Open-Generative-AI

Open Generative AI — Unrestricted Open-Source Alternative to AI Video Platforms

The free, open-source alternative to AI Video Platforms. Generate AI images and videos using 600+ state-of-the-art models across 14 studios — no content filters, no closed ecosystem, no subscription fees.

Community: Join Discord for discussions and support

🎨 Explore 50+ more open-source AI apps →

🤝 New: Open Duet — open-source, self-hosted AI coworkers (named agents) with instructions, memory, routines, their own isolated computer, and your approval before they act. An alternative to Claude Duet, OpenAI Dots, and Manus Cue. Star it on GitHub →

💰 Turn This Into Your Own Product — White Label & Resell

Want to launch this as your own branded AI studio and charge your own customers for it? MuAPI White Label lets you spin up a fully white-labeled version of this app — your logo, your colors, your custom domain, your own pricing — with zero infra to manage. You keep the markup on every generation; MuAPI handles the models, the queue, and the billing plumbing underneath.

• Your branding — logo, color theme, and a custom domain (e.g. studio.yourbrand.com)
• Your pricing — set your own credit/subscription prices for end users, keep the margin
• No infra — no servers, workers, or model hosting to run yourself
• All studios included — Image, Video, Audio, Lip Sync, Cinema, Workflows, and more, depending on plan

Plans start at $49/mo. Get started with White Label →

What similar AI studios charge their users

Consumer AI image/video platforms almost all run on paid monthly subscriptions — this is the same playbook you'd run under your own brand:

Platform	Typical subscription range
Midjourney	~$10–$120/mo (Basic → Mega)
Runway	~$12–$76/mo (Standard → Unlimited), custom Enterprise
Kling AI	~$10–$92/mo across Standard → Premier tiers
Luma Dream Machine	~$10–$100+/mo
Pika	~$8–$58/mo

(Figures are approximate, general-market ranges and change over time — check each platform's current pricing page before quoting them.) With MuAPI White Label, you set these numbers yourself for your own end users — the subscription revenue is yours.

---

API examples and model references

• OpenAI API examples — GPT Image, Sora, and GPT endpoints through Muapi.
• Google Gemini Media API examples — Nano Banana, Imagen, Veo, Gemini Omni, and speech endpoints.
• MiniMax Media API examples — Hailuo, H3 Max, speech, and music workflows.
• Kling Video API examples — Kling text-to-video, image-to-video, and motion control.
• Qwen Image API examples — Qwen image generation, editing, and LoRA workflows through Muapi.

Related Projects

This is a curated set of high-value hubs, popular distribution tools, and model-specific integrations rather than a directory of every related repository.

• awesome-vibecoded-saas — broader catalog of open-source SaaS alternatives featuring this studio.
• Muapi open-source alternatives — compare this studio with the Midjourney workflow and its honest scope.
• awesome-generative-ai-apps — catalog of open-source generative-AI applications.
• awesome-ai-video-models — compare video models by API, price, and capability.
• awesome-ai-image-models — compare image models by API, price, and quality.
• AI-Youtube-Shorts-Generator — Open-source Opus Clip alternative application.
• open-duet — Open-source, self-hosted AI coworkers with memory, routines, isolated computers, and approval-gated actions.
• Open-AI-Design-Agent — Ppen-source autonomous AI design agent.
• muapi-skills — Agent-ready skills for driving generative-media workflows.
• Seedance-2.5-API — Python SDK for Seedance 2.5 video generation.

🌐 Try it Online — No Install Required

Hosted version: 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 — 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 for updates

---

⬇️ Download Desktop App

One-click installers — no Node.js or terminal required.

Platform	Download
macOS Apple Silicon (M1/M2/M3/M4)	Open Generative AI-2.0.0-arm64.dmg
macOS Intel (x64)	Open Generative AI-2.0.0.dmg
Windows (x64)	Open Generative AI Setup 2.0.0.exe
Linux (Ubuntu x64)	.AppImage / .deb, or build locally with npm run electron:build:linux.

