Independent visual references · reviewed 2026-10-13

DLSS 5 before and after: inspect the conversion, not the promise

Every card below pairs an input with one output from our own image pipeline. The file dimensions, controls and provenance are shown beside the image; these are independent references, not NVIDIA screenshots or runtime benchmarks.

Try the free visual enhancer See Studio download details

Every card keeps the input, output, dimensions, record and limit visible; these are independent references, not NVIDIA screenshots.

Portrait detail recovery — 320 × 400 input
320 × 400 input · 320 × 400 px
Portrait detail recovery — 1200 × 1500 reference output
1200 × 1500 reference output · 1200 × 1500 px

portrait-id · Upscale 3.75×

Portrait detail recovery

The same portrait moves from 320 × 400 to 1200 × 1500 (3.75× on each edge). Compare eyes, jawline and hair before treating added texture as factual.

Record: Asset dimensions checked 2026-10-11; per-run model and parameter ledger not attached.

Limit: The asset ledger has dimensions but no per-run model/parameter record; this card is a visual reference, not a benchmark.

View prompt

Upscale 3.75×; preserve facial identity and expression; recover natural hair and skin texture; do not change age or face shape.

Open image upscaler →
Product edge and material detail — 320 × 213 input
320 × 213 input · 320 × 213 px
Product edge and material detail — 1200 × 800 reference output
1200 × 800 reference output · 1200 × 800 px

product · Upscale 3.75×

Product edge and material detail

The product pair uses the same 3.75× edge scale (320 × 213 to 1200 × 800). Check the silhouette, label placement and straight edges at 100%.

Record: Asset dimensions checked 2026-10-11; per-run model and parameter ledger not attached.

Limit: The asset ledger has dimensions but no per-run model/parameter record; inspect labels and logos instead of assuming they were recovered.

View prompt

Upscale 3.75×; preserve product silhouette, label placement and colour; recover material texture; do not invent text.

Open image upscaler →
Architecture lines and render texture — Low-resolution render
Low-resolution render · 320 × 213 px
Architecture lines and render texture — Reference output
Reference output · 1200 × 800 px

architecture-interior · Upscale 3.75×

Architecture lines and render texture

The architecture pair keeps the source and output ratio aligned. Inspect window spacing, facade lines and repeated geometry instead of judging only the thumbnail.

Record: Asset dimensions checked 2026-10-11; per-run model and parameter ledger not attached.

Limit: The asset ledger has dimensions but no per-run model/parameter record; straight lines remain a review point, not a guarantee.

View prompt

Upscale 3.75×; preserve straight architectural lines and window geometry; sharpen edges without halos; keep perspective and materials.

Open image upscaler →
Portrait retouch at the natural level — Portrait source
Portrait source · 600 × 900 px
Portrait retouch at the natural level — Retouched reference
Retouched reference · 600 × 900 px

portrait-id · Same-size retouch

Portrait retouch at the natural level

This is a same-size retouch, not an upscale: 600 × 900 in and 600 × 900 out. The measured run kept the face, hair, jacket and street recognisable.

Record: Measured case: natural level, 59 s wall clock, run recorded 2026-09-25 by the DLSS5NVIDIA team.

Limit: Retouching can still change uncertain pixels. Keep the source beside the result and review eyes, jawline, expression and skin grain.

View prompt

Level: natural. Preserve facial identity and expression; even skin tone and texture without changing face shape.

Open portrait retouch tool →
Background removal with pixel size kept — Photo input
Photo input · 1024 × 1024 px
Background removal with pixel size kept — Transparent cutout preview
Transparent cutout preview · 1024 × 1024 px

object-cutout · Background cutout

Background removal with pixel size kept

The cutout keeps 1024 × 1024 pixels. The real deliverable is a PNG with alpha; the shown result is flattened on a contrasting background.

Record: Measured case: fixed cutout, 2 s wall clock, run recorded 2026-09-25 by the DLSS5NVIDIA team.

Limit: Loose hair, glass and motion blur can retain some background; inspect the edge before publishing.

