[brand-header]
# Does an AI upscaler improve AI-generated video?

### A measured bake-off of nine upscaling routes on synthetic 720p footage

**Short answer: no.** On clean AI-generated video, a plain Lanczos resize beat
every AI upscaler tested — including two paid Topaz models and a local diffusion
restorer that took 69 minutes to make a 60-second film *worse*.

---

## Summary

**Lanczos preserves the exact temporal behaviour of every source frame.** It
cannot introduce frame-to-frame model decisions, sharpening halos, invented
texture, or temporal "corrections." A generative upscaler can produce a sharper
single frame while making the *sequence* less believable — and the sequence is
what an audience watches.

**Photorealistic footage is a trap for restoration models.** A real camera
produces motion blur, defocus, shallow depth of field, lens softness, and detail
that falls away during movement. Nothing in a real shot is uniformly sharp. A
model trained to restore detail reads those properties as missing information
and repairs them — scoring better on a frozen frame while looking less like
something a camera captured. Realism depends more on *consistent imperfect
detail* than on maximum apparent detail.

Temporal processing does not solve this. It can make an invented detail
consistent across frames, but that only propagates a wrong interpretation
smoothly instead of noisily.

**Three different goals, three different answers:**

| Goal | Method |
|---|---|
| Present the video unchanged, at a larger size | **Lanczos** |
| Add modest, stable enhancement | Conservative CNN — Topaz Proteus or Gaia CG, Real-ESRGAN |
| Reconstruct detail that is genuinely absent | Diffusion — SeedVR2, FlashVSR, Topaz Starlight |

This footage sat firmly in the first category. Lanczos is not *failing to
restore* anything — it is correctly refusing to alter the evidence.

**What the measurements found.** Flicker, against the Lanczos floor: SeedVR2 7B
**3.33×**, SeedVR2 3B 2.79×, Real-ESRGAN 1.57×, Topaz Proteus 1.28×, Topaz Gaia
CG 1.10×. The models that added the most apparent detail were the least stable
in motion. Rendering at 4K and downsampling brought every model back to roughly
the noise floor — the one AI route worth considering, if you need one.

**How to judge this yourself.** Do not compare paused frames. Watch 5–10 seconds
of the same shot at normal speed, look away occasionally, and ask **"which
version looks most like a real camera captured this scene?"** — not "which frame
has more detail?" Those two questions have different answers, and only the first
one matters.

*Caveat worth keeping: this is one shot, photoreal and softly graded. On
stylised material — anime, motion graphics, illustration — invented detail reads
as style rather than as a lie, and the trade-off may well go the other way.*

What follows is what was tested, how it was measured, and where the measurement
itself falls down.

---

## The source

A 10-second cut from a short film: a man walking up a wooden pier at dusk.
Generated with Grok Imagine — **1280×720, 24 fps, 240 frames**, graded dark and
soft on purpose.

<a href="BAKEOFF-2026-09-03_images/pier_cut10_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/pier_cut10_1.jpg" alt="the source clip" width="640"></a>

*▶ The untouched 720p source. Click to play in a new tab.*

That "clean" matters more than anything else here. This footage has **no
compression damage, no sensor noise, and very little real high-frequency
detail**. It is the opposite of the case restoration models are built and
marketed for. A model trained to undo damage, handed material with no damage in
it, does not restore — it invents.

**The cut contains two shots.** Frames 0–158 are the pier; a **dissolve runs
frames 159–166** (6.62–6.92 s); frames 167–239 are an interior. ffmpeg's
`select='gt(scene,N)'` reports nothing here, because a dissolve raises the
frame-to-frame difference gently instead of spiking it — it only showed up when
the per-frame difference was plotted directly.

That is worth stating precisely, because it decides which numbers are safe:

- **The measurements below use frames 116–123 — comfortably inside the pier
  shot.** The static patch really is static there, so the flicker figures hold.
- **The eight single-frame comparisons further down are frame 211, which is in
  the interior shot**, not on the pier. They are a fair like-for-like between
  routes, but they are not showing the decking the numbers describe.

---

## The nine routes

| Route | Where it runs |
|---|---|
| Lanczos | ffmpeg, instant |
| Real-ESRGAN x4plus | local GPU |
| SeedVR2 3B → 1080p | local GPU (diffusion) |
| SeedVR2 7B → 1080p | local GPU (diffusion) |
| SeedVR2 → 4K → 1080p | local GPU (supersampled) |
| Topaz Proteus → 1080p | cloud API |
| Topaz Gaia CG → 1080p | cloud API |
| Topaz Proteus → 4K → 1080p | cloud API (supersampled) |
| Topaz Gaia CG → 4K → 1080p | cloud API (supersampled) |

Hardware: RTX 5090, 32 GB.

