Skip to content
Windows desktop · v2.2.1

Forge faceless videos with AI. Publish everywhere.

Give VidForge AI a topic. It researches the subject, writes the script, builds the shot list, voices the narration, finds the footage, cuts the edit, renders on your GPU and uploads the result to every account you connected.

Download for WindowsSee how it works

Free to download. Bring your own API keys — you pay providers directly, at their rates.

  • YouTube
  • TikTok
  • Facebook

Publishes to

  • YouTube
  • TikTok
  • Facebook

Instagram and X are in development.

Built on

Real technology, named honestly.

Every provider and library below is declared in the application's own dependency manifest. We show no customer logos and claim no endorsements, because we would have to invent them.

Language models

  • Anthropic Claude

    Research, scripts and SEO metadata

  • OpenAI GPT

    Scripts, storyboards and scene analysis

Speech and transcription

  • ElevenLabs

    Cloud voice generation

  • VoxCPM2

    Local voice generation and cloning

  • Kokoro

    Lightweight local text-to-speech

  • OpenAI Whisper

    On-device subtitle transcription

Stock media

  • Pexels

    Stock video and image search

  • Pixabay

    Fallback stock media source

Rendering and runtime

  • FFmpeg

    Compositing, audio ducking and encoding

  • NVIDIA NVENC

    Hardware video encoding

  • Intel Quick Sync

    Hardware video encoding

  • AMD AMF

    Hardware video encoding

  • CUDA 12

    GPU inference for local voice models

  • PyTorch

    Local model execution

  • PySide6 / Qt

    Native desktop interface

  • SQLite

    Local project and publishing databases

Publishing APIs

  • YouTube Data API v3

    Channel uploads

  • TikTok Content Posting API

    Inbox draft delivery

  • Facebook Graph API

    Page video publishing

Verifiable

Every technology listed here is declared in the application's own requirements.txt or .env.template. We list no customer logos and no endorsements, because we would have to invent them.

These are integrations, not affiliations. VidForge AI is independent and is not endorsed by or sponsored by any provider named above. You hold your own account with each one.

By the numbers

Engineering figures, not vanity metrics.

Each of these is countable from the source tree. None of them depends on a user count we would rather not publish.

  • 7

    Pipeline stages

    Research through publish, each resumable

  • 15

    Subsystems

    Purpose-built engines, not one model call

  • 3

    Publishing platforms

    YouTube, TikTok and Facebook, live today

  • 4

    Voice providers

    Two cloud, two fully on-device

  • 4

    Interface languages

    English, Spanish, Japanese, Khmer

  • 7

    Visual fallback steps

    A scene can never render empty

  • 4

    Render profiles

    Fast, balanced, high quality, lossless

  • 0

    Servers we operate

    Nothing to upload, nothing to breach

Features

Everything a faceless video needs, in one window.

Fifteen subsystems, from research to upload. No plugin hunting, no round trip through a browser editor, no per-render credits.

All features in detail

The pipeline

One topic in. Seven stages. A published video out.

This is the pipeline the application actually runs, in order, with the timings you can expect on recommended hardware. Any stage can be inspected, re-run or resumed.

  1. Research & Script

    ~30s

    You give a topic and a niche. The script engine researches the subject, then writes a hook, body and payoff sized to your target duration.

  2. Storyboard & Scenes

    ~15s

    The storyboard engine turns the script into a shot list — shot type, camera angle, mood, energy, colour palette and search keywords per beat.

  3. Voice

    ~1–3 min

    Narration is generated with your chosen provider. Long scripts are chunked, quality-checked, retried on failure, crossfade-merged and loudness-normalised to −16 LUFS.

  4. Footage

    ~1–2 min

    Each scene is matched to stock video or images and semantically ranked. A seven-step fallback chain means a scene never ends up empty.

  5. Subtitles

    ~30s

    Whisper transcribes the finished narration locally for word-accurate caption timing — no guessing from the script.

  6. Render

    ~2–6 min

    Cuts, transitions, Ken Burns motion, colour grade, music bed and ducking are assembled, then encoded on your GPU via NVENC, QuickSync or AMF.

  7. Publish

    ~1–4 min

    The Publishing Center writes SEO metadata, queues the export and uploads it to every account you connected — on a schedule, or the moment it finishes.

Publishing

Connect the accounts you already run.

Authorisation uses each platform's official OAuth flow. Tokens are encrypted and stored on your machine, and nothing uploads until you start a job.

  • YouTube

    Available

    Uploads a rendered video to a channel you own, and reads the channel name and avatar so you can tell connected accounts apart.

    Scopes requested

    • https://www.googleapis.com/auth/youtube.upload
    • https://www.googleapis.com/auth/youtube.readonly
    • https://www.googleapis.com/auth/userinfo.email

  • TikTok

    Available

    Sends a rendered video to your TikTok inbox as a draft, and reads your display name and avatar so you can tell connected accounts apart.

