Blur-first by default
The default workflow detects faces and blurs them for local video redaction. Open a video, keep the default, export.
FaceForge
Blur faces in a video without uploading it anywhere. FaceForge is a local-only Windows app for AI face blur and video redaction, optimized for NVIDIA RTX GPUs.
FaceForge is built for people who need to anonymize faces in local media without uploading footage anywhere. The default workflow is blur and redaction. Face replacement exists as an advanced, consent-based Pro workflow, and it is not positioned as a generic deepfake tool.
- Windows 10/11
- NVIDIA RTX + CUDA
- Local-only
- Blur-first
- Public alpha
Why it exists
Redaction without uploads
Most face-blur tools are cloud services: you hand over the footage to protect the people in it. FaceForge takes the opposite approach. Detection, blurring, enhancement, and export all run on your own GPU, so the media that needs protecting never leaves the machine.
GPU acceleration
Optimized for NVIDIA RTX GPUs with CUDA so longer videos process on your own hardware instead of a render queue.
Optional face enhancement
GFPGAN or CodeFormer enhancement can be enabled for higher quality results when a workflow needs it.
Live preview, responsive UI
Real-time preview during processing, with an interface that stays responsive through long exports.
How it works
Video in, redacted video out
- 01
Add a video
Drop a local file into the app. Nothing is uploaded anywhere.
- 02
Keep the default workflow
Blur Faces / Privacy is the default. Adjust blur strength if needed.
- 03
Start Export
The redacted video renders on your GPU and lands next to your files.
Face selection
Three modes, from zero-friction to precise
Auto
Zero-friction default: acts on the largest detected face in each frame.
All Faces
Covers every detected face in each frame, for group shots where everyone should be anonymized.
Locked Face
Advanced mode for multi-person scenes: pick one person in the preview and only they are followed.
The inspector
Tuned with studio controls, not checkboxes
Blur strength, mask boundaries, edge softness, and enhancement are continuous controls in a studio-style inspector. Set what to mask, pick a quality mode, and preview Original, Processed, or Split before you commit to an export.
Consent and privacy model
Local by default, consent-gated by design
App state, logs, cache, and support artifacts live in local per-user folders, and diagnostics are exported only when you choose to. The capable-but-risky feature, face replacement, is deliberately gated rather than deliberately easy.
- Footage, logs, and diagnostics stay on your machine
- Blur and redaction is the default, beginner-facing workflow
- Face replacement is an advanced Pro workflow, blocked in the Free tier
- Replacement requires consent acknowledgement before export
- Diagnostics and repro bundles are exported locally, only when you ask
System requirements
Built for RTX-class Windows machines
- Windows 10 or 11
- NVIDIA RTX GPU (tested on an RTX 3080 Ti with 12GB VRAM)
- CUDA Toolkit 11.8 or 12.x with matching cuDNN
- 16GB RAM minimum, 32GB recommended
Current status
FaceForge v0.1.0 Alpha
The first public alpha is available for Windows/NVIDIA testers. FaceForge focuses on local face blur and video redaction, with consent-based face replacement as an advanced workflow. The current release tier is FaceForge Free: a portable desktop bundle centered on the privacy-first blur flow, with face replacement, enhancement, and alternate-backend controls kept as advanced and Pro surfaces.
The alpha is distributed through itch.io. The GitHub release contains release notes, SHA-256 checksum verification, and support links. The NVIDIA ZIP is hosted externally because the CUDA runtime exceeds GitHub and itch.io file size limits.
Good fit for
Who it is for
- Creators and editors blurring bystanders in footage
- Educators and researchers sharing classroom or study video
- Businesses and agencies redacting client media
- Anyone anonymizing local video without cloud uploads
Not trying to be
Scope, stated plainly
FaceForge is not a cloud service and not a generic deepfake tool. Face replacement stays behind consent acknowledgement and rights to the media involved, and the free tier is about hiding faces, not swapping them. The operator keeps control of the footage from start to finish.
Redact faces without uploading footage
FaceForge is being built in the open enough to hold it accountable: privacy-first defaults, local processing, and honest limits. The v0.1.0 public alpha is available for Windows/NVIDIA testers on itch.io, with release notes and checksums on GitHub.