---

## How it was measured

Two numbers, from a **420×300 patch of decking with no walking figure in it**.
The shot is locked off, so in that patch:

- **Sharpness** — Laplacian variance. How much high-frequency energy is present.
- **Flicker** — mean absolute frame-to-frame change. In a static region of a
  locked shot, *anything that changes was invented by the upscaler.*

Lanczos cannot invent anything, so its numbers are the noise floor everything
else is judged against.

---

## Results

| Route | Sharpness ×Lanczos | **Flicker ×Lanczos** | Added detail | Time (240 frames) | Cost |
|---|---|---|---|---|---|
| **Lanczos** | 1.00× | **1.00×** (floor) | 0.00 | instant | free |
| SeedVR2 7B | 5.54× | **3.33×** | 9.31 | 13.6 min | free (local) |
| SeedVR2 3B | 2.84× | 2.79× | 8.08 | 11.5 min | free (local) |
| Real-ESRGAN | 4.33× | 1.57× | 6.13 | 2.9 min | free (local) |
| Topaz Proteus | 2.18× | 1.28× | 2.75 | 56 s | $0.10 |
| Topaz Gaia CG | 1.28× | 1.10× | 2.26 | 30 s | $0.10 |
| Topaz Proteus via 4K | 1.60× | **1.05×** | 3.22 | 88 s | $0.10 |
| Topaz Gaia CG via 4K | 1.42× | **1.05×** | 2.54 | 96 s | $0.10 |
| SeedVR2 via 4K | 1.57× | **1.04×** | 3.30 | 33.8 min | free (local) |

### Five things the numbers show

**1. Diffusion restorers were the worst option, by a wide margin.** SeedVR2 7B
added 5.5× the high-frequency energy of Lanczos and re-invented a third of it
every frame. Watched as video, that reads as heavy unsharp-mask crunch that
boils continuously.

**2. Bigger is not gentler.** The 7B model measured *worse* than the 3B on both
axes. Reaching for the larger model to reduce artifacts made them stronger.

**3. Deterministic beats generative for stability.** Real-ESRGAN adds nearly as
much sharpness as 7B (4.33×) but only 1.57× the flicker — because the same input
frame always produces the same output, so its invented detail at least sits
still. Diffusion models sample fresh each batch, so theirs boils.

**4. Conservative CNN models behaved far better than any diffusion model.** Gaia
CG landed at 1.10× flicker, Proteus at 1.28×.

**5. Supersampling fixes flicker.** Rendering at 4K and downsampling to 1080p
dropped flicker to the noise floor for *every* model — SeedVR2 from 2.79/3.33×
to 1.04× — and for the Topaz models it *raised* sharpness at the same time.
Proteus via 4K was the best-measured route: 1.60× the detail at 1.05× flicker.

**And none of it mattered.** Watched in motion at delivery size, the gap between
plain Lanczos and the best paid route was not worth the money, the round trip,
or the GPU hours.

---

## Look at it moving

Stills are the wrong tool for this (see below), so start here.

<a href="BAKEOFF-2026-09-03_images/cmp_motion_lanczos_vs_esrgan_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/cmp_motion_lanczos_vs_esrgan_1.jpg" alt="Lanczos vs Real-ESRGAN" width="640"></a>

*▶ **Lanczos (left) vs Real-ESRGAN (right)** — 1:1 crop, no scaling.*

<a href="BAKEOFF-2026-09-03_images/cmp_lanczos_proteus_gaiacg_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/cmp_lanczos_proteus_gaiacg_1.jpg" alt="Lanczos vs Proteus vs Gaia CG" width="640"></a>

*▶ **Lanczos | Topaz Proteus | Topaz Gaia CG** — the comparison that decided it.*

### What each upscaler invented

These show **only the difference** between the upscaled frame and the Lanczos
frame, amplified 10×. Black means "added nothing." This isolates the invention
from the picture — and makes the flicker impossible to miss.

