Video summary

The $7 Performance Boost: Lossless Scaling & Lossless Frame Generation Image Quality

Main summary

Key takeaways

Technology

Overview: Lossless Scaling as an agnostic “frame-gen + upscaling” tool

  • Lossless Scaling (Steam, ~$7) is presented as a third-party utility (with a “cult following”) that can perform:
    • Frame generation up to very high multipliers (up to ~20x)
    • Multiple upscaling methods
  • It aims to be vendor-agnostic: unlike DLSS/FSR/XESS, it doesn’t require engine/game developer integration.
  • The video stresses it’s not a replacement for first-party, tightly integrated upscaling/frame-gen solutions when those exist.
  • Main potential use cases:
    • Helping older GPUs
    • Situations where DLSS/FSR/XESS aren’t available
    • Enthusiast tweaking and niche scenarios

What the video tests / does (methodology + scope)

  • The review is image-quality-focused, not a fully comprehensive feature audit.
  • It emphasizes common usage paths and mentions prior testing for DLSS and FSR, with plans to cover XESS later.
  • Testing method
    • They intentionally slow playback and analyze frame-by-frame to expose artifacts.
    • They also watch at normal speed (with a zoomed-out scale view) to estimate “game feel.”
  • They explicitly did not test emulation use cases (policy/allowed-content constraints).

Frame Generation (LSFG): key concepts and modes

Core limitation

  • Frame generation can smooth motion, but it can’t truly convert 30 fps into true 60 fps.
  • Expected failure modes include:
    • distortions
    • duplication/splitting of moving objects
    • added latency

Two LSFG multiplier modes

  • Fixed mode
    • Uses a static frame ratio (multiplier 1–20).
    • Example: multiplier 4 inserts more interpolated frames.
  • Adaptive mode
    • Dynamically adjusts fractional multipliers to maintain a target FPS.
    • Prioritizes smoothness over quality/latency.

Adaptive mode use case highlighted

  • Particularly relevant for games with hard FPS caps that don’t align with monitor refresh rates (example: Dark Souls II capped at 60 fps).
  • The video discusses potential “microstutter” / “simulation vs visual time” oddities, referencing a separate benchmarking methodology video.

Upscaling (LS1): key concepts and limitations

How LS1 differs from DLSS/FSR-style methods

  • The video selects LS1 as the primary upscaling method for detailed testing.
  • LS1 is spatial-only
    • It doesn’t use game engine motion/extra data the way DLSS/FSR do.
    • It runs on finished frames captured after the game renders them.

Settings / controls

  • Supports scaling factors 1–10 and includes an auto mode.
  • Includes a sharpness slider (0–4, default ~1).

Operational caveats

  • The target app generally must run in windowed or borderless mode.
  • Borderless may fail in some games (example mentioned: Baldur’s Gate 3).
  • Some fullscreen behaviors can be inconsistent depending on how the game implements fullscreen exclusivity.

How the software works internally (capturing + processing pipeline)

  • Upscaling and frame gen are performed after the original frames render by capturing frames via DXGI API.

Frame generation flow (high level)

  1. Capture finished game frames
  2. Downscale them
  3. Estimate motion between downscaled frames
  4. Render a full-res warped frame from the prior real frame
  5. Output generated frames sequentially (real frame → fake → next real)

Flow scale slider

  • Controls how aggressively downscaled frames are used for motion estimation.
  • Increasing it can:
    • add warp steps
    • increase GPU overhead
    • affect latency and reduce base performance

Dual GPU mode

  • Can move processing work to a secondary GPU to reduce burden on the primary GPU.
  • Not standard SLI/AFR; it’s described as task offloading.

Testing setup

Frame generation testing benchmark

  • CPU: 9800X3D
  • GPU: RTX 5090
  • Capture examples at 4K / 120 fps, using 2x frame gen
  • They cap/adjust FPS at higher multipliers to stay within capture limitations.
  • Example stated: for 4x, they allow ~3.33 ms extra gap between real frames that LSFG must bridge.

Results: LSFG frame generation (artifacts + per-game findings)

Overall pattern

  • At higher multipliers (and/or under slowed frame-by-frame inspection), LSFG often shows:
    • object splitting
    • warping
    • flicker
    • double images
  • This is especially common when motion is complex, such as an object crossing or occluding another.

Final Fantasy XVI (scripted Dragoon scene)

  • 2x
    • blur/strangeness; lance shape breaks down midair
  • 25% slow-motion
    • breakdown when the lance passes over armor
  • Frame-by-frame
    • lance splits into two (“quantum superposition” look), with severe motion errors
  • UI stays relatively stable, but subtitle morphing isn’t perfect
  • 4x
    • additional warping; starry sky flickers/churns
  • Conclusion: impressive smoothness in-play, but fails strongly under detailed inspection.

