ComfyUI-DLSS5-Enhancer-v15
NVIDIA DLSS 5 Neural Rendering and RTX Upscaling for ComfyUI.
ComfyUI-DLSS5-Enhancer-v15
Run NVIDIA DLSS 5 Neural Rendering over your frames and videos, directly inside ComfyUI.
DLSS 5 is NVIDIA's neural rendering pass. In a game, it reconstructs the material and lighting detail that real-time rendering has to leave out, such as skin subsurface scattering, light transmission through hair, and contact shadows. This node pack drives that same native renderer over video frames and image batches, with optional 1x to 4x RTX upscaling and DLSS Frame Generation.
It is not a filter that imitates a look. Frames go to the native renderer and come back reconstructed. The model reconstructs materials and recovers structure (skin, hair, fabric) that the source only implied, rather than just sharpening edges.
Unlike older community implementations that relied on ReShade injection, this pack drives the official v16.5 headless CLI (VE_CLI.exe), providing a stable, native, and highly optimized pipeline.
How it Works
This pack acts as a bridge between ComfyUI and the native DLSS 5 Visual Enhancer CLI.
- Neural Rendering: The node exports your frames or passes your video file path to
VE_CLI.exe. The CLI handles the D3D12 process, hosting the native NVIDIA runtime to reconstruct high-frequency material details. - RTX Upscaling: If requested, the reconstructed frames are passed through NVIDIA's RTX Super Resolution engine to intelligently enlarge the image up to 4x its original size.
- DLSS Frame Generation: Optionally interpolates video to a target FPS using DLSS Frame Generation for ultra-smooth playback. (Requires RTX 40 or 50 Series GPU).
- Video Pipeline: For video files, the CLI processes the entire file natively in a single pass. This allows the AI to maintain temporal consistency, preventing the flickering and jitter common in frame-by-frame upscaling, while efficiently handling audio and metadata.
Requirements
| Item | Requirement |
| --- | --- |
| GPU | NVIDIA RTX 20, 30, 40, or 50 Series. (GTX, AMD, and Intel GPUs are not supported as they lack the required Tensor Cores). |
| GPU (Frame Gen) | DLSS Frame Generation requires RTX 40 or 50 Series. Enable Hardware-accelerated GPU scheduling in Windows. |
| Driver | A recent NVIDIA Game Ready or Studio Driver (560.xx or newer recommended). |
| OS | Windows only. |
| ComfyUI | A recent build supporting standard IMAGE and VIDEO tensor types. |
Installation
Step 1: Install the Node Pack
Clone this repository into your ComfyUI/custom_nodes/ folder:
git clone https://github.com/Strix-3D/ComfyUI-DLSS5-Enhancer-v15.git
(Alternatively, if published to ComfyUI Manager, search for DLSS5-Enhancer-v15 and click Install).
Important for Manual Installations: If you cloned the repository manually (rather than using ComfyUI Manager), you must install the Python dependencies. Open a PowerShell window in your main ComfyUI folder and run:
.\python_embeded\python.exe -m pip install -r ComfyUI\custom_nodes\ComfyUI-DLSS5-Enhancer-v15\requirements.txt
Step 2: Download the Runtime & Accept the License
This node pack does not include the proprietary NVIDIA binaries. You must download them using the included installer script.
- Open a PowerShell window in your main ComfyUI folder (the one containing
python_embeded). - Run the following command:
.\python_embeded\python.exe ComfyUI\custom_nodes\ComfyUI-DLSS5-Enhancer-v15\install.py
- Read the License Notice: The script will display a list of third-party components (NVIDIA, FFmpeg, mpv, etc.) and their respective licenses.
- Accept the Terms: Type
yand press Enter to confirm you accept the terms and wish to proceed. - Wait for the ~700MB runtime to download and extract into the
runtime/folder.
Step 3: Restart ComfyUI
Once the installation is complete, restart ComfyUI. The nodes will now appear in your menu under the DLSS5_v15 category.
Nodes & Controls
This pack includes two highly optimized nodes:
1. DLSS5 Enhance (Image)
Takes an IMAGE batch, applies Neural Rendering, and optionally upscales it. Outputs an IMAGE batch.
2. DLSS5 Enhance (Video)
Takes a file path string, processes the video natively via the CLI, and saves the result to your output directory. Outputs the new file path string (output_path) and the frame count (frames).