<a href="BAKEOFF-2026-09-03_images/resid_esrgan_vs_7b_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/resid_esrgan_vs_7b_1.jpg" alt="Real-ESRGAN vs SeedVR2 7B residual" width="640"></a>

*▶ **Real-ESRGAN (left) vs SeedVR2 7B (right)**, invention only. The right side boils.*

<a href="BAKEOFF-2026-09-03_images/resid_proteus_vs_7b_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/resid_proteus_vs_7b_1.jpg" alt="Proteus vs SeedVR2 7B residual" width="640"></a>

*▶ **Topaz Proteus (left) vs SeedVR2 7B (right)**, invention only.*

---

## The same frame, eight ways

Frame 211 (00:08.8) from each route — this is the **interior shot** that follows
the dissolve, not the pier. Click any figure for the full-resolution original.

| | |
|---|---|
| [![source 720p](BAKEOFF-2026-09-03_images/4_source_720p_1.jpg)](4_source_720p.png)<br>**Source, 720p** | [![Lanczos](BAKEOFF-2026-09-03_images/1_lanczos_1080p_1.jpg)](1_lanczos_1080p.png)<br>**Lanczos → 1080p** |
| [![Gaia CG](BAKEOFF-2026-09-03_images/2_topaz_gaia_cg_1080p_1.jpg)](2_topaz_gaia_cg_1080p.png)<br>**Topaz Gaia CG → 1080p** | [![Proteus](BAKEOFF-2026-09-03_images/3_topaz_proteus_1080p_1.jpg)](3_topaz_proteus_1080p.png)<br>**Topaz Proteus → 1080p** |
| [![Proteus via 4K](BAKEOFF-2026-09-03_images/5_topaz_proteus_4k_to_1080_1.jpg)](5_topaz_proteus_4k_to_1080.png)<br>**Proteus 4K → 1080p** | [![Gaia CG via 4K](BAKEOFF-2026-09-03_images/6_topaz_gaia_cg_4k_to_1080_1.jpg)](6_topaz_gaia_cg_4k_to_1080.png)<br>**Gaia CG 4K → 1080p** |
| [![Proteus 4K native](BAKEOFF-2026-09-03_images/7_topaz_proteus_4K_native_1.jpg)](7_topaz_proteus_4K_native.png)<br>**Proteus, native 4K** | [![Gaia CG 4K native](BAKEOFF-2026-09-03_images/8_topaz_gaia_cg_4K_native_1.jpg)](8_topaz_gaia_cg_4K_native.png)<br>**Gaia CG, native 4K** |

### Crop grids at 1:1

[![four-way grid](BAKEOFF-2026-09-03_images/pier_grid4_lifted_1.jpg)](pier_grid4_lifted.png)
*Top-left **Lanczos**, top-right **Real-ESRGAN**, bottom-left **SeedVR2 3B**,
bottom-right **SeedVR2 7B**. Brightness lifted for inspection — this is not how
the shot looks. Note the hard horizontal striations the SeedVR2 panels put into
the decking.*
([true-levels version](BAKEOFF-2026-09-03_images/pier_grid4_1.png))

[![all five routes](BAKEOFF-2026-09-03_images/pier_all5_lifted_1.jpg)](pier_all5_lifted.png)
*Left to right: **Lanczos · Real-ESRGAN · SeedVR2 3B · SeedVR2 7B · SeedVR2 via 4K**.*
([true-levels version](BAKEOFF-2026-09-03_images/pier_all5_1.png))

---

## The methodological lesson

**Judge video in motion. Never from a still.**

Every quality judgement made from a frozen frame during this bake-off was wrong:

- The SeedVR2 4K-supersampled route looked **best of all** as a still. Watched as
  video, it was rejected on sight.
- SeedVR2 7B read as "marginally less aggressive" than 3B in a still. Measured,
  it was worse on both sharpening and flicker.

A still frame rewards whichever model invents the most detail, and is completely
blind to the one artifact that decides whether a shot is usable. **Flicker only
exists in motion.**

### Where this measurement falls down

The static-patch metric has a blind spot worth stating plainly, because it
produced a result that contradicts the human verdict.

The patch is a piece of decking with **no figure in it**. It catches invented
texture crawling on flat surfaces. It is blind to everything happening on the
subject — warping on the coat, edges swimming around the legs, features shifting.
On a shot whose subject is a man walking, that is a large omission.

Concretely: the table ranks SeedVR2-via-4K as the **most stable route of all**,
and a human watched that exact file and rejected it. The number is not wrong; it
is answering a narrower question than the eye is. **Background stability is not
picture quality.** A fuller comparison would measure a window that tracks the
subject as well.

---

## Practical notes for anyone running SeedVR2 locally

Two limits found by measurement, both contradicting the guidance in the docs.

