NextGenUp is a free, open-source AI photo and video studio: upscale images to 8K and videos to 4K, remove backgrounds, erase unwanted objects, denoise and deblur photos, and colorize old black-and-white pictures — using Real-ESRGAN, BiRefNet, LaMa, SCUNet, NAFNet, DDColor, and GFPGAN face restoration, entirely on your own machine. A free alternative to Topaz Gigapixel AI, Topaz Video AI, remove.bg, and Magic Eraser with no cloud, no subscription, no watermarks, and no upload of your files to anyone's server.
A matily.org product. Free and open source under the MIT license.
| NextGenUp | Topaz Photo / Video AI | remove.bg | Upscayl | Video2X | |
|---|---|---|---|---|---|
| Price | Free, MIT | ~$199/year | ~$0.20+/image | Free | Free |
| Image upscaling | Yes, to 8K | Yes | No | Yes | No |
| Video upscaling | Yes, to 4K | Yes | No | No | Yes |
| Background removal | Yes (BiRefNet) | No | Yes | No | No |
| Object eraser (magic eraser) | Yes (LaMa) | No | No | No | No |
| Denoise and deblur | Yes (SCUNet / NAFNet) | Yes | No | No | No |
| Old photo colorization | Yes (DDColor) | No | No | No | No |
| Face restoration | Yes (GFPGAN) | Yes | No | No | No |
| Same-size enhance mode | Yes | Yes | No | No | No |
| Batch + zip export | Yes | Yes | Paid API | Yes | Partial |
| Runs fully offline | Yes | Yes | No (cloud) | Yes | Yes |
| Also usable from any browser on your network | Yes | No | n/a | No | No |
Excellent tools all around — NextGenUp's niche is covering images, video, faces, backgrounds, and restoration in one simple app that is free forever.
| Platform | File | Notes |
|---|---|---|
| macOS (Apple Silicon) | Download .dmg | M1/M2/M3/M4 Macs |
| macOS (Intel) | Download .dmg | x64 build |
| Windows 10/11 (x64) | Download installer | NSIS .exe or .msi |
| Linux (x64) | Download .AppImage / .deb | Most distributions |
Every installer is fully self-contained — the AI engine, FFmpeg, and the upscaling models are bundled. Install, open, upscale. The app checks for updates on launch and installs them with one click (signed releases).
macOS note: builds are not notarized with Apple; on first launch, right-click the app and choose Open.
Watch the 30-second demo — a full tutorial is coming soon on our YouTube channel, subscribe to catch it.
| Image upscaling | Four modes — Quick (browser AI), Quality (FFmpeg), Enhance (same size, AI cleanup), Ultra (server AI, works from any device) |
| Background remover | BiRefNet segmentation with feathered edges — transparent PNG, white, or any custom color background |
| Denoise and deblur | SCUNet removes grain and ISO noise; NAFNet removes motion blur and camera shake |
| Magic Eraser | Brush over people, wires, trash, or photobombers and LaMa inpainting removes them — repeat as often as you like, with undo |
| Old Photo Restore | DDColor colorizes black-and-white photos; combine with face restoration and detail enhancement in one pass |
| Video upscaling | Basic (FFmpeg Lanczos + CAS sharpening) and Pro (Real-ESRGAN, frame by frame), single or bulk |
| Enhance mode | Keep the original resolution but reconstruct detail — noise, blur, and JPEG artifacts removed |
| Face restoration | GFPGAN v1.4 — detected faces are aligned, restored, and seamlessly blended back |
| Batch processing | Drop multiple images or videos, process them all, download results individually or as one zip |
| On-demand AI models | The optional models download on first use with a progress bar, so the installer stays small |
| Before/after slider | Drag a divider across the result to compare against the original |
| Cancellable jobs | Every job can be cancelled mid-run; switch modes and restart instantly |
| Portrait and landscape | Orientation handled automatically; output capped at 4K (video) / 8K (image) |
| Auto-updates | Desktop app updates itself from GitHub releases (cryptographically signed) |
-
Open NextGenUp and pick a tool in the sidebar:
Tool What it does Image Upscale Quick (browser AI), Quality (FFmpeg), Enhance (same size), or Ultra (server AI) — 2x or 4x, up to 8K Background Remover Cut the subject out to a transparent PNG, or place it on white or any color Denoise & Deblur Remove Noise for grainy low-light photos; Remove Blur for motion blur and camera shake Magic Eraser Brush over anything unwanted, press Erase, repeat — with undo Old Photo Restore Colorize black-and-white photos, restore faces, and enhance detail in one pass Video Upscale Basic (fast FFmpeg) or Pro (frame-by-frame AI) — drop several videos for a batch queue -
Drop a file onto the dropzone (or click to browse — select several files for batch mode).