Shared Controls
| Widget | Options / Range | Effect | | --- | --- | --- | | style | Default, Natural, Cinematic | Default: Balanced detail reconstruction.<br>Natural: Subtle enhancement, preserves original textures.<br>Cinematic: Aggressive contrast and micro-detail enhancement. | | strength | 0.00 to 2.00 (Default: 1.0) | The intensity of the Neural Rendering effect. 1.0 is the standard NVIDIA tuning. Values > 1.5 can cause AI "hallucinations" (weird artifacts). | | scale | 1.0 to 4.0 (Default: 1.0) | The RTX upscaling multiplier. Set to 1.0 to skip upscaling entirely and only apply Neural Rendering. |
Video File Specific Controls
| Widget | Options / Range | Effect |
| --- | --- | --- |
| video_path | STRING | Relative path to video file within ComfyUI input folder (e.g., my_video.mp4). Absolute paths are blocked for security. |
| codec | HEVC, H264, AV1, ProRes Proxy | The video encoder to use. H264 is recommended for maximum compatibility. |
| container | MKV, MP4, MOV | The output file container format. |
| quality | Auto, Good, Best, Max | Encoding quality/bitrate. |
| target_fps | INT (0-120, Default: 0) | 0 = disabled. Target FPS for DLSS Frame Generation interpolation. (Requires RTX 40/50 Series GPU). |
| filename_prefix | STRING | Prefix for the output filename (a timestamp is automatically appended). |
| output_directory | STRING | Relative subfolder path within ComfyUI output directory. Leave blank to use the default output folder. |
| copy_audio | BOOLEAN | Whether to mux the original audio into the output file. |
Recommended Settings
The defaults below are starting points based on how the v16.5 native runtime behaves:
| Goal | Settings |
| --- | --- |
| Clean up generated video, same resolution | Defaults: scale 1.0, style Default, strength 1.0 |
| Clean up and upscale | Same as above, with scale 2.0 |
| Smooth 30fps to 60fps | Defaults plus target_fps 60 (RTX 40/50 only) |
| Most realistic faces | Defaults plus style Natural, strength 0.8 |
| More contrast and punch | style Cinematic, strength 1.0 (check the shadows) |
| Output looks too strong/plastic | Drop strength to 0.6 or 0.7, or switch style to Natural |
Note: The neural pass works on what is in the frame. It reconstructs skin and material response, but it does not restore a face that the source no longer contains. For heavily degraded footage, run a generative restoration pass first and use this node as the final cleanup.
Example Workflows
Image Enhancement
Because the DLSS5 nodes focus purely on AI reconstruction and upscaling, it is highly recommended to use standard ComfyUI nodes for post-processing.
<img width="1255" height="883" alt="dlss5_example" src="https://github.com/user-attachments/assets/c38672f6-a992-4598-81cd-274112ee62c6" />Load Image ➔ DLSS5 Enhance (Image) ➔ Color Correction / Sharpening ➔ Save Image
Video File Enhancement
The video node accepts a relative file path and outputs a new file path. To preview the result and preserve audio, feed the output path into a Load Video (Path) node, and connect both the IMAGE and audio outputs to Video Combine.
DLSS5 Enhance (Video) ➔ (output_path) ➔ Load Video (Path) ➔ (IMAGE & audio) ➔ Video Combine
Troubleshooting
| Message / Issue | Cause and Fix |
| --- | --- |
| Windows SmartScreen blocks VE_CLI.exe | This is normal for unsigned community tools. Right-click the .exe in the runtime folder, select Properties, check Unblock, and click OK. |
| Runtime not found | The ~700MB runtime hasn't been downloaded. Run install.py manually and accept the license. |
| Node fails on GTX / AMD cards | These GPUs lack the physical Tensor Cores required for DLSS/RTX AI processing. The nodes will not work on these cards. |
| Frame Generation fails on RTX 20/30 | DLSS Frame Generation strictly requires an RTX 40 or 50 Series GPU. Set target_fps to 0 to disable it. |
| Video processing is very slow | Neural Rendering is computationally heavy. Processing a long video natively through the AI pipeline takes time. Ensure no other heavy GPU tasks (like games) are running. |
| Output looks "plastic" or weird | Your strength is likely too high. Drop it to 0.7 or 1.0, or switch the style to Natural. |
Project Layout
ComfyUI-DLSS5-Enhancer-v15/
install.py Downloads the runtime and prompts for license acceptance
nodes.py The ComfyUI node definitions (Image & Video File)
paths.py Runtime discovery and validation
__init__.py ComfyUI extension entrypoint
README.md This file
LICENSE MIT License for the Python source code
requirements.txt Python dependencies (imageio-ffmpeg)
runtime/ Native runtime (VE_CLI.exe, NVIDIA DLLs) - *Not tracked in Git*
config.json Local path configuration - *Not tracked in Git*
Licensing and Attribution
The Python source code in this repository is licensed under the MIT License.
This repository does not host or redistribute the AI runtime binaries.
The install.py script downloads the official DLSS 5 Visual Enhancer v16.5 release from Merserk/dlss5-visual-enhancer. The downloaded runtime contains proprietary components owned by their respective creators:
- NVIDIA Corporation: DLSS Super Resolution, Neural Rendering, Frame Generation, and RTX Video SDKs (Proprietary).
- Merserk: Visual Enhancer CLI and Engine (Merserk Source License).
- FFmpeg/mpv: Media processing binaries (GPLv2/v3).
By running the installer and typing y, you agree to use these components only under the terms permitted by their respective licenses. The worker protocol was reimplemented for this pack. The runtime and the original application come from Merserk/dlss5-visual-enhancer, which makes NVIDIA's renderer usable outside a game engine. This project is not affiliated with or endorsed by NVIDIA or that project.