View prompt

Fixed cutout instruction; keep the teapot and remove the background to alpha transparency.

Open object eraser →
Object erased and the table rebuilt — Photo input
Photo input · 1024 × 1024 px
Object erased and the table rebuilt — Erased-object reference
Erased-object reference · 1024 × 1024 px

object-cutout · Generative erase

Object erased and the table rebuilt

A 1024 × 1024 erase run rebuilds the surface under the spoon while leaving the mug, napkin and light direction as review anchors.

Record: Measured case: fixed erase prompt, 60 s wall clock, run recorded 2026-09-25 by the DLSS5NVIDIA team.

Limit: The full frame is re-rendered, so fine text elsewhere can drift. Do not use it as pixel-exact restoration.

View prompt

Remove the spoon from the table and rebuild the wooden surface behind it; keep the mug, napkin and lighting.

Open object eraser →
A garment reference placed on a person — Person reference
Person reference · Person photo
A garment reference placed on a person — Try-on reference
Try-on reference · 768 × 1024 px output

two-reference · Try-on · two references

A garment reference placed on a person

The person image is the slider input; a second garment photo drives the jacket. Compare face, hands, hem, logos and folds before use.

Second reference: /examples/tryon-garment.jpg

Record: Measured case: top garment, 58 s wall clock, run recorded 2026-09-25 by the DLSS5NVIDIA team.

Limit: One garment per run. The result is a preview and can drift around hands, hems and logos.

View prompt

Garment: a top. References: person first, garment second. Keep face, pose and background stable.

Open visual enhancer →
Cyberpunk game frame 1 — 269 × 179 px input
269 × 179 px input · 269 × 179 px
Cyberpunk game frame 1 — 768 × 512 px reference output
768 × 512 px reference output · 768 × 512 px

game-frames · Game-frame style conversion

Cyberpunk game frame 1

A game-scene visual reference. Inspect silhouette, thin geometry, foliage and small text at 100%; it is not a game-runtime or frame-rate result.

Record: Asset dimensions checked 2026-10-13; per-run model and parameter ledger not attached; visual reference only.

Limit: The reference can alter texture, text and small geometry; it does not establish in-game DLSS behavior.

View prompt

Convert the supplied game frame while keeping the character silhouette, pose, lighting and scene layout available for review.

Open character-style tool →
Cyberpunk game frame 2 — 269 × 179 px input
269 × 179 px input · 269 × 179 px
Cyberpunk game frame 2 — 768 × 512 px reference output
768 × 512 px reference output · 768 × 512 px

game-frames · Game-frame style conversion

Cyberpunk game frame 2

A game-scene visual reference. Inspect silhouette, thin geometry, foliage and small text at 100%; it is not a game-runtime or frame-rate result.

Record: Asset dimensions checked 2026-10-13; per-run model and parameter ledger not attached; visual reference only.

Limit: The reference can alter texture, text and small geometry; it does not establish in-game DLSS behavior.

View prompt

Convert the supplied game frame while keeping the character silhouette, pose, lighting and scene layout available for review.

Open character-style tool →
Fantasy game frame — 240 × 200 px input
240 × 200 px input · 240 × 200 px
Fantasy game frame — 687 × 572 px reference output
687 × 572 px reference output · 687 × 572 px

game-frames · Game-frame style conversion

Fantasy game frame

A game-scene visual reference. Inspect silhouette, thin geometry, foliage and small text at 100%; it is not a game-runtime or frame-rate result.

Record: Asset dimensions checked 2026-10-13; per-run model and parameter ledger not attached; visual reference only.

Limit: The reference can alter texture, text and small geometry; it does not establish in-game DLSS behavior.

View prompt

Convert the supplied game frame while keeping the character silhouette, pose, lighting and scene layout available for review.

