What are the best settings for Topaz Video AI to upscale SD and 1080p footage to 4K in 2026?

What "Best" Actually Means in Topaz Video AI

There is no single "best" configuration in Topaz Video AI because the right settings depend on the source material, your GPU, and the final delivery format. The software, originally released as Video Enhance AI in 2020 and rebranded to Video AI, ships with multiple neural models (Artemis, Gaia, Proteus, Nyx, Rhea, Apollo) that each target a different problem. According to Topaz Labs' own documentation and reviews such as the Fstoppers Image Enhancer comparison, users who simply pick the highest-quality preset at 4× scale tend to get worse results than users who match the model to the artifact they are trying to fix.

Also worth reading: How Do AI Footage Restoration Tools Upscale Videos to 4K? · How Do Professionals Restore and Upscale Archival Footage to 4K Without Losing Historical Integrity? · What Are the Best K60 Video Export Settings for Faster, Cleaner 4K Exports?

For most users upscaling Standard Definition (480i/p) or 1080p footage to 4K (3840×2160) in 2026, the practical recommendation is: choose Proteus v4 for clean modern sources, Artemis v13 for noisy or low-light footage, Gaia for computer-generated or animated content, and Nyx for footage that already has acceptable detail but needs motion-aware sharpening at higher frame rates. A "one-click" preset can serve as a starting point, but the model, the enhancement strength, dehalo radius, and noise reduction parameters all interact in ways that a single slider cannot capture.

The remainder of this guide walks through the realistic settings that consistently produce usable 4K output without hours of trial-and-error, and it highlights where the defaults are misleading.

Matching the Model to the Source Material

The single most important decision in Topaz Video AI is which model to select, and this decision should be made before you touch any sliders. For live-action footage shot on a modern camera, Proteus offers the best balance between detail recovery and stability. Proteus has three internal "quality" modes — Quality, Balanced, and Speed — and for 4K upscaling the Quality preset is almost always worth the extra render time because it uses a higher-resolution intermediate pass.

For older, noisy, or grain-heavy footage, Artemis outperforms Proteus. In a real-world test covered by Red Shark News converting SD footage to 4K, Artemis produced visibly cleaner results on VHS-style sources because it was trained on noisy pairs and can hallucinate detail without amplifying grain. The downside is that Artemis can introduce a "synthetic" look on already-clean footage, which is why matching the model to the source is non-negotiable. Gaia remains the standard for animation, screen recordings, and anime, because it was trained on CGI-style data and tends to preserve flat, vector-like edges without over-sharpening.

Nyx is a newer addition designed primarily for low-light and nighttime scenes. If you are upscaling surveillance footage, concert recordings, or any material shot at ISO 3200 and above, Nyx typically beats both Proteus and Artemis at preserving chroma fidelity while suppressing luminance noise.

Recommended Settings for SD to 4K Upscaling

For Standard Definition (640×480 NTSC or 720×576 PAL) upscaling to 3840×2160, the most reliable workflow in 2026 starts with the Artemis v13 model at the Quality preset. Set the output resolution to 4K, keep the frame rate at "Auto" unless the source is clearly 23.976 or 25 fps, and use a single-pass output rather than two-pass. Within the enhancement panel, set Enhancement Strength between 60 and 80 percent. Anything above 85 percent tends to introduce edge halos on faces and text. Set Noise Reduction between 20 and 40 for VHS-grade sources and between 10 and 20 for cleaner DVD captures.

Dehalo is often overlooked but critical for SD material. A value of 50 to 70 helps remove the ringing that NTSC composite encoding leaves around hard edges. Anti-aliasing should remain off for SD sources because it tends to soften the very edges you are trying to recover. Deblur can be left at 0 unless the source shows obvious motion blur, in which case 25 to 50 helps.

For DVD and broadcast footage specifically, several reviewers including Red Shark News observed that combining Artemis with a separate grain-management pass through Neat Video (or Topaz's own Photo AI on an extracted frame sequence) produces cleaner 4K output than asking the Video AI model to do everything in one pass. This is a workflow consideration, not just a setting.

Recommended Settings for 1080p to 4K Upscaling

For 1080p (1920×1080) footage, the target ratio is a 2× upscale, which is gentler than the 4× or 6× jump required for SD sources. Here, Proteus v4 typically produces the most natural-looking 4K result. Use the Quality preset, set Enhancement Strength to 50 to 70, and leave Noise Reduction between 0 and 10 unless the source has visible grain. Because 2× is well within the comfort zone of modern AI models, pushing Enhancement Strength above 80 almost always introduces artifacts that look like fine shimmering on textured surfaces.

For modern digital cameras with clean sensor output — Sony A7S III, Canon R5, RED Komodo — many users in 2025 and 2026 have reported that Proteus with all sliders at 50 produces output that is indistinguishable from a native 4K render. For 1080p footage with mild compression artifacts from streaming or webcam capture, switching to a dedicated artifact-removal model such as Nyx with Detail Recovery set to High can recover more usable texture.

The frame rate is another lever. If you have 1080p60 footage and want a 4K60 output, Nyx v2 supports frame rate conversion natively and tends to do a better job than the older Gaia-based interpolation. If you only need 4K30, leaving the frame rate alone is usually the safest choice.

