Version 2.2.1 · Released 24 July 2026
Download VidForge AI
A single installer for Windows 10 / 11 (64-bit). No account, no licence key to activate, and nothing phones home. You supply your own API keys and pay providers directly.
VidForgeAI-Setup-2.2.1-x64.exe
412 MB · Windows 10 / 11 (64-bit) · Includes FFmpeg and the Whisper model
Only download VidForge AI from vidforgeai.online. We do not distribute through software aggregators, torrent sites or third-party download portals, and any build you find there is not one we published. Verify the installer's digital signature before running it.
Before you install
System requirements
The minimum column will produce videos. The recommended column is what makes batch mode and local voice models comfortable.
Minimum
- Operating system
- Windows 10 64-bit (build 19041) or later
- Processor
- 4-core x86-64 CPU
- Memory
- 8 GB RAM
- Storage
- 12 GB free space (SSD recommended)
- Graphics
- Any GPU — VidForge AI falls back to CPU encoding
- Network
- Broadband connection for AI providers and uploads
Recommended
- Operating system
- Windows 11 64-bit
- Processor
- 8-core CPU or better
- Memory
- 16 GB RAM (32 GB for local voice models)
- Storage
- 50 GB free NVMe space for footage cache and exports
- Graphics
- NVIDIA RTX with 8 GB+ VRAM and CUDA 12 for NVENC and local voice
- Network
- 25 Mbps upload for batch publishing
Bundled with the installer
- FFmpeg build used for encoding, audio ducking and Ken Burns motion
- SQLite databases for the render pipeline and the Publishing Center
- OpenAI Whisper model files for on-device subtitle transcription
Hardware in detail
Will it run well on your machine?
The summary table above is the short answer. This is the long one, so you can check the specific parts that matter before downloading.
Graphics and hardware encoding
Rendering uses your GPU's dedicated video encoder where one is available. At startup VidForge AI runs a real one-frame test encode against each candidate and selects the first that genuinely works, logging the exact reason any was rejected.
- NVIDIA
- GTX 1050 or newer with NVENC. RTX with 8 GB+ VRAM recommended.
- Intel
- 6th-generation Core or newer with Quick Sync Video.
- AMD
- RX 400 series or newer with AMF.
- No GPU
- Supported. Encoding falls back to CPU with the reason logged.
- Drivers
- Keep graphics drivers current — stale drivers are the most common cause of hardware encoding being unavailable.
CUDA and local voice models
CUDA is only required if you run voice generation locally with VoxCPM2. Cloud voices, Kokoro on CPU, and every other stage work without it.
- CUDA runtime
- 12.0 or later for GPU inference.
- VRAM
- 8 GB or more for comfortable VoxCPM2 inference.
- CPU fallback
- Supported and considerably slower. Kokoro is the lighter CPU option.
- Apple silicon
- MPS is detected by the voice engine, but the application itself ships for Windows.
- Not required for
- Scripts, storyboards, footage search, subtitles, rendering or publishing.
FFmpeg
A tested FFmpeg build ships with the installer, so there is nothing to install separately and no PATH configuration to get wrong.
- Bundled
- Yes. No separate download or system install needed.
- Used for
- Compositing, Ken Burns motion, transitions, audio ducking and final encoding.
- System FFmpeg
- Ignored by default. The bundled build is used to guarantee a known encoder set.
Memory, CPU and storage
Batch mode and local voice models are the memory-hungry parts. Everything else is comfortable on a mid-range machine.
- CPU minimum
- 4-core x86-64.
- CPU recommended
- 8-core or better for parallel batch rendering.
- RAM minimum
- 8 GB.
- RAM recommended
- 16 GB, or 32 GB when running local voice models alongside a batch.
- Storage
- 12 GB free to install. 50 GB NVMe recommended for the footage cache and exports.
- Network
- Broadband to reach providers. 25 Mbps upload or better for batch publishing.
Integrity
Check what you downloaded.
Desktop software asks you to run an executable with full user privileges. Verifying it first is a reasonable thing to expect, and we make it straightforward.
Verify your download
Confirm the installer you received is the one we published, and that it was not modified in transit or by a mirror.
- File name
- VidForgeAI-Setup-2.2.1-x64.exe
- SHA-256 digest
Not yet published for this build
We publish the digest here with each signed release. We deliberately do not show a placeholder value — a hash that always matches would train you to skip the check that matters. Until it appears, verify the installer's Authenticode signature in its Windows file properties instead, and only download from vidforgeai.online.
- How to check on Windows
certutil -hashfile VidForgeAI-Setup-2.2.1-x64.exe SHA256
Run it in Command Prompt from the folder containing the installer, then compare the output with the digest above. They must match exactly.
Installation
Five steps to your first render.
Budget about twenty minutes, most of it waiting for the installer and gathering keys.
Download and run the installer
Windows SmartScreen may warn you about an unrecognised publisher on a newly signed build. Choose More info, then Run anyway, after confirming the download came from vidforgeai.online.
Let it install FFmpeg and the local models
The installer places a bundled FFmpeg build and the Whisper subtitle model alongside the application. This is the slowest step and needs a few gigabytes of free space.
Add your provider keys
Open Settings and paste in your keys. At minimum you need one text provider — Anthropic or OpenAI — plus either an ElevenLabs key or a local voice model, and a Pexels key for stock footage.
Register your publishing apps (optional)
To upload, create your own OAuth application in the Google, TikTok or Meta developer console and paste the client ID and secret into Settings. Register the loopback port the app displays.
Run one job end to end
Press New Video, enter a topic, and watch the pipeline view. Getting one video through before configuring batch mode makes any misconfiguration obvious.
Credentials
What you'll need to supply.
These are your accounts with your own quotas. Nothing is resold through us, and no keys ship inside the installer.
| Provider | Status | What it's for |
|---|---|---|
| Anthropic or OpenAI | Required | Research, scripts, storyboards and SEO. |
| ElevenLabs, or a local model | Required | Narration. VoxCPM2 and Kokoro run locally at no per-word cost. |
| Pexels | Required | Stock video and image search. |
| Pixabay | Recommended | Second free source so scenes rarely fall back. |
| Platform OAuth apps | Optional | Only needed to publish. Rendering works without them. |
Keys are stored in your own configuration file on your own disk, and OAuth tokens are encrypted at rest with a key held only on your machine. See the security section of the Privacy Policy for exactly how.
Release notes · v2.2.1
Publishing Center hardening and multilingual interface
Released 24 July 2026. Earlier versions are in the full changelog.
- added
Interface localisation for English, Spanish, Japanese and Khmer.
- added
Publishing diagnostics tab reporting per-account OAuth health with a classified failure reason for every attempt.
- added
Chrome profile isolation so separate publishing identities no longer share browser state.
- improved
TikTok authorisation now uses hex-encoded PKCE challenges as TikTok's desktop flow requires, fixing token exchange rejections.
- improved
Token exchange surfaces the platform's own error text instead of failing with an opaque key error.
- improved
Uploads interrupted by a crash are reset to pending on next launch rather than retried automatically.
- fixed
Voice metadata commits that succeeded on the GPU but failed to persist are now replayed at startup before the window opens.
- security
OAuth token encryption key permissions tightened to owner-only where the operating system supports it.
Trouble installing? The help centre covers the common failures, or write to support@vidforgeai.online.