Skip to content

Ray Reconstruction & Motion

Visual Enhancer provides six stages dedicated to surface reconstruction, smooth high-frame-rate synthesis, and lightweight NVIDIA VFX video acceleration: DLSS Ray Reconstruction, DLSS Frame Generation, FSR Frame Generation, Video Super Resolution, Upscale (Lightweight), and Video Frame Generation.


DLSS Ray Reconstruction runs at native 1× resolution to clean noisy reflections, dappled shadows, global illumination shimmer, and fine specular textures without changing output dimensions.

  • Hardware: NVIDIA RTX GPU (compute capability >= 7.5), max 16,384 px per side.
  • Exclusivity: Mutually exclusive with DLSS Super Resolution.
  • Video HDR requirement: HDR videos processed without RTX Video HDR must carry smpte2084 (PQ) or arib-std-b67 (HLG) transfer signaling.
Control Mode Options Default Notes
Card Toggle Image & Video On / Off Off Image mode has no extra controls
Optical Flow Quality Video only High, Medium, Low High Balances temporal motion stability (High = finest grid) vs speed

Frame Generation Engines (DLSS, FSR 3, and NVIDIA VFX)

Section titled “Frame Generation Engines (DLSS, FSR 3, and NVIDIA VFX)”

All three frame-generation stages synthesize intermediate video frames to raise output cadence. Only one frame generator can be active per workflow.

Both videos play silently in a synchronized loop. Drag or use the arrow keys to compare. Press Space to play or pause, or Escape to pause.

DLSS Frame Generation (Frame-rate mode: Target FPS, Target Output Frame Rate: 120, Optical Flow Quality: High), synthesizing 5× cadence (24 → 120 FPS) across rapid highway panning and tire smoke.
  • Why this demo stayed clean at 120 FPS: Fast lateral camera movement past thin overhead streetlamps, striped highway barriers, glowing neon signage, and semi-transparent tire smoke easily causes tearing or pole halos with coarse optical flow. Using DLSS Frame Generation (--interpolator dlss --fps 120) with Optical Flow Quality on High solves motion vectors on the finest grid so edges and smoke trails stay cohesive across 5× synthesis (24 → 120 FPS).
  • Choosing Target FPS vs. Multiplier: Pick Target FPS (60 or 120) when targeting a fixed display refresh rate or normalizing mixed clips (23.976 / 24 / 30 FPS). Switch Frame-rate mode to Multiplier (2x–8x) when you want exact integer cadence multiples (24 → 48 / 96 FPS) with zero fractional timing adjustment.
Comparison DLSS Frame Generation FSR Frame Generation (FSR 3) Video Frame Generation (NVIDIA VFX)
GPU Support NVIDIA RTX (compute capability >= 7.5; RTX 40 / 50 + HAGS recommended) Any DirectX 12 GPU (NVIDIA, AMD, Intel) NVIDIA RTX 40 (Ada) or RTX 50 (Blackwell)
Resolution Limits Up to 16,384 px per side 64 × 64 px min, 16,384 px max per side Same as VFX pipeline (2 × 2 px up to codec/pipeline max)
HDR Support Supported (with HDR export pipeline) Supported with Preserve / Export 10-bit HDR + 10-bit codec Supported (10-bit / HDR), except when Upscale (Lightweight) is active
Cadence Handling Binary/integer multi-pass up to target (supports VFR & fractional targets) Binary/integer multi-pass up to target Constant Frame Rate (CFR) output
Control Range / Options Default Notes
Frame-rate mode Target FPS, Multiplier Target FPS Choose fixed output frame rate or exact multiplier
Target FPS 23.976, 25, 29.97, 30, 50, 59.94, 60, 90, 119.88, 120, 144, 165, 180, 240, 360, 480 60 Must exceed source/incoming frame rate
Multiplier 2x – 8x (2, 3, 4, 5, 6, 7, 8) 2x Multiplies incoming cadence by an integer factor
Optical Flow Quality High, Medium, Low High High uses finest motion vectors; Medium/Low increase throughput

Select your preferred DirectX 12 adapter under Settings > AI Processing GPU (DirectX 12 / FidelityFX).

Control Range / Options Default Notes
Frame-rate mode Target FPS, Multiplier Target FPS Identical target and multiplier presets to DLSS FG
Target FPS 23.976 – 480 60 Must exceed incoming frame rate
Multiplier 2x – 8x 2x Integer multiplier

The three lightweight NVIDIA VFX stages at the bottom of the Video pipeline share CUDA surfaces directly when placed consecutively in ascending order, avoiding CPU memory copies.

Fast NVIDIA VFX video restoration and upscaling with dedicated artifact modes.

Control Range / Options Default Notes
Mode family Standard Upscaling Modes, Denoise Modes, Deblur Modes, High-Bitrate Modes Standard Upscaling Modes Choose Denoise/Deblur for 1× cleanup, or Standard/High-Bitrate for upscaling
Quality Bicubic (Standard only), Low, Medium, High, Ultra High Higher levels improve edge reconstruction at modest GPU cost
Strength 0.00 – 1.00 (step 0.01) 1.00 Blends processed output with input/fallback strength
Sizing Mode Scale factor, Custom dimensions Scale factor Hidden when Denoise Modes or Deblur Modes is active (1× native)
Scale factor 1x, 1.5x, 2x, 3x, 4x 2x Available in Scale factor mode
Custom dimensions Width (px), Height (px), Lock Aspect Ratio (On/Off) 3840 × 2160, Locked Available in Custom dimensions mode

High-throughput spatial upscaler with built-in sharpening for 8-bit SDR video.

Control Range / Options Default Notes
Scale factor 4/3x, 1.5x, 2x, 3x, 4x 2x Incoming width and height multiplied by factor must produce exact integers
Sharpening strength 0.00 – 1.00 (step 0.01) 0.40 Controls post-upscale crispness

Hardware-accelerated VFX motion interpolation producing constant-frame-rate (CFR) video on RTX 40 (Ada) and RTX 50 (Blackwell) GPUs.

Control Range / Options Default Notes
Model Low, Medium, High Medium Selects complexity vs processing speed
Frame-rate mode Multiplier, Target FPS Multiplier Multiplier or custom FPS input
Multiplier 2x – 8x 2x Integer multiplier (Frame-rate mode = Multiplier)
Target FPS Custom numeric FPS (must be > incoming FPS and <= 1000) 60 Constant frame rate target (Frame-rate mode = Target FPS)
Automatic scene-cut detection On / Off switch On Resets optical flow across hard cuts to prevent cross-fade ghosting

  • Upscale vs. frame generation order: Upscaling first (4K input to frame generation) feeds finer edges to optical flow at higher VRAM cost, whereas interpolating at 1080p first before lightweight spatial upscaling reduces optical-flow workload on 4K exports.
  • Keep VFX stages adjacent: Because Visual Enhancer runs DLSS Frame Generation and FSR Frame Generation separately from NVIDIA VFX, use Video Frame Generation when pairing directly with Video Super Resolution or Upscale (Lightweight) to preserve zero-copy CUDA chaining.
  • Avoid double-filtering motion: Only one frame generator (DLSS, FSR, or Video Frame Generation) can be active per workflow.
  • Respect SDR-before-HDR ordering: Place RTX Video HDR after Upscale (Lightweight) and after all SDR-only upscalers so 10-bit PQ HDR frames are never fed into an SDR-only stage.
  • Enable scene-cut detection: Keep Automatic scene-cut detection enabled on edited footage with frequent camera cuts so transitions stay crisp instead of morphing between shots.