Hardware, Encoding, and Output Settings

Topaz Video AI is one of the most GPU-hungry consumer applications available. The 2025 and 2026 versions require an NVIDIA GPU with at least 8 GB of VRAM for 4K output, and the company recommends 16 GB for workflows involving 6K or 8K sources. Apple Silicon support has improved dramatically since the M1 generation, and Scott Simmons' ProVideo Coalition review of the M4 Mac mini confirmed that an M4 Pro chip with 24 GB of unified memory can process 4K60 material in real time or better using ProRes hardware acceleration.

For the output container and codec, the practical answer in 2026 is ProRes 422 HQ for archival work and H.265 at a constant quality of around RF 18 to 20 for delivery. H.264 is acceptable for web delivery but wastes storage space compared to H.265 at equivalent quality. Avoid the in-app "YouTube" preset unless you specifically understand what it does — it re-encodes twice and tends to over-compress mid-tones.

If you are processing long-form material (over 30 minutes), break the job into chapters of 5 to 10 minutes. This allows you to recover from a crash and makes memory pressure manageable on 8 GB cards.

Comparison of Topaz Models

The following table summarizes the practical differences between the four models most users choose between in 2026. Settings are general recommendations for 4K output; specific strengths vary with source material.

The numbers in the speed row are not official benchmarks but approximate figures gathered from user reports in 2025–2026 testing.

Common Mistakes and How to Avoid Them

The most frequent mistake is using the "highest quality" preset on every clip. The presets in Topaz Video AI are starting points, not optimal values, and the Quality preset on Artemis with a clean 1080p source often produces a waxy, over-processed look that betrays the AI immediately. A second common error is leaving Noise Reduction at the default when upscaling clean footage — the default of around 25 tends to eat fine texture and skin pores, which is fine for VHS but wrong for modern DSLR output.

Another mistake is using the same model for an entire timeline. If you have a project that mixes talking-head interviews, B-roll, and archival footage, you should batch your clips by content type and apply a different model and slider set to each batch. Topaz Video AI does not currently offer automatic per-shot model selection, so this is a manual step, but it has a measurable impact on quality.

A fourth mistake is ignoring color space. If your source is in Rec. 709 and you output tagged as Rec. 2020, you will see slightly desaturated and washed-out results in QuickTime and most NLE timelines. Match the output color space to the source, and only convert to Rec. 2020 if your delivery target specifically requires it (most HDR mastering pipelines). Finally, do not rely solely on the preview window at quarter resolution — preview at full resolution before committing to a render that may take hours.

Alternatives Worth Considering in 2026

Topaz Video AI remains the most popular consumer-facing AI upscaler, but it is no longer the only serious option. HitPaw Video Enhancer, Video2X, and the open-source Real-ESRGAN and VapourSynth-based anime upscalers all produce competitive results in specific scenarios. According to a 2026 roundup published by We Rave You covering AI video upscalers, Real-ESRGAN-NC (the anime-tuned variant) consistently outperforms Topaz Gaia on hand-drawn animation while running on much less hardware.

For professionals working in DaVinci Resolve, the built-in Super Scale feature has improved significantly since version 19 and is now a reasonable choice for 2× scaling where the speed advantage matters more than the last 5 percent of detail. For film restoration houses, tools like MTI Cortex and DaVinci Neural Engine still dominate at the high end, but their learning curve and pricing put them out of reach for most consumers.

The honest takeaway is that Topaz Video AI is the right tool when you need a single application that handles a wide range of source material with minimal setup, and the wrong tool when you need maximum quality in a narrow niche like anime restoration or 8K archival work.

When to Use These Settings and When to Wait

If you are working on a project with a real deadline, apply the recommendations in this guide and run a 10-second test before committing to a full render. If you are a hobbyist working with personal footage, the value of Topaz Video AI at its 2026 retail price (around $299 for a perpetual license, with subscription options near $30 per month) is harder to justify against free alternatives that produce 80 to 90 percent of the quality. The Verge reported in 2019 on AI upscaling of old games, and the field has matured enough that the free options are now genuinely competitive for non-commercial work.

For commercial work, time saved is usually worth more than the license fee. A 30-minute documentary that would take a week to upscale frame-by-frame in After Effects can be processed overnight in Topaz Video AI with results that are often indistinguishable to viewers. That trade-off is what the $299 license is actually buying.

Finally, the field is moving quickly. New models ship every few months, and the recommendations here reflect the state of the software in late 2026. Before starting a major project, check Topaz Labs' release notes for any new models and re-run a short test clip, because the "best" settings in six months may differ from the ones above.

Summary of Practical Recommendations

For the typical user upscaling SD or 1080p to 4K in 2026, the configuration below is a reliable starting point that can be adjusted based on test renders.

These are not universal answers — they are the configuration most often recommended by Topaz Labs' own documentation, by independent reviewers at Fstoppers, PetaPixel, PCMag, and by user reports aggregated across forums in 2025 and 2026. Treat them as a starting point, render a short test, and adjust based on the specific source footage in front of you.