Topaz Upscale Video Review (2026): Models, Speed, Quality & a Best Alternative

What Is Topaz Upscale and How Does It Work?

Here's the short answer: Topaz's video upscaler now ships as Topaz Video (formerly Topaz Video AI), a desktop app that reconstructs missing detail with model-per-content-type AI. If you're searching for it today, the two things you want to know are whether it still exists, and whether the new subscription is worth it over the alternatives — this review answers both.

Unlike conventional scalers that simply stretch pixels, Topaz Video analyzes motion vectors, edges, textures, and noise patterns frame-by-frame, then uses trained AI models to predict and reconstruct missing detail at the target resolution.

Each Topaz upscale model is trained for a specific footage type — Proteus for general live-action, Rhea for texture-heavy 4× upscales, Gaia for anime/CGI, Iris for faces, Starlight for severely degraded sources. That model-per-content-type approach is what separates Topaz from simple bicubic or Lanczos upscalers.

Platforms: Windows 10/11 (64-bit) and macOS 12+ (Intel + Apple Silicon M1/M2/M3/M4) 

Required GPU: NVIDIA RTX 30-series or newer with ≥ 8 GB VRAM recommended

What Happened to Topaz Video AI?

Topaz Labs discontinued Topaz Video AI in September 2025, with v7.1.5 as the final legacy build. It has been replaced by a product called simply Topaz Video, which is subscription-only. Existing installs of v7.1.5 still run, but new users get Topaz Video, and continued access to the latest models requires an active subscription.

Topaz Upscale Models Explained

Before testing performance, here's what each Topaz Video upscaler model does and when to use it. Topaz's 2026 lineup keeps the earlier Proteus / Rhea / Gaia / Iris / Starlight core and adds several refreshed models on top.

My default picks: reach for Proteus (or Astra 2, if you want the newest flagship) when in doubt; save Rhea for texture-critical stills-like shots; use Starlight only when the source is genuinely broken. Of the 2026 additions, Astra 2 is the one that matters most for upscaling — Nyx XL and Hyperion 2 are supporting engines, not upscalers.

Real-World Performance Tests of Topaz AI Upscale Models

To evaluate how each model performs in real scenarios, we tested Topaz AI video upscale using the following system:

Test system: 

GPU: NVIDIA GeForce RTX 4070
CPU: AMD Ryzen 7 5700G with Radeon Graphics

Proteus Model (General Live-Action)

The default “go-to” for most live-action content, excelling at balancing sharpness with natural textures.

• Source: 30-second urban landscape, 360 × 640, 30 fps
• Target: 4K
• Processing time: 1 min 47 s (~7.85 fps)

Result: Sharp, artifact-controlled output. Foliage and skyline edges stay clean. Good Edge definition without over-sharpening. Proteus is the model you should reach for when you don't know which one to pick.

Rhea Model (Texture Detail)

This model specializes in 4x upscaling and detail boosting, ideal for enhancing fine structure and surface fidelity.

• Source: 30-second cityscape, 720 × 1280, 30 fps
• Target: 4× upscale
• Processing time: 13 min 30 s (~1.10 fps)

Result: Excellent brickwork and surface reconstruction. Trade-off: Rhea is ~7× slower than Proteus — not a practical pick for long-form or batch jobs unless maximum detail matters.

Gaia Model (Anime & CGI)

This is Topaz anime model, designed for high-quality and animation inputs, ensuring crisp linework, vibrant color, and refined CGI output.

• Source: 1 min 59 s anime clip, 720 × 720, 30 fps
• Processing time: 9 min 37 s (~6.19 fps)

Result: Clean line work, stable color consistency. Weakness: slight softening around motion edges and occasional shadow flattening during fast action sequences.

Iris Model (Face Enhancement)

Focused on face enhancement, Iris intelligently restores facial details, particularly in low- to mid-quality or interlaced footage.

• Source: 25-second facial clip, 640 × 360, 25 fps
• Processing time: 1 min 13 s

Result: Clear eye detail, realistic skin texture, subtle facial feature recovery — without the "uncanny smoothness" artifact that can appear in aggressive face models.