All releases: 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 — Mount the DMG and drag the app to /Applications

Step 2 — Open Terminal and run:

xattr -cr "/Applications/Open Generative AI.app"

Step 3 — Right-click the app in /Applications → click Open → 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 — macOS will block it
2. Go to System Settings → Privacy & Security
3. Scroll down to find "Open Generative AI was blocked"
4. Click Open Anyway → Open

Windows Installation — 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:

# Build Linux installers (AppImage + .deb)
npm run electron:build:linux

Generated files are written to the release/ folder:

• AppImage — portable, run directly after making executable:

  chmod +x "release/Open Generative AI-*.AppImage"
  ./release/Open\ Generative\ AI-*.AppImage
  

• .deb — install on Debian/Ubuntu:

  sudo apt install ./release/open-generative-ai_*_amd64.deb
  

If AppImage fails to start on older systems, install libfuse2:

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 — 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 — Temporary system fix (AppImage users):

sudo sysctl -w kernel.apparmor_restrict_unprivileged_userns=0

This lasts until next reboot. To make it permanent:

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 — just full creative freedom. Powered by Muapi.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 — all from a sleek, modern interface you can self-host and customize.

Why Open Generative AI instead of other AI Video Platforms?

• No filters — no content filters, no nanny guardrails, no prompt rejections
• Free & open-source — no subscription, no vendor lock-in
• Self-hosted — your data stays on your machine, full creative control
• 200+ models — text-to-image, image-to-image, text-to-video, image-to-video, lip sync
• Multi-image input — feed up to 14 reference images into compatible models
• Lip Sync Studio — animate portraits or sync lips to any audio with 9 dedicated models
• Extensible — 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.

⚡ 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, 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 install)	A user-installed Wan2GP: either its folder on this machine (the app runs wgp.py itself) or its MCP server on another machine. Python + PyTorch on a CUDA/ROCm GPU.	Video models (Wan 2.1/2.2, Hunyuan, LTX-2) in Video Studio and large image models (Flux, Qwen-Image). NVIDIA/AMD GPU required on the Wan2GP machine; the desktop app itself can run on a Mac.

Both engines share the same UI: open Settings → Local Models to configure each.

Engine 1 — sd.cpp (bundled)

Model	Type	Size	Notes
Z-Image Turbo ⚡	Diffusion Transformer	2.5 GB + 2.7 GB aux	8-step turbo. Heavy on memory.
Z-Image Base ⚡	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 — 2.4 GB
• FLUX VAE — 335 MB

How to use:

1. Open Settings → Local Models in the desktop app
2. Install the sd.cpp inference engine (one click — auto-downloaded)
3. Download your chosen model (and auxiliary files for Z-Image)
4. In Image Studio, click the ⚡ Local toggle next to the model selector
5. Select your local model and generate — 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 — Wan2GP (your own install)

The app does not bundle Python or model weights for Wan2GP. Install Wan2GP yourself on a machine with a CUDA or ROCm GPU:

git clone https://github.com/deepbeepmeep/Wan2GP
cd Wan2GP
./install.sh                          # or install.bat on Windows

Same machine (recommended). In the desktop app open Settings → Local Models → Wan2GP, set the Wan2GP folder (Python defaults to <folder>/venv/bin/python), click Save, then Check — it runs wgp.py --dry-run on a test task. Each generation runs Wan2GP's headless CLI (wgp.py --process settings.json); weights download automatically the first time a model is used, and the download shows up in the progress bar.

Another machine. Start WanGP's MCP server there and put its URL (e.g. http://192.168.1.42:7866/mcp) in the same settings panel:

python wgp.py --mcp --mcp-api-version 1 --mcp-transport streamable-http --mcp-host 0.0.0.0 --mcp-port 7866

Use --mcp-host 0.0.0.0 only on a trusted network (see Wan2GP's docs/AUTHENTICATION.md for OAuth/HTTPS). The Gradio web UI port (7860) is not an API and cannot be used here.

