Visual Enhancer: DLSS 5 Upscaling for Images, Video, and Live Playback
Explore Visual Enhancer, a portable Windows tool combining DLSS 5, RTX Video Super Resolution, HDR, frame generation, and live playback.
Why Visual Enhancer matters
Upscaling a low-resolution image is easy to demonstrate but difficult to make consistently useful. Different media needs different processing: a photograph may benefit from denoising and detail reconstruction, while compressed video needs temporal cleanup, frame interpolation, HDR conversion, and audio-preserving export. Visual Enhancer brings these operations into one portable Windows application for NVIDIA RTX GPUs.
The project is built around configurable pipelines rather than a single “enhance” button. You can arrange individual stages, enable only the operations required for a job, preview the result, and export one file or an entire batch. The same application also offers Live playback for local videos, direct streams, YouTube, and Twitch.
Installation
Visual Enhancer is distributed as a portable application:
- Download the latest release from the GitHub repository.
- Extract the complete ZIP archive to a folder.
- Run
Visual Enhancer.exe.
The application requires 64-bit Windows 11 with Direct3D 12 and a compatible NVIDIA GeForce RTX GPU. Current NVIDIA drivers are required for NVIDIA-powered stages. DLSS 5 Neural Rendering availability depends on supported hardware, driver support, and successful initialization on the selected AI Processing GPU.
Hardware-accelerated GPU scheduling should be enabled for DLSS Frame Generation. RTX Video Super Resolution and RTX Video HDR require their respective runtimes. Video EWA Lanczos scaling additionally requires an available Vulkan device and an FFmpeg build with libplacebo support.
A pipeline designed for experimentation
Image and Video have independent stage lists. Stages execute from top to bottom, and each card can be enabled, disabled, or dragged into a different position. The initial image order is:
Denoising → DLSS Neural Rendering → Scaling → DLSS Super Resolution
→ RTX Super Resolution → Coloring → Sharpening
The video pipeline also includes DLSS Frame Generation before Coloring. These are available stages, not mandatory effects. Rendering performs preflight checks after every enabled stage, validating dimensions, frame rates, hardware capability, and HDR compatibility.
This ordering matters. Video Scaling, DLSS Super Resolution, and RTX Super Resolution cannot process HDR video at arbitrary positions, so they should normally run before HDR conversion. Nonzero video sharpening must also precede HDR conversion. A Frame Generation target must be higher than the frame rate entering that stage.
DLSS 5 Neural Rendering
The Neuroframe Engine makes DLSS 5 Neural Rendering available directly inside Visual Enhancer without an external graphics injector or game add-on. It can process still images, videos, and Live playback.
Controls include:
- NR Style: Default, Natural, or Cinematic
- NR Intensity:
0.00–2.00 - Local Tone Strength:
0.00–2.00 - Local Structure Strength:
0.00–2.00 - Skin Structure Strength:
-1.00–2.00 - Automatic Mask: optional face and skin masking
- NR Passes: one to four passes
- NR Color Strength, Tone Preservation, Face/Skin Protection, and Grain Preservation
- Mask Feather:
0–128output pixels - Shimmer Suppression:
0.00–1.00for video and Live
The Neural Rendering stage operates at the dimensions supplied by earlier stages. Its input must be at least 64×64, while supported Neural Rendering processing is bounded at 7680×4320. Use the separate Scaling stage when you want to change dimensions before rendering.
The Detail-Only preset is useful when you want reconstructed structure without a strong color transformation. It sets NR Color Strength to 0.00 and Tone Preservation to 1.00, while leaving the other controls editable. A custom image mask can restrict where Neural Rendering is applied; the same mask is available to Live during the current application session.
Additional passes can produce a stronger cumulative effect, but they increase processing time. For video, Shimmer Suppression helps stabilize fine detail between frames and reduce distracting temporal sparkle.
Upscaling options are deliberately separate
Visual Enhancer includes three different approaches that should not be confused:
- Conventional Scaling changes dimensions using filters such as Lanczos4, Area, Bicubic, Bilinear, or Nearest for images, and Spline36, Lanczos, Bicubic, Area, Bilinear, or EWA Lanczos for video.
- DLSS Super Resolution provides DLAA at
1×, Quality at1.5×, Balanced at approximately1.72×, Performance at2×, and Ultra Performance at3×. Presets J, K, L, and M are available in addition to the Default preset. - RTX Video Super Resolution offers quality levels 1 through 4 and output scaling from
1×to4×, plus custom dimensions.
RTX Video Super Resolution can enhance an image at native resolution with 1× sizing, or enlarge it while applying reconstruction. Output dimensions are limited to 16384 pixels per dimension, and custom dimensions cannot be smaller than the incoming image when VSR is enabled.
HDR conversion and video output
The RTX Super Resolution video stage also contains RTX Video HDR. VSR and HDR can run together, or VSR can be disabled for HDR-only processing. HDR controls include contrast and saturation from 0–200, middle gray from 10–100, peak luminance from 400–2000 nits, and Packed 10-bit or Packed 10-bit FP16 processing.