-
Press the action button. If a tool's AI model is not installed yet, it downloads once with a progress bar and then runs.
-
Watch the live progress (cancel any time), compare the result, and download — single files or everything as one zip.
For portraits, tick Restore faces (Ultra mode and Old Photo Restore) to run GFPGAN face restoration on every detected face.
The same interface also runs as a self-hosted web app that any device on your network can use.
Requirements: Python 3.10–3.13 and FFmpeg (brew install ffmpeg / apt install ffmpeg).
git clone https://github.com/riponcm/nextgenUp.git
cd nextgenUp
./setup.sh
source venv/bin/activate
python app.pyOpen http://localhost:5000, or http://<your-ip>:5000 from other devices. Browser AI modes need WebGPU (Chrome/Edge 113+) on secure origins; the Ultra mode works from any browser on any device.
Browser / app window Local server
-------------------- ------------
upload image or video ───────────► Pillow / ffprobe read metadata
Quality & Basic modes ───────────► FFmpeg: Lanczos scale + CAS + unsharp
Quick / Enhance / Pro
ONNX Runtime Web (WebGPU/WASM)
Real-ESRGAN in a Web Worker
tile-based inference Ultra mode
frames sent back ───────────► Real-ESRGAN via ONNX Runtime (CPU)
GFPGAN face restoration
FFmpeg assembles video + audio
The upscaling model is Real-ESRGAN (realesr-general-x4v3, ~5 MB ONNX, bundled). The other tools use optional ONNX models that download on first use into the app's data folder: BiRefNet-lite (~224 MB, background removal), SCUNet (~77 MB, denoise), NAFNet (~92 MB, deblur), DDColor-tiny (~135 MB, colorization), and LaMa (~208 MB, object removal) — all MIT or Apache-2.0 licensed. Images are processed in overlapping tiles or windows so any resolution fits in memory; inference runs in a Web Worker (browser) or a background thread (server) so the interface never freezes. Completed tasks persist in SQLite, and files older than 24 hours are cleaned up automatically.
| Environment variable | Default | Purpose |
|---|---|---|
PORT |
5000 |
Server port |
CLEANUP_HOURS |
24 |
Files older than this are deleted |
FLASK_DEBUG |
0 |
Set to 1 for the Werkzeug debugger (never on an exposed host) |
Upload limit is 2 GB. Face restoration requires the GFPGAN model (~340 MB): answer yes in setup.sh, or for the desktop app drop GFPGANv1.4.onnx into the app-data models/ folder.
app.py Flask server: upload, process, progress, download, zip
ai_engine.py Server-side Real-ESRGAN (ONNX Runtime, tiled)
bg_remove.py BiRefNet background removal
restore_engine.py SCUNet denoise, NAFNet deblur, DDColor colorize, LaMa inpaint
model_store.py Registry + on-demand downloader for the optional AI models
face_restore.py GFPGAN face restoration + YuNet detection
paths.py Dev vs. packaged path resolution
templates/index.html Dashboard (all tools)
static/js/app.js Video controller (single + batch), navigation, About dialog
static/js/image-app.js Image upscale controller: modes, batch, compare slider
static/js/bg-app.js Background remover controller
static/js/tools-app.js Denoise/deblur, Magic Eraser, and Old Photo Restore controllers
static/js/pro-worker.js ONNX inference worker (tiling, WebGPU/WASM)
src-tauri/ Desktop shell (Tauri 2): sidecar, auto-updater
nextgenup-server.spec PyInstaller build for the bundled server
.github/workflows/ CI: signed installers for all platforms
Releases are built by GitHub Actions. Bump the version in src-tauri/tauri.conf.json and app.py, then:
git tag v1.0.1
git push origin v1.0.1Installers for macOS (Intel and Apple Silicon), Windows x64, and Linux are compiled, signed for auto-update, and attached to the GitHub release automatically. Existing installations pick the update up on next launch.
- Frame interpolation and AI slow motion (RIFE) — 24 to 60/120 fps
- Video background removal (Robust Video Matting)
- Larger server models (Real-ESRGAN x4plus, HAT) for a maximum-quality tier
- Face restoration for video frames
- One-button automatic photo restoration pipeline
Is NextGenUp really free? Yes. NextGenUp is completely free and open source under the MIT license — no trial, no subscription, no watermark, no account. It is maintained by Matily.