**Batch size has a hard ceiling far below what the arithmetic suggests.**
SeedVR2's latents pack 4 frames, so batch size must follow 4n+1 (1, 5, 9, 13…),
and the documentation suggests matching batch size to shot length. On a 32 GB
card at 1080p, `--batch_size 21` *fits* — 31,715 MiB used of 32,607, 892 MiB
free — and then collapses. VAE decode went from ~5 seconds per batch at batch 5
to **13 minutes per batch** at 21. Five hours in, it had finished one chunk of
three. **Batch 5 was the stable operating point.**

**720p → 4K needs VAE tiling on 32 GB.** Without it, the run OOMs — in the VAE
decode, not the transformer, so block-swapping does not help. With tiling:
33.8 minutes for 240 frames.

---

## Practical notes on the Topaz API

Topaz's models are available per-second through third-party API resellers, with
**no subscription** — which is worth knowing before paying $299/year for the
desktop app. Measured cost: **$0.10 per 10 seconds of output at any resolution**.
1080p and 4K cost the same, despite published rate cards claiming resolution
tiers.


**Model selection lives in `preset`, by Topaz's internal codes:**

| Code | Model | | Code | Model |
|---|---|---|---|---|
| `prob-4` | Proteus | | `nyx-3` `nxf-1` `nxl-1` `nxhf-1` | Nyx family |
| `gcg-5` | Gaia CG | | `ahq-12` `amq-13` `alq-13` `alqs-2` `amqs-2` `aaa-9` | Artemis family |
| `ghq-5` | Gaia HQ | | `thd-3` `thf-4` `thm-2` | Theia |
| `rhea-1` | Rhea | | `dtv-4` `dtd-4` `dtds-2` `dtvs-2` `ddv-3` | Dione |
| `iris-2` `iris-3` | Iris | | `color-1` | Color |

No Starlight — that is Topaz's cloud-diffusion tier, kept behind their own
subscription.

**Total spend for this entire bake-off: $0.40.**

---

## Conclusion

**For AI-generated source, use a plain resize.** It is instant, free, and
invents nothing — so it cannot flicker, cannot crawl, and cannot rewrite a face.

Reach for an AI upscaler when the source is *genuinely degraded* real footage,
which is the job these models were built for. Then:

- Prefer a **conservative CNN** (Topaz Proteus or Gaia CG, or Real-ESRGAN) over a
  **diffusion restorer**, unless there is real damage to undo.
- If you use one, **render at 4K and downsample** — it dropped flicker to the
  noise floor for every model tested and cost nothing extra through an API.
- **Test on 5–10 seconds and watch it move** before committing a whole film.

The most expensive route tested produced the worst picture. The cheapest
produced the best. That is not a general law of upscaling — it is what happens
when you point a restoration tool at material with nothing to restore.

---

## All files

Click any thumbnail to play that clip in a new tab. All are the same 240 frames.

| | Clip | What it is | Output |
|---|---|---|---|
| <a href="BAKEOFF-2026-09-03_images/pier_cut10_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/pier_cut10_1.jpg" width="200"></a> | **Source** | The untouched Grok render everything else was made from | 1280×720 |
| <a href="BAKEOFF-2026-09-03_images/pier_lanczos_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/pier_lanczos_1.jpg" width="200"></a> | **Lanczos** | Plain resize. Invents nothing. **The winner.** | 1920×1080 |
| <a href="BAKEOFF-2026-09-03_images/pier_esrgan_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/pier_esrgan_1.jpg" width="200"></a> | **Real-ESRGAN** | 4× per frame, deterministic. 1.57× flicker | 1920×1080 |
| <a href="BAKEOFF-2026-09-03_images/pier_topaz_proteus_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/pier_topaz_proteus_1.jpg" width="200"></a> | **Topaz Proteus** | Cloud CNN. 1.28× flicker, 56 s, $0.10 | 1920×1080 |
| <a href="BAKEOFF-2026-09-03_images/pier_topaz_gaiacg_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/pier_topaz_gaiacg_1.jpg" width="200"></a> | **Topaz Gaia CG** | Cloud CNN for clean CG. 1.10× flicker, 30 s | 1920×1080 |
| <a href="BAKEOFF-2026-09-03_images/pier_topaz_proteus_4k_to_1080_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/pier_topaz_proteus_4k_to_1080_1.jpg" width="200"></a> | **Proteus via 4K** | Supersampled. Best measured: 1.60× detail, 1.05× flicker | 1920×1080 |
| <a href="BAKEOFF-2026-09-03_images/pier_topaz_gaiacg_4k_to_1080_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/pier_topaz_gaiacg_4k_to_1080_1.jpg" width="200"></a> | **Gaia CG via 4K** | Supersampled. 1.42× detail, 1.05× flicker | 1920×1080 |
| <a href="BAKEOFF-2026-09-03_images/pier_seedvr2_3b_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/pier_seedvr2_3b_1.jpg" width="200"></a> | **SeedVR2 3B** | Local diffusion. 2.79× flicker, 11.5 min | 1920×1080 |
| <a href="BAKEOFF-2026-09-03_images/pier_seedvr2_7b_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/pier_seedvr2_7b_1.jpg" width="200"></a> | **SeedVR2 7B** | Local diffusion. **Worst tested** — 3.33× flicker | 1920×1080 |
| <a href="BAKEOFF-2026-09-03_images/pier_seedvr2_4k_to_1080_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/pier_seedvr2_4k_to_1080_1.jpg" width="200"></a> | **SeedVR2 via 4K** | Supersampled. Measures best, rejected on sight | 1920×1080 |
| <a href="BAKEOFF-2026-09-03_images/pier_topaz_proteus_4k_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/pier_topaz_proteus_4k_1.jpg" width="200"></a> | **Proteus, native 4K** | The master the downsample came from | 3840×2160 |
| <a href="BAKEOFF-2026-09-03_images/pier_topaz_gaiacg_4k_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/pier_topaz_gaiacg_4k_1.jpg" width="200"></a> | **Gaia CG, native 4K** | The master the downsample came from | 3840×2160 |
| <a href="BAKEOFF-2026-09-03_images/pier_seedvr2_4k_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/pier_seedvr2_4k_1.jpg" width="200"></a> | **SeedVR2, native 4K** | 33.8 min, needed VAE tiling to fit at all | 3840×2160 |