Open character-style tool →
Sci-fi game frame — 269 × 179 px input
269 × 179 px input · 269 × 179 px
Sci-fi game frame — 768 × 512 px reference output
768 × 512 px reference output · 768 × 512 px

game-frames · Game-frame style conversion

Sci-fi game frame

A game-scene visual reference. Inspect silhouette, thin geometry, foliage and small text at 100%; it is not a game-runtime or frame-rate result.

Record: Asset dimensions checked 2026-10-13; per-run model and parameter ledger not attached; visual reference only.

Limit: The reference can alter texture, text and small geometry; it does not establish in-game DLSS behavior.

View prompt

Convert the supplied game frame while keeping the character silhouette, pose, lighting and scene layout available for review.

Open character-style tool →
Modern game frame — 269 × 179 px input
269 × 179 px input · 269 × 179 px
Modern game frame — 768 × 512 px reference output
768 × 512 px reference output · 768 × 512 px

game-frames · Game-frame style conversion

Modern game frame

A game-scene visual reference. Inspect silhouette, thin geometry, foliage and small text at 100%; it is not a game-runtime or frame-rate result.

Record: Asset dimensions checked 2026-10-13; per-run model and parameter ledger not attached; visual reference only.

Limit: The reference can alter texture, text and small geometry; it does not establish in-game DLSS behavior.

View prompt

Convert the supplied game frame while keeping the character silhouette, pose, lighting and scene layout available for review.

Open character-style tool →
Anime game frame — 269 × 179 px input
269 × 179 px input · 269 × 179 px
Anime game frame — 768 × 512 px reference output
768 × 512 px reference output · 768 × 512 px

game-frames · Game-frame style conversion

Anime game frame

A game-scene visual reference. Inspect silhouette, thin geometry, foliage and small text at 100%; it is not a game-runtime or frame-rate result.

Record: Asset dimensions checked 2026-10-13; per-run model and parameter ledger not attached; visual reference only.

Limit: The reference can alter texture, text and small geometry; it does not establish in-game DLSS behavior.

View prompt

Convert the supplied game frame while keeping the character silhouette, pose, lighting and scene layout available for review.

Open character-style tool →
Product square detail 1 — 219 × 219 px input
219 × 219 px input · 219 × 219 px
Product square detail 1 — 627 × 627 px reference output
627 × 627 px reference output · 627 × 627 px

product · Upscale 2.86×

Product square detail 1

A square product reference for checking logos, labels, silhouette and material edges at 100%.

Record: Asset dimensions checked 2026-10-13; per-run model and parameter ledger not attached; visual reference only.

Limit: Generated lettering and fine logos require manual comparison with the source.

View prompt

Preserve the product silhouette, label placement and material edges while enlarging the supplied product image.

Open image upscaler →
Product square detail 2 — 219 × 219 px input
219 × 219 px input · 219 × 219 px
Product square detail 2 — 627 × 627 px reference output
627 × 627 px reference output · 627 × 627 px

product · Upscale 2.86×

Product square detail 2

A square product reference for checking logos, labels, silhouette and material edges at 100%.

Record: Asset dimensions checked 2026-10-13; per-run model and parameter ledger not attached; visual reference only.

Limit: Generated lettering and fine logos require manual comparison with the source.

View prompt

Preserve the product silhouette, label placement and material edges while enlarging the supplied product image.

Open image upscaler →
Badge bitmap to vector reference — 1024 × 1024 px bitmap
1024 × 1024 px bitmap · 1024 × 1024 px
Badge bitmap to vector reference — 1024 × 1024 px preview
1024 × 1024 px preview · 1024 × 1024 px

product · Bitmap to vector

Badge bitmap to vector reference

A same-size badge reference for checking edges, holes and emblem geometry after vector-oriented conversion.

Record: Asset dimensions checked 2026-10-13; per-run model and parameter ledger not attached; visual reference only.

Limit: Small lettering and ambiguous edges still need source comparison; this is not a guaranteed production SVG.

View prompt

Convert the supplied badge bitmap into a clean vector-style result while keeping the visible emblem geometry.