Is NextGenUp a good free alternative to Topaz Gigapixel AI or Topaz Video AI? Yes, for most use cases. NextGenUp uses Real-ESRGAN — the same family of super-resolution models used by many commercial tools — and covers both image and video upscaling plus GFPGAN face restoration in one app. Topaz still holds an edge on some fine detail at extreme zoom, but for web, social media, prints, and AI-generated media, results are comparable at zero cost.
Does NextGenUp upload my photos or videos to the cloud? No. Everything runs locally on your machine — the AI models, FFmpeg, and the server all execute on your own hardware. Your files never leave your computer.
What AI models does NextGenUp use? Real-ESRGAN for super-resolution, BiRefNet for background removal, LaMa for object removal, SCUNet for denoising, NAFNet for deblurring, DDColor for colorization, GFPGAN v1.4 for face restoration, and YuNet for face detection — all running locally through ONNX Runtime, all under MIT or Apache-2.0 licenses.
Is NextGenUp a free alternative to remove.bg? Yes. The Background Remover uses BiRefNet, a state-of-the-art segmentation model, and outputs a transparent PNG (or any background color you choose) with no per-image fees, no signup, and no upload — including batch removal with zip export.
How do I remove an object or a person from a photo? Open Magic Eraser, drop your photo, brush over whatever you want gone, and press Erase. LaMa inpainting reconstructs the background behind the object. You can brush and erase repeatedly, and undo any step.
Can it colorize old black-and-white photos? Yes. Old Photo Restore uses DDColor to add realistic color to black-and-white or sepia photos. Tick "Restore faces" and "Enhance detail" to also repair faces and recover sharpness in the same pass.
Do the AI models make the app download huge? No. The installer ships only the small upscaling model. The optional tools (background removal, denoise, deblur, colorize, eraser) each download their model once on first use, with a progress bar, ranging from 77 MB to 224 MB. Tools you never open download nothing.
How do I upscale a video to 4K? Open the Video Upscale tab, drop in your video, choose Basic (fast) or Pro (AI), select 4x, and press Start Upscaling. A 720p video becomes 4K (3840x2160); portrait videos are handled automatically. Audio is preserved.
Can it fix blurry faces in old photos? Yes. Select Ultra mode and tick "Restore faces" — every detected face is aligned, restored with GFPGAN, and blended back seamlessly. This works especially well on old, scanned, or compressed photos.
Can I make an image clearer without changing its size? Yes — that is the Enhance mode. The AI upscales 4x internally to reconstruct detail, then downscales back to the original dimensions, removing noise, blur, and JPEG artifacts while keeping the exact resolution.
What platforms does NextGenUp support? Desktop apps for macOS (Apple Silicon and Intel), Windows 10/11 x64, and Linux (AppImage/deb). It also runs as a self-hosted web app usable from any browser on your network, including phones and tablets.
Do I need a GPU? No. Browser modes use WebGPU when available and fall back to WASM; the server modes run on CPU. A GPU makes things faster but nothing requires one.
How is NextGenUp different from Upscayl or Video2X? Upscayl upscales images only; Video2X handles video but is more technical to set up. NextGenUp covers images, video, and face restoration in one point-and-click app, and can also serve its interface to other devices on your network.
If NextGenUp is useful to you, please star the repository — it is the simplest way to help other people discover a free alternative to paid upscalers. Sharing the project with anyone who works with old photos, AI-generated media, or low-resolution video helps just as much. Issues and pull requests are welcome.
NextGenUp stands on excellent open work:
- Real-ESRGAN (Xintao Wang et al.) — super-resolution models
- BiRefNet (Zheng Peng et al.) — background segmentation
- LaMa (Samsung AI) — large-mask inpainting for the Magic Eraser
- SCUNet (Kai Zhang et al.) — real-world image denoising
- NAFNet (Megvii Research) — image deblurring
- DDColor (Xiaoyang Kang et al.) — photo colorization
- GFPGAN (Tencent ARC) — face restoration
- YuNet (OpenCV Zoo) — face detection
- FFmpeg — video processing (bundled builds are GPL; FFmpeg keeps its own license)
- ONNX Runtime — AI inference on server and in the browser
- Tauri — desktop shell
- Flask — local server
Model weights keep the licenses of their upstream projects (MIT / Apache-2.0); they are downloaded from their public releases on first use, never bundled or re-hosted.
Built and maintained by Matily, with projectmem providing persistent project memory during development.
MIT © 2026 Ripon Chandra Malo / matily.org