & playground
βββ next.config.mjs # transpilePackages: ['studio']
βββ tailwind.config.js
βββ package.json # workspaces: ["packages/studio"]
```
The `packages/studio` library is also consumed by the hosted version on [muapi.ai](https://muapi.ai?utm_source=github&utm_medium=readme&utm_campaign=open-generative-ai) β model updates made in `packages/studio/src/models.js` apply to both the self-hosted app and the hosted version automatically.
## π API Integration
The app communicates with [Muapi.ai](https://muapi.ai?utm_source=github&utm_medium=readme&utm_campaign=open-generative-ai) using a two-step pattern:
1. **Submit** β `POST /api/v1/{model-endpoint}` with prompt and parameters
2. **Poll** β `GET /api/v1/predictions/{request_id}/result` until status is `completed`
Authentication uses the `x-api-key` header. During development, a Vite proxy handles CORS by routing `/api` requests to `https://api.muapi.ai`.
File uploads use `POST /api/v1/upload_file` (multipart/form-data) and return a hosted URL that is passed to image-conditioned models. For multi-image models the full `images_list` array is forwarded to the API in one request.
Lip sync jobs use the same two-step pattern: a dedicated `processLipSync()` method accepts `image_url` or `video_url` alongside `audio_url`, dispatches to the model's endpoint, and polls until the output video URL is available.
## π¨ Supported Model Categories
| Category | Count | Examples |
|---|---|---|
| **Text-to-Image** | 50+ | Flux Dev, Nano Banana 2, Seedream 5.0, Ideogram v3, Midjourney v7, GPT-4o, SDXL |
| **Image-to-Image** | 55+ | Nano Banana 2 Edit (Γ14), Flux Kontext Pro, GPT-4o Edit, Seededit v3, Upscaler, Background Remover |
| **Text-to-Video** | 40+ | Kling v3, Sora 2, Veo 3, Wan 2.6, Seedance 2.0, Seedance 2.0 Extend, Seedance Pro, Hailuo 2.3, Runway Gen-3 |
| **Image-to-Video** | 60+ | Kling v2.1 I2V, Veo3 I2V, Runway I2V, Seedance 2.0 I2V, Midjourney v7 I2V, Hunyuan I2V, Wan2.2 I2V |
| **Lip Sync** | 9 | Infinite Talk I2V, Wan 2.2 Speech to Video, LTX 2.3 Lipsync, LTX 2 19B Lipsync, Sync, LatentSync, Creatify, Veed, Infinite Talk V2V |
## π οΈ Tech Stack
- **Next.js 14** β App Router, server components, fast dev server
- **React 18** β Studio UI components
- **Tailwind CSS v3** β Utility-first styling
- **npm workspaces** β Monorepo with shared `packages/studio` library
- **Muapi.ai** β AI model API gateway
## π€ How is this different from other AI Video Platforms?
**Open Generative AI** is a community-driven, open-source alternative that provides similar creative capabilities without the closed ecosystem:
| | Other providers | Open Generative AI |
| :--- | :--- | :--- |
| **Cost** | Subscription-based | Free (open-source) |
| **Content filters** | Yes β prompts blocked or altered | None |
| **Restrictions** | Platform guardrails enforced | Full creative freedom |
| **Models** | Proprietary | 200+ open & commercial models |
| **Multi-image input** | Limited | Up to 14 images per request |
| **Lip sync** | No | 9 models, image & video modes |
| **Hosted version** | Subscription | Free at [muapi.ai/open-generative-ai](https://muapi.ai/open-generative-ai?utm_source=github&utm_medium=readme&utm_campaign=open-generative-ai) |
| **Self-hosting** | No | Yes |
| **Customizable** | No | Fully hackable |
| **Data privacy** | Cloud-based | Your data stays local |
| **Source code** | Closed | MIT licensed |
## π License
MIT
## π Credits
Built with [Muapi.ai](https://muapi.ai?utm_source=github&utm_medium=readme&utm_campaign=open-generative-ai) β the unified API for AI image and video generation models.
---
**Deep Dive**: For more details on the "AI Influencer" engine, upcoming "Popcorn" storyboarding features, and the future of this project, read the [full technical overview](https://medium.com/@anilmatcha/).
---
*Looking for a free, open-source AI Video Platform? Open Generative AI is an open-source AI image and video generation studio β with no content filters that you can self-host, customize, and extend.*