Then switch Video Studio to ⚡ Local and pick a model. Models your Wan2GP version doesn't have are listed with the reason; models whose weights aren't downloaded yet are marked.

Model	WanGP model_type	Type	Notes
Wan 2.1 1.3B	t2v_1.3B	Video	Lightest option, 480p, ~8 GB VRAM
Wan 2.2 14B T2V / I2V	t2v_2_2 / i2v_2_2	Video	480p, slow on consumer GPUs; I2V needs a start frame
Hunyuan Video 1.5	hunyuan_1_5_480_t2v	Video	480p T2V
LTX-2 Distilled	ltx2_22B_distilled	Video	8 steps, 720p
Flux.1 Dev	flux	Image	Image Studio
Qwen Image	qwen_image_20B	Image	Image Studio

Why the remote option? Wan2GP's runtime (Sage attention, flash-attn, AWQ/GGUF kernels) is CUDA/ROCm-only — there is no MPS / Apple Silicon path. The MCP 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 — 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 — stick to SD 1.5 there.
• For SD 1.5 on M2: expect ~1–2 s/step with the Metal dylib active. If you see ~10 s/step instead, the binary may have fallen back to CPU — 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.

# 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 512×512 PNG. If VRAM is 0.00MB instead, the dylib is CPU-only — check otool -L "$APP_DATA/bin/libstable-diffusion.dylib" | grep -i metal and reinstall the engine from Settings → Local Models if Metal is missing.

---

✨ Features

• Image Studio — 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 — 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 (your own install — local folder or remote MCP server) for Flux, Qwen-Image, and video models in Video Studio (Wan 2.1/2.2, Hunyuan, LTX-2). Configure both in Settings → Local Models.
• Multi-Image Input — 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 — 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.
• Audio Studio — Generate and edit AI audio/music from text prompts.
• AI Clipping — Auto-clip and extract highlights from longer video content.
• Vibe Motion Studio — Motion/animation generation studio for stylized video effects.
• Lip Sync Studio — Animate portrait images or sync lips on existing videos using audio. 9 dedicated models across two modes: portrait image + audio → talking video, and video + audio → lipsync video.
• Body Swap (Recast) Studio — Swap/recast a subject's body or appearance in an image or video.
• Cinema Studio — Interface for photorealistic cinematic shots with pro camera controls (Lens, Focal Length, Aperture)
• Marketing Studio — Generate ad and marketing-ready creative variations from a single input.
• Workflow Studio — 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.
• Agent Studio — Multi-turn creative agent that plans and executes generation tasks conversationally.
• Design Agent Studio — Canvas-based autonomous design agent for iterative visual work.
• Explore Apps — Directory of app templates and use-cases built on the same model catalog.
• AI Influencer Studio — Tools for creating and managing consistent AI persona/influencer content.
• Upload History — Reference images are uploaded once and stored locally. A picker panel lets you reuse any previously uploaded image across sessions — no re-uploading.
• Smart Controls — 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 — Browse, revisit, and download all past generations (persisted in browser storage)
• Image & Video Download — One-click download of generated outputs in full resolution
• API Key Management — Secure API key storage in browser localStorage (never sent to any server except Muapi)
• Responsive Design — Works seamlessly on desktop and mobile with dark glassmorphism UI

🖼️ Image Studio — 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, Midjourney…)	Required
Image-to-Image	Reference image uploaded	55+ i2i models (Kontext, Nano Banana 2 Edit, Seedream 5.0 Edit, Seededit, Upscaler…)	Optional

Newly Added Models

Model	Type	Key Features
Nano Banana 2	Text-to-Image	Google Gemini 3.1 Flash Image · Resolution 1K/2K/4K · Google Search enhancement · aspect ratio auto
Nano Banana 2 Edit	Image-to-Image	Up to 14 reference images · Resolution 1K/2K/4K · Google Search enhancement
Seedream 5.0	Text-to-Image	ByteDance · Quality basic/high · 8 aspect ratios · up to 4K
Seedream 5.0 Edit	Image-to-Image	ByteDance · Natural language style transfer · Quality basic/high
MiniMax Image 01	Text-to-Image	MiniMax · 8 aspect ratios · up to 4 images per request · 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	2–3
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 — images are sent to the model in the order you select them
• Batch upload — 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