Open visual enhancer →
Architecture line reference 2 — 253 × 190 px input
253 × 190 px input · 253 × 190 px
Architecture line reference 2 — 724 × 543 px reference output
724 × 543 px reference output · 724 × 543 px

architecture-interior · Style conversion

Architecture line reference 2

An architectural reference for checking repeated windows, facade lines and perspective.

Record: Asset dimensions checked 2026-10-13; per-run model and parameter ledger not attached; visual reference only.

Limit: Straight lines and repeated geometry can drift; inspect the source and output together.

View prompt

Convert the architectural reference while retaining perspective, window spacing and facade lines for inspection.

Open visual enhancer →
Architecture line reference 3 — 253 × 190 px input
253 × 190 px input · 253 × 190 px
Architecture line reference 3 — 724 × 543 px reference output
724 × 543 px reference output · 724 × 543 px

architecture-interior · Style conversion

Architecture line reference 3

An architectural reference for checking repeated windows, facade lines and perspective.

Record: Asset dimensions checked 2026-10-13; per-run model and parameter ledger not attached; visual reference only.

Limit: Straight lines and repeated geometry can drift; inspect the source and output together.

View prompt

Convert the architectural reference while retaining perspective, window spacing and facade lines for inspection.

Open visual enhancer →
Interior style conversion reference — 900 × 900 reference input
900 × 900 reference input · 900 × 900 px
Interior style conversion reference — 1100 × 733 reference output
1100 × 733 reference output · 1100 × 733 px

architecture-interior · Style conversion

Interior style conversion reference

A room-style reference for checking materials, lighting and large structural lines.

Record: Asset dimensions checked 2026-10-13; per-run model and parameter ledger not attached; visual reference only.

Limit: A style conversion can change small objects and proportions; review the room layout before use.

View prompt

Use the supplied interior reference to guide materials, lighting and room composition.

Open visual enhancer →
Creative object redraw 1 — 269 × 179 px input
269 × 179 px input · 269 × 179 px
Creative object redraw 1 — 768 × 512 px reference output
768 × 512 px reference output · 768 × 512 px

object-cutout · Object redraw

Creative object redraw 1

A creative redraw reference for checking object edges, background continuity and lighting.

Record: Asset dimensions checked 2026-10-13; per-run model and parameter ledger not attached; visual reference only.

Limit: The surrounding scene may be regenerated; do not treat small text or exact geometry as preserved.

View prompt

Redraw the selected object and surrounding context while keeping the main visual subject available for review.

Open visual enhancer →
Creative object redraw 2 — 269 × 179 px input
269 × 179 px input · 269 × 179 px
Creative object redraw 2 — 768 × 512 px reference output
768 × 512 px reference output · 768 × 512 px

object-cutout · Object redraw

Creative object redraw 2

A creative redraw reference for checking object edges, background continuity and lighting.

Record: Asset dimensions checked 2026-10-13; per-run model and parameter ledger not attached; visual reference only.

Limit: The surrounding scene may be regenerated; do not treat small text or exact geometry as preserved.

View prompt

Redraw the selected object and surrounding context while keeping the main visual subject available for review.

Open visual enhancer →
Makeup guided by a second reference — 700 × 466 source
700 × 466 source · 700 × 466 px
Makeup guided by a second reference — 800 × 800 reference output
800 × 800 reference output · 800 × 800 px

two-reference · Makeup · two references

Makeup guided by a second reference

A two-reference makeup case; compare identity, eye area, skin texture and color placement.

Second reference: /examples/makeup-after.jpg

Record: Asset dimensions checked 2026-10-13; per-run model and parameter ledger not attached; visual reference only.

Limit: Makeup placement can drift with pose and lighting; it is a visual reference, not a cosmetic guarantee.

View prompt

Use the source portrait and a second makeup reference to guide color and placement while keeping identity reviewable.

Open visual enhancer →
Natural scene conversion 1 — 269 × 179 px input
269 × 179 px input · 269 × 179 px
Natural scene conversion 1 — 768 × 512 px reference output
768 × 512 px reference output · 768 × 512 px

nature · Look conversion

Natural scene conversion 1

A natural-scene reference for checking light direction, horizon and broad texture.