Starlight Model (Diffusion Restoration)

Starlight stands out as Topaz's most ambitious model. In severely degraded or artifact-heavy clips, it produced excellent temporal consistency and noticeably smoother motion reconstruction.

However, processing times increased dramatically, and hardware utilization spiked significantly. Rendering was substantially slower than other models.

One practical tip from my own runs: on genuinely low-res sources (around 480p), Proteus at 2× produces cleaner output than pushing straight to 4× — the aggressive factor tends to hallucinate texture where the source has none. And be honest with expectations: results are strong on decent 1080p sources but degrade on very low-res or out-of-focus footage regardless of which model you pick.

How Long Does Topaz Take to Upscale?

The short answer: on an RTX 4070, Proteus runs at roughly 7–8 fps, Rhea drops to about 1 fps, and Starlight is slowest of all. In practical terms, one minute of 720p source takes around four minutes on Proteus, thirteen-plus minutes on Rhea, and can stretch to 30–45 minutes on Starlight.

Performance Summary

For more details on these models, you can watch this video:

How to Upscale a Video in Topaz (Step-by-Step)

If you’re wondering how to use Topaz Video AI, and you already own a license or are running the watermarked demo, here’s the shortest path to a completed upscale.

1. Install Topaz Video AI from topazlabs.com and launch.
2. Drop your source file into the main timeline (drag-and-drop or File → Open).
3. Pick a model based on content type — Proteus for general clips, Gaia for anime, Iris for faces, Rhea for texture, Starlight for severely damaged sources.
4. Set output resolution — 4K, 8K, or custom — in the right-hand inspector. Leave frame rate matching unless you also want slow motion.
5. Configure export options: container (MP4/MOV/MKV), codec (H.264 / HEVC / ProRes / DNxHR), bitrate, quality preset.
6. Click Export. Topaz queues the job; you can stack additional clips to run overnight.
7. Verify the output in a frame viewer at 100% zoom before celebrating — over-sharpening and color-shift artifacts are easiest to spot paused.

Topaz Upscaler Pricing & Trial (2026 Subscription)

In October 2025, Topaz Labs retired the perpetual license. All new users are now on annual subscriptions.

Pricing as of July 2026. One thing worth calling out: the annual plan locks you in for a full year — there is no mid-year cancel-and-refund, so factor that into the long-term math in the TCO section below.

Free Trial — Topaz offers a demo (no credit card required), but every exported clip is watermarked. You can preview output quality and test hardware speed, but you can't ship production work.

Important: the old "buy-it-once" perpetual license is no longer available for new users. Continued access to the latest AI models requires ongoing subscription payments.

Key Benefits of Topaz AI Video Upscale

• Industry-leading quality on heavy restoration. Rhea, Starlight, and Iris are genuinely best-in-class for their specific niches.
• Deep model library. Five dedicated upscaling models plus supporting frame interpolation and enhancement engines — more fine-grained control than any competitor.
• Granular parameter tuning. Power users can adjust sharpening, dehalo, anti-aliasing, and denoise independently.
• Batch queue. Load multiple clips with different models; Topaz processes overnight without supervision.
• Apple Silicon support. Topaz Video AI runs natively on M-series chips (M1 through M4).

Drawbacks of Topaz Video Upscale

• Annual subscription forever. After October 2025 there's no one-time buy; stop paying and you lose the app.
• Slow on complex models. Rhea drops to ~1 fps on RTX 4070; Starlight is even slower. Long-form restoration takes hours.
• Demanding hardware. Topaz recommends NVIDIA RTX 30-series+ with 8 GB VRAM minimum. Mid-range and integrated GPUs struggle.
• Steep learning curve. The upscaling models (plus additional enhancement / frame-interp / HDR models) are a lot to learn for a first-time user.
• Watermarked trial. You can't evaluate finished-quality output without paying.
• No built-in editor. Pure upscaling engine; you still need a separate NLE for trimming and finishing.
• Real-world quality is source-dependent. Users consistently report strong results on decent 1080p footage but uneven output on very low-res, out-of-focus, or heavily compressed sources — no upscaler, Topaz included, can invent detail that was never captured.