🎬 Video Studio — 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, Runway…)	Required
Image-to-Video	Start frame uploaded	60+ i2v models (Kling I2V, Veo3 I2V, Runway I2V, Wan I2V, Seedance 2.0 I2V, Midjourney I2V…)	Optional

Newly Added Models

Model	Type	Key Features
Seedance 2.0	Text-to-Video	ByteDance · Aspect ratios 16:9 / 9:16 / 4:3 / 3:4 · Duration 5 / 10 / 15s · Quality basic/high
Seedance 2.0 I2V	Image-to-Video	ByteDance · Animate images into video · Up to 9 reference images · Aspect ratios 16:9 / 9:16 / 4:3 / 3:4 · Duration 5 / 10 / 15s · Quality basic/high
Seedance 2.0 Extend	Video Extension	ByteDance · Seamlessly continue any Seedance 2.0 generation · Preserves style, motion & audio · Optional continuation prompt · Duration 5 / 10 / 15s · Quality basic/high
Grok Imagine T2V	Text-to-Video	xAI · Duration 6 / 10 / 15s · Modes: fun / normal / spicy · Aspect ratios 9:16 / 16:9 / 2:3 / 3:2 / 1:1
Grok Imagine I2V	Image-to-Video	xAI · Duration 6 / 10 / 15s · Modes: fun / normal / spicy · Cinematic motion from still images
MiniMax Hailuo 02 / 2.3 Standard & Pro	Text-to-Video / Image-to-Video	MiniMax · Full HD video · Multiple aspect ratios · Fast variant included

🎙️ 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 → animated talking video
Video	Switch to Video mode	Upload an existing video + audio file → lipsync video

Image-based Models (Portrait Image + Audio → 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 → Lipsync Video)

Model	Endpoint	Resolutions	Prompt
Sync Lipsync	sync-lipsync	—	—
LatentSync	latentsync-video	—	—
Creatify Lipsync	creatify-lipsync	—	—
Veed Lipsync	veed-lipsync	—	—
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.

🔀 Workflow Studio

The Workflow Studio lets you build and run multi-step AI pipelines without writing code.

Key capabilities:

• Templates — Start from pre-built workflows (image chains, video pipelines, and more)
• My Workflows — Save and manage your own custom pipelines
• Community — Browse and run workflows published by other users
• Node-based Builder — Drag-and-drop visual editor to connect models and route outputs between steps
• Playground — Run any workflow interactively with a form UI; results render inline
• API execution — Every workflow is also callable via the Muapi API

💡 Want to add workflows to your own app? Check out Vibe Workflow — the open-source workflow engine powering this feature. Drop it into any project.

🎥 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)

📁 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 — click a thumbnail to instantly select and close
• Multi-image models — 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 ✕ button
• History persists across browser sessions (stored in localStorage)

🚀 Quick Start

Prerequisites

• Node.js (v18+)
• A Muapi.ai access key. 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 instead — no Node.js required. The instructions below are for contributors building from source.

Pick the entry point that matches your goal:

• Desktop app (Electron) → npm run electron:dev
• Hosted web version (Next.js) → npm run dev

# Clone the repository (with submodules — 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 — `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) — recommended
npm run dev            # Hosted web version (Next.js) → 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 — 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

npm run build
npm run start

Desktop App Build

Build native desktop apps with Electron:

# macOS (DMG — Intel + Apple Silicon)
npm run electron:build

# Windows (NSIS installer — x64 + ARM64)
npm run electron:build:win

# Linux (AppImage + DEB — 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.

🏗️ Architecture

The app is a Next.js monorepo with a shared packages/studio component library.