Record: Asset dimensions checked 2026-10-13; per-run model and parameter ledger not attached; visual reference only.

Limit: Fine foliage and distant detail can be invented or softened; compare against the input before sharing.

View prompt

Convert the natural scene while keeping the main light direction, horizon and large shapes available for comparison.

Open visual enhancer →
Natural scene conversion 2 — 269 × 179 px input
269 × 179 px input · 269 × 179 px
Natural scene conversion 2 — 768 × 512 px reference output
768 × 512 px reference output · 768 × 512 px

nature · Look conversion

Natural scene conversion 2

A natural-scene reference for checking light direction, horizon and broad texture.

Record: Asset dimensions checked 2026-10-13; per-run model and parameter ledger not attached; visual reference only.

Limit: Fine foliage and distant detail can be invented or softened; compare against the input before sharing.

View prompt

Convert the natural scene while keeping the main light direction, horizon and large shapes available for comparison.

Open visual enhancer →

Case index

Browse every case by category, dimensions and record status.

CategoryCaseInputOutputWorkflowRecord
portrait-idPortrait detail recovery320 × 400 px1200 × 1500 pxUpscale 3.75×checked 2026-10-11 · no run ledger
productProduct edge and material detail320 × 213 px1200 × 800 pxUpscale 3.75×checked 2026-10-11 · no run ledger
architecture-interiorArchitecture lines and render texture320 × 213 px1200 × 800 pxUpscale 3.75×checked 2026-10-11 · no run ledger
portrait-idPortrait retouch at the natural level600 × 900 px600 × 900 pxSame-size retouchMeasured: natural level, 59 s wall clock, run recorded 2026-09-25 by the DLSS5NVIDIA team.
object-cutoutBackground removal with pixel size kept1024 × 1024 px1024 × 1024 pxBackground cutoutMeasured: fixed cutout, 2 s wall clock, run recorded 2026-09-25 by the DLSS5NVIDIA team.
object-cutoutObject erased and the table rebuilt1024 × 1024 px1024 × 1024 pxGenerative eraseMeasured: fixed erase prompt, 60 s wall clock, run recorded 2026-09-25 by the DLSS5NVIDIA team.
two-referenceA garment reference placed on a personPerson photo768 × 1024 px outputTry-on · two referencesMeasured: top garment, 58 s wall clock, run recorded 2026-09-25 by the DLSS5NVIDIA team.
game-framesCyberpunk game frame 1269 × 179 px768 × 512 pxGame-frame style conversionchecked 2026-10-13 · no run ledger
game-framesCyberpunk game frame 2269 × 179 px768 × 512 pxGame-frame style conversionchecked 2026-10-13 · no run ledger
game-framesFantasy game frame240 × 200 px687 × 572 pxGame-frame style conversionchecked 2026-10-13 · no run ledger
game-framesSci-fi game frame269 × 179 px768 × 512 pxGame-frame style conversionchecked 2026-10-13 · no run ledger
game-framesModern game frame269 × 179 px768 × 512 pxGame-frame style conversionchecked 2026-10-13 · no run ledger
game-framesAnime game frame269 × 179 px768 × 512 pxGame-frame style conversionchecked 2026-10-13 · no run ledger
productProduct square detail 1219 × 219 px627 × 627 pxUpscale 2.86×checked 2026-10-13 · no run ledger
productProduct square detail 2219 × 219 px627 × 627 pxUpscale 2.86×checked 2026-10-13 · no run ledger
productBadge bitmap to vector reference1024 × 1024 px1024 × 1024 pxBitmap to vectorchecked 2026-10-13 · no run ledger
architecture-interiorArchitecture line reference 2253 × 190 px724 × 543 pxStyle conversionchecked 2026-10-13 · no run ledger
architecture-interiorArchitecture line reference 3253 × 190 px724 × 543 pxStyle conversionchecked 2026-10-13 · no run ledger
architecture-interiorInterior style conversion reference900 × 900 px1100 × 733 pxStyle conversionchecked 2026-10-13 · no run ledger
object-cutoutCreative object redraw 1269 × 179 px768 × 512 pxObject redrawchecked 2026-10-13 · no run ledger
object-cutoutCreative object redraw 2269 × 179 px768 × 512 pxObject redrawchecked 2026-10-13 · no run ledger
two-referenceMakeup guided by a second reference700 × 466 px800 × 800 pxMakeup · two referenceschecked 2026-10-13 · no run ledger
natureNatural scene conversion 1269 × 179 px768 × 512 pxLook conversionchecked 2026-10-13 · no run ledger
natureNatural scene conversion 2269 × 179 px768 × 512 pxLook conversionchecked 2026-10-13 · no run ledger