Cyberpunk 2077 (Corpo Plaza)

  • 2x
    • may look “not bad” at full speed; slowed reveals issues
  • Flickering columns and warped high-frequency geometry (highway divider grids)
  • Frame-by-frame
    • arrows and passing cars show broken/wobbly geometry
  • 4x
    • “totally passable” in motion, but slowdown shows big warps
  • 10x
    • severe failures (phasing poles, “impressionist paintings,” wreck chaos)
  • Conclusion: artifact severity increases dramatically with multiplier.

Marvel Rivals (built-in benchmark)

  • 2x at normal speed: generally okay initially
  • Frame-by-frame: double images and morphing (staff, horns, background geometry, graffiti/masks)
  • 4x
    • many double images (tentacles)
    • UI elements (names/health bars) can warp due to limited UI/world separation
  • Conclusion: without engine-level UI separation, LSFG can’t consistently distinguish UI vs 3D world.

Another Marvel Rivals training scenario (wall run)

  • Repeating geometric patterns (triangular ridge/rails) become jumbled/morph under slowdown
  • 20x described as mostly “funny/amusing,” with widespread “acid trip” instability

Space Marine 2 (decapitation mission)

  • Considered one of LSFG’s better cases
  • 2x
    • fewer obvious anomalies at play speed
  • Frame-by-frame still finds issues like brief warping at armor edges
  • Generated frames are blurrier overall
  • Conclusion: works better when motion/patterns are easier to interpolate

Results: LS1 upscaling image quality (artifact types + scene comparisons)

Important framing

  • The video says LS1 isn’t a direct competitor to DLSS/FSR at current-gen quality levels because it lacks game engine information.
  • Comparisons include:
    1. Native 1080p vs LS1 upscaled to 4K
    2. Native 4K vs LS1
  • They also use a “minimum acceptability” check: whether LS1 makes 1080p → 4K look better than native 1080p.

Marvel Rivals

  • 1080p → LS1 4K
    • Pros: clearer some wood textures; some UI elements sharper
    • Cons: shadow floor becomes grainy/undesirable; oversharpening
  • Versus native 4K
    • LS1 looks “crunchy” and oversharpened
    • fine edges/details can look more aliased than true 4K
  • Overall: more like blur removal + sharpening than reconstruction.

Baldur’s Gate 3

  • Borderless mode issues prevented consistent scaling; they used windowed mode
  • Improved over native 1080p (sharper, better clarity)
  • Downsides:
    • aliased “crunch” on fine character edges
    • some details can’t be restored to true 4K fidelity (e.g., tail ridges, shoulder zigzag detail)
  • Conclusion: partial improvements, but cannot match native high-res detail.

Space Marine 2

  • Considered a stronger LS1 showcase
  • 1080p → LS1 4K
    • Pros: better definition on statue/armor base; more clarity mid-to-far
    • Cons: distant/fine details limited by missing detail from 1080p source
  • Versus native 4K
    • native still wins for very fine geometry and small details
  • Conclusion: can be “good enough” in some scenes, but not equivalent to native 4K.

Overall conclusions (what the video recommends / doesn’t)

Frame generation (LSFG)

  • Useful for smoothness if you accept:
    • clarity loss
    • artifact risk
    • added latency
  • Not “magic”: generated frames can distort/duplicate moving objects.
  • Reviewer’s taste summary:
    • 2x frame gen isn’t especially useful (“not useful” stated for their taste)
    • adaptive mode is described as genuinely impressive and unique

Upscaling (LS1)

  • Mixed results:
    • sometimes mild clarity improvement (e.g., Space Marine 2)
    • sometimes oversharpening and worse fine-edge artifacts (e.g., Baldur’s Gate 3)
  • Because it’s spatial-only, it’s generally limited compared to DLSS/FSR/XESS when those are available.

Where Lossless Scaling fits best

  • Valuable as a third-party agnostic solution when vendor methods aren’t available.
  • Potential niche value:
    • older hardware
    • flexible use beyond only gaming
    • pixel art and community-guided modes
    • emulation is commonly discussed (but not tested in this video)

Main speakers / sources

  • The video appears to be from Gamers Nexus (GN), reviewing/testing Lossless Scaling.
  • Specific speaker identity isn’t clearly stated in the subtitles; it’s presented as a GN-style multi-test bench review with the main reviewers speaking through the script.

Original video