    Scopes requested

    • video.upload
    • user.info.basic

  • Facebook

    Available

    Publishes a rendered video to a Facebook Page you manage, and lists your Pages so you can pick the destination.

    Scopes requested

    • pages_show_list
    • pages_manage_posts
    • pages_read_engagement

  • Instagram

    In development

    Planned. Instagram Reels publishing through the Instagram Graph API is in development and is not part of v2.2.1.

  • X

    In development

    Planned. Video posting to X is in development and is not part of v2.2.1.

You create your own OAuth application in each platform's developer console and paste the credentials into Settings. Authorisation happens on the platform's own consent screen, in your own browser, and the redirect is received by a loopback server on your computer — nothing passes through a server we operate, because we don't operate one.

Inside the app

Built like professional software, not a wizard.

Four working surfaces: what's running now, what one job is doing, what's queued for upload, and which voices you've saved.

VidForge AI — Dashboard
  • Active job — Deep sea creaturesRendering · 68%
  • Queued jobs4 waiting
  • Recent exports12 this week
  • EncoderNVENC verified
  • Provider healthAll reachable

What is running now, what finished, and whether every provider is reachable.

Interface diagrams, not screenshots — drawn to match the application's real working surfaces and kept in step with each release rather than shipped as images that go stale.

Desktop or cloud

Why this runs on your machine.

A comparison of approaches rather than products. Three of the twelve dimensions favour a hosted service, and they are marked as such.

Where rendering happens

Desktop
Your GPU, via NVENC, Quick Sync or AMF. No queue.
Cloud
A shared render farm, subject to queue position at peak times.

Where your footage lives

Desktop
On your disk. Nothing is uploaded to render it.
Cloud
Uploaded to the provider, then downloaded back once rendered.

Model access

Desktop
Your own API keys. Switch provider or model per job.
Cloud
Whichever model the vendor selected, at the tier you bought.

Cost structure

Desktop
Provider rates at cost, with no markup and no per-render fee.
Cloud
Subscription plus credits, priced above underlying provider cost.

Output ceiling

Desktop
Bounded by your hardware and provider quotas.
Cloud
Bounded by your plan's monthly credit allocation.

Working offline

Desktop
Voice and subtitle stages run locally with no connection.
Cloud
Nothing works without a connection.

Confidential material

Desktop
Prompts go straight to the provider you contracted with.
Cloud
Passes through the vendor's systems under their retention policy.

Batch generation

Desktop
Queue dozens of topics overnight with a shared asset cache.
Cloud
Usually rate-limited or metered per render.

Multi-account publishing

Desktop
Unlimited connected accounts with isolated browser profiles.
Cloud
Commonly limited per seat or per plan tier.

Setup effort

Desktop
Install, then register your own keys and OAuth applications.
Cloud
Sign up and start immediately.

Hardware requirement

Desktop
Needs a reasonably capable Windows machine; a GPU helps a lot.
Cloud
Runs on any device with a browser.

Maintenance

Desktop
You update the app and manage your own keys and drivers.
Cloud
The vendor handles updates and infrastructure.

Three of these twelve dimensions favour a cloud workflow, and they are marked accordingly. If you need to work from a tablet, avoid managing API keys, or start producing in the next five minutes, a hosted service is the better fit and we would rather say so here than have you discover it after installing.

Who it's for

Different reasons to stop editing by hand.

The bottleneck is rarely ideas. It's the four hours between having one and having a finished, captioned, uploaded video.

  • YouTubers

    One video a week, and most of it is editing.

    Queue seven topics on Sunday, wake up to seven finished videos with captions, music and metadata already written. Your time goes back into what the videos are about.

  • Content creators

    Every platform wants a different aspect ratio and caption style.

    Render once per aspect ratio from the same job, with caption styling tuned per format, then publish each version to the account it belongs on.

  • Marketing agencies

    Client accounts can't share credentials or browser sessions.

    Unlimited connected accounts per platform, isolated Chrome profiles per identity, and a local history of every upload attempt you can show a client.

  • Businesses

    Scripts and product details can't go through someone else's SaaS.

    There is no intermediary. Prompts go straight from your machine to the AI provider you chose under your own contract, and nothing is retained by us.

  • AI automation builders

    Per-seat pricing and rate limits you don't control.

    Your own API keys, your own quotas, your own hardware. Batch mode and folder watchers make it a component in a larger pipeline rather than a destination.

  • Anyone tired of subscriptions

    Monthly fees for credits you may not use.

    The application is free to download. You pay OpenAI, Anthropic, ElevenLabs or Pexels directly, at their published rates, only for what you actually generate.

Output volume and quality depend on your hardware, the providers you configure and the topics you choose. Nothing here is a guarantee of views, revenue or platform performance.

Questions

The things people ask first.

More in the help centre, including troubleshooting and account setup.

Your next hundred videos are one topic list away.

VidForge AI 2.2.1 for Windows 10 / 11 (64-bit). Free to download, no account required, and your projects never leave your machine.

Download for WindowsExplore every feature