Skip to content

Presets, Models & Settings

Visual Enhancer separates workflow configurations (Presets), downloadable weights and runtimes (Models), and machine-specific hardware choices (Settings).

Use the Presets menu in the header bar to save and reload complete workflow setups:

  • Export Preset: Saves active Image and Video pipeline card orders, toggles, parameters, ReShade overrides, 3D LUT adjustments, and export settings to a .json file (up to 1 MiB, 1–120 character preset name). Per-session paint mask buffers are omitted so presets remain compact and portable.
  • Import Preset: Disabled while a render is running (Idle required). Importing validates types, ranges, stage-order rules, and filename suffixes before applying settings—showing a non-blocking banner confirmation or validation error.

Click Models in the header bar to open the model library dialog. Core NVIDIA RTX/DLSS features, ReShade, LUT grading, scaling, and format conversion do not require downloads; AI cleanup, AI upscaling, and ReShade AI depth install on demand into %LOCALAPPDATA%\Merserk\Visual Enhancer\Models (or VE_MODELS_DIRECTORY).

Models dialog over the Neural Rendering workspace listing Upscale, Denoising, Deblurring, ReShade depth, and NeuroFrame Runtime packages with Download and Remove buttons

Category Package Used By
Upscale VisionNext Image & Video Upscale stage
Upscale RazorNext Image & Video Upscale stage (high-detail reconstruction with VRAM budget control)
Upscale PeakMagnetar Image (1x, 2x, 4x, 6x, 8x) & Video (2x, 4x / Target Resolution) Upscale stage
Upscale OptiPulsar Fast Image (1x, 2x, 4x, 6x, 8x) & Video (2x, 4x / Target Resolution) Upscale stage
Denoising ClearMagnetar Denoising stage (with optional Noise Adaptation)
Deblurring FocusPulsar Deblurring stage
ReShade MoGe-2 (moge-2-vitb) ReShade AI Depth / Auto depth estimation (ONNX Runtime DirectML)
Runtime NeuroFrame Runtime Execution runtime (CUDA, XPU, or ROCm / HIP) required by PyTorch AI models (ClearMagnetar, FocusPulsar, VisionNext, RazorNext, PeakMagnetar, OptiPulsar)
  • Clicking Download on any PyTorch model automatically queues NeuroFrame Runtime first if the matching backend (CUDA, XPU, or ROCm / HIP) is not yet installed.
  • Downloads run in chunks with SHA-256 verification and automatic resume markers (.part / .resume.json), so interrupted transfers show a Resume button.
  • A live progress bar appears directly under the Models header tab while transfers are active. Remove prompts for confirmation and is disabled while a model is actively loaded by a render.
  1. Extract or copy your offline Models bundle (containing models-manifest.json, model folders, and optionally neuroframe-runtime) to %LOCALAPPDATA%\Merserk\Visual Enhancer\Models—or store it on any drive and set the VE_MODELS_DIRECTORY environment variable.
  2. Launch Visual Enhancer.exe or VE_CLI.exe. Startup automatically verifies and migrates compatible offline layouts.

Settings (Preferences & System Configuration)

Section titled “Settings (Preferences & System Configuration)”

Open Settings from the header bar to configure global preferences stored per user:

  • Interface Language: Switch dynamically between 12 UI localizations (English (US), العربية, Deutsch, Español, Français, हिन्दी, 日本語, 한국어, Polski, Українська, 中文 (简体), 中文 (繁體)).
  • AI Processing GPU (Tensor Cores / NGX): Select Auto or a specific NVIDIA GPU for DLSS Neural Rendering, DLSS Super Resolution, DLSS Ray Reconstruction, DLSS Frame Generation, RTX Video Super Resolution, RTX Video HDR, NVIDIA VFX, and ReShade CUDA interop.
  • AI Processing GPU (DirectX 12 / FidelityFX): Select Auto or a DXGI GPU adapter for AMD FSR 3 Frame Generation.
  • Decoding / Encoding Device: Choose Auto, CPU (Software decoding / encoding), or a Vulkan device adapter for FFmpeg decode/encode and GPU pixel filtering.
  • GPU: Select Auto or a specific DXGI adapter for AI upscalers.
  • AI Inferencing Framework: Choose Auto, CUDA, XPU, or ROCm / HIP for NeuroFrame Runtime execution.
  • Realtime Preview: Toggles automatic preview re-evaluation when you adjust sliders or scrub the video timeline. Disable this on slower machines and trigger previews manually with the Preview button.
  • Where to set Video Preview Length: In Video mode on Neural Rendering, use the duration selector (1 s, 3 s, 5 s, 10 s, 20 s, 30 s) directly beside Preview in the bottom inspector footer.
  • Where to configure ReShade search folders: Inside the ReShade pipeline card, set semicolon-separated (;) Shader Search Folder (.fx / .fxh) and Texture & LUT Search Folder paths.
  • Clean Cache: Displays current temporary preview and clip cache footprint on disk and reclaims all unreferenced cache files while keeping files currently loaded in viewports or queues untouched. Disabled while renders, previews, or LUT analyses are active.
  • Open Logs: Opens <app-folder>\logs\ in File Explorer for troubleshooting diagnostics.
  • Reset All Settings to Defaults: Restores every pipeline setting and clears active control masks after a confirmation prompt.