Upscaling Without a High-End GPU: Online and Cloud Options

Topaz Video is desktop-only and wants an NVIDIA RTX 30-series or newer card with at least 8 GB of VRAM. Many readers simply don't have that hardware — a MacBook Air, a mid-range laptop, or an older desktop won't clear the bar. Here are the genuine no-GPU paths worth knowing about.

• Topaz's own online upscaler. Topaz Labs offers a browser-based upscale tool that runs on their servers instead of your machine. Handy for a one-off clip; not built for hour-long timelines. Best for: short clips when you don't want to install anything. Not ideal for: batch or long-form work.
• Pay-per-use cloud APIs. Services like fal.ai (roughly $0.01 per up-to-720p clip) and Replicate host open-source upscaling models you can call from the browser. Cheap for occasional jobs, but you're stitching together a workflow yourself. Best for: developers or one-off experiments. Not ideal for: non-technical users who want a finished-app experience.
• UniFab FabCloud fallback. UniFab Video Upscaler AI is a desktop app, but when it detects your local GPU can't keep up, it can hand the job to FabCloud and finish in the cloud. You get the desktop UI and the cloud's horsepower on the same license. Best for: Mac users and mid-range PCs that want one predictable workflow. Not ideal for: fully offline environments.

Best Alternative to Topaz Upscaler: UniFab Video Upscaler AI

If you want a Topaz ai video upscaler alternative that delivers comparable quality at faster speed, lower long-term cost, and simpler UX, UniFab Video Upscaler AI is the most direct replacement in 2026.

UniFab Video Upscaler AI is part of UniFab All-In-One — a $319.99 lifetime-license suite containing 22 AI tools (19 paid modules plus 3 forever-free tools: Video Converter, Background Remover, Vocal Remover). It runs on both Windows and macOS (Intel + Apple Silicon) and falls back to FabCloud cloud processing when your GPU can't keep up.

Core Model Matchup: UniFab vs Topaz Upscale

Topaz Upscale Video vs UniFab Upscale Video — Head-to-Head

To move beyond the spec sheets, we conducted a series of hands-on, real-world comparisons—using the exact same footage, hardware, and settings—between Topaz and UniFab’s key models. This approach ensured that our findings reflect practical performance and user experience, not just theoretical benchmarks. What follows are objective observations on speed, output quality, and workflow efficiency for both platforms.

Test Environment & Methodology

To ensure a fair, repeatable assessment, we ran all upscaling tests on the same hardware—NVIDIA GeForce RTX 4070 GPU and AMD Ryzen 7 5700G CPU—with standardized sample clips. This approach allowed us to isolate the true differences between Topaz and UniFab, free from hardware or software version bias.

Proteus vs Equinox — Everyday 4K Upscaling

Source: 30-second urban landscape, 360 × 640, 30 fps → target 4K

Verdict: Equinox is ~70% faster with identical visual output. For batch work, the efficiency Edge is material.

Rhea vs Vellum — Texture & Detail Enhancement

Source: 30-second cityscape, 720 × 1280, 30 fps

Verdict: Vellum wraps a texture-heavy 4× job ~33% faster than Rhea. Quality is virtually indistinguishable to the naked eye. Trade-off: Vellum is fixed at 4×; Rhea lets you choose any scale factor.

Gaia vs Kairo — Animation / Cartoon Upscaling

Source: 1 min 59 s anime clip, 720 × 720, 30 fps

Verdict: Kairo is over 2× faster AND visibly sharper than Gaia on anime sources — the clearest win in this comparison.

Starlight vs Titanus — Cinematic Restoration

Topaz Starlight remains the gold standard for VHS / 360p-era restoration — diffusion AI produces temporal consistency nothing else matches today. But render times explode and output caps at HD. 

UniFab's Titanus (Ultra-Fast / Fast / HQ) delivers 4K output significantly faster on film & TV sources, and runs on lower-tier hardware, but doesn't match Starlight on the most degraded sources.