### Side-by-side comparisons

| | Clip | What it shows |
|---|---|---|
| <a href="BAKEOFF-2026-09-03_images/cmp_motion_lanczos_vs_esrgan_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/cmp_motion_lanczos_vs_esrgan_1.jpg" width="200"></a> | **Lanczos vs Real-ESRGAN** | 1:1 crop, no scaling |
| <a href="BAKEOFF-2026-09-03_images/cmp_lanczos_proteus_gaiacg_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/cmp_lanczos_proteus_gaiacg_1.jpg" width="200"></a> | **Lanczos / Proteus / Gaia CG** | The comparison that decided it |
| <a href="BAKEOFF-2026-09-03_images/cmp_motion_three_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/cmp_motion_three_1.jpg" width="200"></a> | **Lanczos / ESRGAN / SeedVR2-4K** | Three-up |

### Residuals — what each upscaler invented, amplified 10×

Black means "added nothing." This isolates the invention from the picture.

| | Clip | What it shows |
|---|---|---|
| <a href="BAKEOFF-2026-09-03_images/resid_esrgan_vs_7b_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/resid_esrgan_vs_7b_1.jpg" width="200"></a> | **ESRGAN vs SeedVR2 7B** | Left sits still; right boils |
| <a href="BAKEOFF-2026-09-03_images/resid_proteus_vs_7b_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/resid_proteus_vs_7b_1.jpg" width="200"></a> | **Proteus vs SeedVR2 7B** | Same contrast, conservative CNN vs diffusion |
| <a href="BAKEOFF-2026-09-03_images/resid_esrgan_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/resid_esrgan_1.jpg" width="200"></a> | **Real-ESRGAN alone** | Invention only |
| <a href="BAKEOFF-2026-09-03_images/resid_proteus_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/resid_proteus_1.jpg" width="200"></a> | **Topaz Proteus alone** | Invention only |
| <a href="BAKEOFF-2026-09-03_images/resid_7b_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/resid_7b_1.jpg" width="200"></a> | **SeedVR2 7B alone** | Invention only |

### From an earlier test on a different shot

These three come from a first pass on an interior shot, before the pier was
chosen as the harder test. Kept for completeness; not part of the results above.

| | Clip | What it is |
|---|---|---|
| <a href="BAKEOFF-2026-09-03_images/kettle_cut10_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/kettle_cut10_1.jpg" width="200"></a> | **Source** | Interior cut, 1280×720 |
| <a href="BAKEOFF-2026-09-03_images/bake_lanczos_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/bake_lanczos_1.jpg" width="200"></a> | **Lanczos** | 1920×1080 |
| <a href="BAKEOFF-2026-09-03_images/bake_esrgan_1.mp4" target="_blank" rel="noopener"><img src="BAKEOFF-2026-09-03_images/bake_esrgan_1.jpg" width="200"></a> | **Real-ESRGAN** | 1920×1080 |

---