RTX Video HDR accepts SDR input rather than already-HDR video. To export HDR, enable 10-bit HDR Mode and select a compatible codec. Supported HDR-capable outputs include H.265, AV1, ProRes Proxy, ProRes HQ, and FFV1 Lossless RGB 10-bit.
Available video outputs are:
| Codec | Container | Notes |
|---|---|---|
| H.264 | MP4 | CPU or NVIDIA NVENC, 8-bit SDR |
| H.265 | MKV | CPU or NVENC, supports 10-bit HDR |
| AV1 | MKV | CPU or NVENC, supports 10-bit HDR |
| ProRes Proxy | MOV | CPU-based 10-bit 4:2:2 |
| ProRes HQ | MOV | CPU-based 10-bit 4:2:2 |
| FFV1 Lossless RGB 10-bit | MKV | CPU-based lossless 10-bit RGB |
The default codec is H.264 with NVIDIA NVENC. Encoding quality can be Auto, Good, Best, or Max; ProRes HQ and FFV1 use codec-fixed quality.
Frame generation from 23.976 to 480 FPS
DLSS Frame Generation is a reorderable Video stage. It can create intermediate frames at targets ranging from 23.976 through 480 FPS, including 60, 120, 144, 240, and 360 FPS. The selected target must exceed the frame rate entering the stage; otherwise preflight rejects the pipeline.
The DLSSG engine offers Auto, Native DLSSG, and Cascade modes. Before committing to a full export, render a three-, five-, ten-, twenty-, or thirty-second preview from the selected video position. This is especially useful for checking motion artifacts and ensuring the target frame rate is appropriate.
Live playback
Live mode applies Neural Rendering while watching local videos or supported online sources. It supports direct network streams, YouTube, and Twitch, with controls for playback, volume, mute, expanded viewing, and fullscreen mode.
Important Live settings include:
- Source quality from Auto through 2160p
- Maximum input resolution from 480p through 2160p
- One-, two-, or four-second segments
- Auto, Source, 60, 30, or 24 FPS modes
- A playback buffer from 2 to 30 seconds
- Independent Live Scale from 25% through 200%
Live keeps its own Neural Rendering settings, so tuning playback does not overwrite the main Image or Video configuration. Most Neural Rendering controls can change while playback is active. Source, quality, resolution, segment length, target FPS, buffer, and Live Scale changes take effect the next time Live starts.
Preview, batching, and export safety
The unified media viewer supports Split, 2-Up, and Output comparisons, 100% inspection, image fitting, and video timeline scrubbing. Realtime Preview can refresh supported changes automatically, while a smart preview cache reuses unchanged intermediate results when you adjust later stages.
For production work, the batch queue accepts files, folders, drag-and-drop media, clipboard bitmap images, copied local files, and supported browser image URLs. Failed items can be retried, completed items cleared, and finished files revealed in Explorer. Each queue item reports state, progress, elapsed time, dimensions, processing details, and output path.
Images can be exported as PNG, JPEG, WebP, AVIF, or TIFF. PNG and TIFF support 16-bit output. Video exports preserve source audio, metadata, and chapters where the container and codec allow it, but subtitle streams are not mapped by the unified exporter. Image exports do not preserve original EXIF metadata, although decoding applies EXIF orientation, handles supported color profiles, and retains transparency where the format supports it.
By default, completed files are written to outputs/, while troubleshooting logs are stored in logs/. Existing files are not silently overwritten, and incomplete output is cleaned up when necessary.
Practical starting configurations
For a clean still-image enhancement, start with Denoising, DLSS Neural Rendering, and moderate sharpening. Add Scaling or RTX Super Resolution only if the output needs a larger resolution.
For older SDR video, a sensible order is:
Denoising → DLSS Neural Rendering → RTX Super Resolution/HDR
→ DLSS Frame Generation → Coloring → Sharpening
Adjust this based on the source and preflight requirements. If the goal is HDR conversion rather than enlargement, disable VSR while leaving RTX Video HDR enabled. For high-frame-rate playback, preview Frame Generation before exporting and verify that the chosen codec and HDR mode match the intended display workflow.
Project status and licensing
The repository is primarily implemented in Python (81.7%) and QML (18.2%), with Lua accounting for 0.1%. At the time of the supplied repository snapshot, it had 1.1k stars, 90 forks, 16 watchers, 68 commits, 14 releases, and one listed contributor. The latest release is Visual Enhancer v11.0.
Project-owned material is distributed under the Merserk Source License 1.0. Personal, professional, and commercial use is permitted, including commercial use of processed outputs. Redistribution, repackaging, rebranding, resale, sublicensing, publishing modified builds, and similar activities require prior written permission unless the license states otherwise. Third-party components remain governed by their own licenses; consult LICENSE and THIRD-PARTY-NOTICES before redistributing the application.
Visual Enhancer is an independent community project and is not affiliated with, sponsored by, or endorsed by NVIDIA. NVIDIA, GeForce RTX, DLSS, and RTX Video are trademarks or registered trademarks of NVIDIA Corporation.