Head-to-Head Summary

• Animation / cartoon — Kairo beats Gaia on speed AND quality ✨
• Texture / detail — Vellum beats Rhea on speed; quality is a tie
• General live-action — Equinox beats Proteus on speed; quality is a tie
• Face restoration — Iris ≈ UniFab Face Enhancer AI (slight Topaz speed Edge)

Full Feature Comparison: UniFab All-In-One vs Topaz Video

UniFab Video Upscaler AI is bundled inside UniFab All-In-One — so one license covers far more than just upscaling.

Topaz Video vs UniFab — 5-Year TCO

After the October 2025 subscription switch, total cost of ownership matters more than sticker price. Here's what you pay over time:

UniFab breaks even at month 13. Every month after is pure savings.

My Verdict — Which Tool Should You Buy?

Use Topaz Video if you're a professional film restorer leaning heavily on Starlight for damaged sources, you need granular per-parameter tuning, and you already own an RTX 30-series or newer GPU. The annual subscription is defensible for working pros.

Use UniFab All-In-One if you upscale anime, restore general live-action, process batches, or work on a Mac / mid-range PC. UniFab matches Topaz on quality for most real-world clips, beats it outright on anime (Kairo > Gaia in our tests), runs faster, adds 15+ tools Topaz doesn't have at all, and costs $319.99 once instead of $299 every year.

Our full UniFab vs Topaz head-to-head goes deeper on the side-by-side benchmarks. 

🎁 Ready to Test a Lifetime-License Alternative?

FAQs about Topaz Upscale

Can Topaz upscale video?

Yes. Topaz Video AI is a dedicated topaz video upscaler designed specifically for AI-based video upscaling. Five models (Proteus, Rhea, Gaia, Iris, Starlight) handle different content types — live-action, anime, faces, degraded footage — and output up to 4K or 8K from SD/HD sources. Workflow: drop a file, pick a model, click Export.

Is Topaz upscaler free?

No. Topaz Video AI requires a $299/year subscription as of October 2025 (the one-time perpetual license was retired). A free demo is available, but every exported clip is watermarked, so you can't use the output for finished projects. If you need a free full trial, UniFab Video Upscaler AI offers a 30-day no-watermark trial.

Is Topaz Video AI discontinued?

Yes. The app formerly named Topaz Video AI was discontinued in September 2025 — v7.1.5 is the final legacy build. Topaz Labs replaced it with a product called Topaz Video, which is subscription-only. Existing v7.1.5 installs still run; new users get Topaz Video.

Can Topaz upscale images too, or only video?

Topaz Video handles video only. For still-image upscaling, Topaz Labs sells separate products — Gigapixel and Photo AI. This review and Topaz Video itself are strictly about video; image users should look at Gigapixel instead.

What are the system requirements for the Topaz video upscaler?

Windows 10/11 (64-bit) or macOS 12+ with Apple Silicon support. Topaz officially recommends an NVIDIA RTX 30-series or newer GPU with ≥ 8 GB VRAM, 16 GB RAM (32 GB recommended), and a modern CPU with AVX2 support. Lower-tier GPUs will crash or slow to a crawl on Rhea, Gaia, and Starlight.

Does Topaz Video AI work on Mac (Apple Silicon)?

Yes. Topaz Video AI is compatible with macOS 12+ and optimized for M1/M2/M3/M4 chips — often running cooler than on comparable Intel Macs. However, Apple Silicon performance still lags high-end NVIDIA GPUs on heavy 4K+ restoration. UniFab's FabCloud sidesteps the hardware limit entirely for Mac users.

Is Topaz Video safe to download?

Yes. Topaz Video is a legitimate desktop application from Topaz Labs with no malware or spyware, and downloads from the official Topaz site are safe. Avoid unauthorized third-party builds — they often carry trojans and can't receive model updates anyway.

How do I upscale a video without paying for Topaz?

The most direct free-with-real-output route is UniFab Video Upscaler AI's 30-day trial (no watermark, full features). Open-source pipelines like Video2X and DaVinci Resolve's Super Scale are also improving; our roundup of Topaz Video AI alternatives covers the full free-and-paid shortlist so you can match a tool to your source type.