GUI v.1.4b
🎬 SeedVR2.5 GUI by HB2k — v.1.4 beta
A professional, high-end AI Video & Image Upscaling GUI for SeedVR2 — no command line needed.
Instantly upscale footage up to 4×, preview quality before committing, and compare results side-by-side in full screen.
📋 Table of Contents
✨ What's New in v1.4 Beta
📊 Dual Status Metadata
Real-time display of both Input and Output file information simultaneously in the status bar. Instantly see resolution, codec, duration, and size for both the source and the upscaled result — no need to switch views.
🖼️ Advanced Preview System
One-click frame capture from video to instantly test upscale quality with Batch Size 1:
- Pause the video at any frame using the built-in player
- Click "📸 Preview Frame" to extract that exact frame
- The GUI automatically runs a single-frame upscale so you can judge quality before committing to a full render
🖥️ Full Screen Comparison (Split View)
Professional Topaz-style side-by-side comparison now available in dedicated full-screen mode:
- Toggle between Input, Output, and Split View from the control bar
- Full-screen mode maximizes the comparison area for detailed quality inspection
- Drag the split divider to reveal more of either the original or upscaled feed
⚡ VRAM Optimization
expandable_segments is automatically enabled for NVIDIA RTX 30/40/50 series GPUs, delivering:
- Reduced VRAM fragmentation during long encode/decode runs
- More stable memory usage across multiple upscale batches
- Better performance on GPUs with ≥ 8 GB VRAM (e.g. RTX 5070 Ti, 4090, 3090)
🔗 GitHub Link
A dedicated "GitHub" button inside the GUI provides direct one-click access to this repository for updates, issues, and discussions.
🚀 Installation & Setup
No Python environment setup required if you use the packaged
.exe.
Step 1 — Download the GUI
Download SeedVR2_GUI.exe from the dist/ folder in this repository.
Step 2 — Select your Python interpreter
On first launch, click "Browse…" next to the Python field and point it to your python.exe. Both options are supported:
- Embedded Python (recommended): the standalone Python bundled alongside SeedVR2
- System Python: any Python 3.12+ installation on your machine
Step 3 — Point to the SeedVR2 core folder
Click "Browse…" next to the SeedVR2 Folder field and select the directory that contains inference_cli.py. This is the core SeedVR2 engine the GUI drives.
Step 4 — Done! Settings are saved automatically
All paths and preferences are persisted via Qt QSettings — just launch the .exe next time and you're ready to go.
🎮 Usage
Recommended Workflow
📂 Load File → 📸 Preview Frame → ⚙️ Adjust Settings → ▶️ Run Full Upscale
| Step | Action | Description |
|---|---|---|
| 1 | 📂 Open | Click the folder icon in the player bar or use Settings → Input to load a video or image |
| 2 | ⏯️ Seek | Scrub to a representative frame using the playback controls |
| 3 | 📸 Preview Frame | Click the preview button — the GUI captures the frame and runs a Batch Size 1 upscale |
| 4 | 🔍 Split View | Switch to Split View mode to compare input vs output at full resolution |
| 5 | ⚙️ Adjust | Tune upscale factor, resolution cap, DiT quantisation, VAE offload, and output format |
| 6 | Click Start to upscale the full file — dual progress bars show per-batch and overall progress |
🔍 Split View Comparison
- Use the [Input] [Output] [Split View] buttons in the control bar to switch modes
- In Split View, drag the divider left/right for a live before/after comparison
- Click ⛶ Full Screen to enter full-screen comparison mode
🎛️ Key Settings
| Setting | Default | Notes |
|---|---|---|
| Resolution | 720p | Output height cap |
| Batch Size | 1 | Frames per inference step (increase for speed, decrease for VRAM) |
| DiT Quantisation | Q8_0 | Balances quality and VRAM |
| GPU Device | Auto | Multi-GPU selection supported |
| Output Format | Auto-detect | H.264 / H.265 / AV1 / PNG / JPG / WEBP / TIFF |
💻 System Requirements
| Component | Minimum | Recommended |
|---|---|---|
| GPU | NVIDIA RTX 20 series, 8 GB VRAM | RTX 5070 Ti / 4090 (Blackwell / Ada Lovelace) |
| VRAM | 8 GB | 16 GB+ |
| Python | 3.12 | 3.12 (embedded environment preferred) |
| OS | Windows 10 64-bit | Windows 11 64-bit |
| CUDA | 11.8+ | 12.x |
| RAM | 16 GB | 32 GB+ |
⚠️ Note: AMD / CPU / MPS inference is possible via the underlying SeedVR2 CLI but the GUI is optimised for NVIDIA hardware.
🙏 Credits & Links
| 🔗 Repository | https://github.com/naxci1/ComfyUI-SeedVR2.5_new |
| 👤 GUI Author | HB2k — GUI design, Windows integration, preview system, split-view comparison |
| 🏗️ Core Engine | SeedVR2 by ByteDance Seed Team |
| 🔌 ComfyUI Node | ComfyUI-SeedVR2_VideoUpscaler by numz |
Special thanks to the original SeedVR research and engineering team at ByteDance for releasing SeedVR2 and enabling community-built tools like this GUI.
Built with ❤️ on top of SeedVR2 — the state-of-the-art open-source video restoration model.