Method and reproduction

Inputs come from uploads to our own image pipeline. For a reproduction, record the original input, output size, model or preset, hardware, driver, date and elapsed time; these are not NVIDIA runtime screenshots, game frames or cross-GPU comparisons.

Open the benchmark evidence center → · See 20 character-style cases →

Six-second video references

The clips loop and carry a text conclusion so the page remains understandable without playback.

Low-resolution input to an upscale reference

Text conclusion: this original reference shows a source-to-output transition; it is not a model speed or quality benchmark.

Game character style conversion reference

Text conclusion: identity, pose and silhouette are the review anchors; this is an independent style reference, not NVIDIA runtime footage.

Our only published video-pipeline measurement: a three-second 720p clip (36 frames) took 87.4 seconds of batch time on an RTX 4090 reference machine with driver 591.86 — 2.43 seconds per frame. That is local batch time, not real-time performance, and not the render time of the two clips above, which are transition references only.

Measurement method and limits: open the benchmark evidence center →

Honest limits

Very small sources, heavy compression, tiny text, multiple faces and motion blur can make results unreliable. Enlargement is an estimate, not factual recovery; inspect at 100% and keep the original before publishing.

Frequently asked questions

Are these NVIDIA DLSS 5 runtime screenshots?

No. These are independent before-and-after references from the DLSS5NVIDIA image pipeline. They are not NVIDIA captures and do not prove an in-game DLSS runtime result.

What does “before and after” mean here?

Each card keeps the input beside one output from the named workflow. The three low/high pairs show verified file dimensions; the retouch, cutout, erase and try-on cards show measured workflow runs with their controls and date.

Does upscaling recover factual detail?

No. It estimates a more usable visual result from the input. Faces, labels, numbers, thin geometry and repeated patterns can change, so compare them at 100% and keep the original.

Can I try the converter without an RTX GPU?

Yes. The browser workflow runs on the site service, so a modern browser is enough for a cached example or an account workflow. The separate DLSS5 Studio request page is a Windows RTX product and is still being developed.

Why is the retouch card not labelled as an upscale?

Its input and output are both 600 × 900. It is a same-size portrait retouch case, included to show the difference between restoring an existing image and enlarging it.

What inputs usually fail?

Extremely small or heavily compressed images, tiny text, multiple faces and scenes with motion blur are hard cases. Start with the largest original file, then inspect the result before sharing or publishing it.

How is this page different from the benchmark center?

This page is a visual gallery: it pairs existing inputs and outputs so you can inspect edges, faces and composition. The benchmark center defines measurement fields and separates recorded runs from visual references.

Can I reproduce one of these cases?

Yes. Open the linked tool on a comparable input, use the control shown on the card, and record the original file, output size, model or preset, hardware, driver, date and elapsed time. Results may vary between runs.

Why does this page not use NVIDIA screenshots?

The gallery documents our own image pipeline. Keeping independent assets separate from NVIDIA material makes the provenance clear and avoids presenting a visual reference as an official runtime capture.

Continue with a real workflow

Image upscaler · Visual enhancer · Game character cases · Studio download details

Review what changed in a neural redraw before sharing it

Inspect game HUD and text before sharing a redraw