Open-Generative-AI/
├── app/                        # Next.js App Router
│   ├── layout.js               # Root layout (Tailwind, fonts)
│   ├── page.js                 # Redirects → /studio
│   └── studio/
│       └── page.js             # Studio page — renders StandaloneShell
├── components/
│   ├── StandaloneShell.js      # Tab nav + BYOK (API key from localStorage)
│   └── ApiKeyModal.js          # API key entry modal
├── packages/
│   └── studio/                 # Shared React component library
│       └── src/
│           ├── index.js        # Exports: ImageStudio, VideoStudio, AudioStudio, ClippingStudio, VibeMotionStudio, LipSyncStudio, RecastStudio, CinemaStudio, MarketingStudio, WorkflowStudio, AgentStudio, DesignAgentStudio, AppsStudio, AiInfluencerStudio, McpCliStudio
│           ├── models.js       # 400+ model definitions (single source of truth)
│           ├── muapi.js        # API client (named exports, apiKey as first param)
│           └── components/
│               ├── ImageStudio.jsx    # Dual-mode t2i/i2i studio
│               ├── VideoStudio.jsx    # Dual-mode t2v/i2v studio
│               ├── LipSyncStudio.jsx  # Portrait/video + audio → talking video
│               ├── CinemaStudio.jsx   # Pro studio with camera controls
│               └── WorkflowStudio.jsx # Multi-step pipeline builder & playground
├── next.config.mjs             # transpilePackages: ['studio']
├── tailwind.config.js
└── package.json                # workspaces: ["packages/studio"]

The packages/studio library is also consumed by the hosted version on muapi.ai — model updates made in packages/studio/src/models.js apply to both the self-hosted app and the hosted version automatically.

🔌 API Integration

The app communicates with Muapi.ai using a two-step pattern:

1. Submit — POST /api/v1/{model-endpoint} with prompt and parameters
2. Poll — 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.

🎨 Supported Model Categories

Category	Count	Examples
Text-to-Image	70+	Flux Dev, Nano Banana 2, Seedream 5.0, Ideogram v3, Midjourney v7, GPT-4o, SDXL
Image-to-Image	70+	Nano Banana 2 Edit (×14), Flux Kontext Pro, GPT-4o Edit, Seededit v3, Upscaler, Background Remover
Text-to-Video	85+	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	120+	Kling v2.1 I2V, Veo3 I2V, Runway I2V, Seedance 2.0 I2V, Midjourney v7 I2V, Hunyuan I2V, Wan2.2 I2V
Video-to-Video	35+	Video effects, AI Clipping, Vibe Motion, video-conditioned edits
Lip Sync	15	Infinite Talk I2V, Wan 2.2 Speech to Video, LTX 2.3 Lipsync, LTX 2 19B Lipsync, Sync, LatentSync, Creatify, Veed, Infinite Talk V2V
Body Swap / Recast	3	Subject/appearance recast across image and video
Audio	15+	Text-to-music, remix, and audio editing models

(Counts verified against packages/studio/src/models.js — total 420+ models across these 8 categories, plus additional models surfaced through Marketing, Agent, and Design Agent studios.)

🛠️ Tech Stack

• Next.js 14 — App Router, server components, fast dev server
• React 18 — Studio UI components
• Tailwind CSS v3 — Utility-first styling
• npm workspaces — Monorepo with shared packages/studio library
• Muapi.ai — AI model API gateway

🤔 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 — prompts blocked or altered	None
Restrictions	Platform guardrails enforced	Full creative freedom
Models	Proprietary	400+ 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
Self-hosting	No	Yes
Customizable	No	Fully hackable
Data privacy	Cloud-based	Your data stays local
Source code	Closed	MIT licensed

📄 License

MIT

🙏 Credits

Built with Muapi.ai — 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.

---

Looking for a free, open-source AI Video Platform? Open Generative AI is an open-source AI image and video generation studio — with no content filters that you can self-host, customize, and extend.