Top 10 Results Using Topaz Video Enhance AI: Before & After ExamplesTopaz Video Enhance AI (VVEAI) is one of the most popular consumer tools for video upscaling, denoising, and frame-interpolation. In this article I showcase ten notable before-and-after examples that demonstrate what VVEAI can — and cannot — do. Each example includes the source footage characteristics, the model and settings used, the concrete outcome, and practical tips so you can reproduce similar results.
1) 480p TV promo → 1080p (face clarity and text legibility)
Source: A 480×270 (approx. 480p) TV promo clip with heavily compressed text overlays and small, low-contrast faces.
Settings used
- Model: Artemis (or Artemis HQ for better faces)
- Output: 1080p (2×)
- Denoise: Mild
- Sharpening: Low
- Face Refinement: On
Outcome: Faces became visibly cleaner and more natural, compression artifacts around text were reduced, and small captions regained legibility. Edges appear less smeared compared with simple bicubic upscaling.
Tips: Turn on Face Refinement for human subjects and use Artemis HQ if your GPU can handle it. If text remains soft, do a second pass focused on sharpening the text area or isolate text with masks in an editor.
2) Old 720×480 home video → 4K (detail recovery and grain handling)
Source: Interlaced VHS capture at 720×480, scan contains chroma noise, combing, and heavy analog grain.
Settings used
- Model: Standard or Proteus (for custom tuning)
- Output: 4K (4×)
- Deinterlace: Enabled (if necessary)
- Denoise: Moderate (preserve some grain)
- Grain: Preserve a little to avoid plastic look
Outcome: Significant perceived detail increase on faces and fabrics; chroma noise and combing largely removed while a natural film-like grain remained. Some fine texture is synthetic but plausible.
Tips: Deinterlace first if the source is interlaced. Proteus lets you tweak model strength per-preset; use that to balance denoising vs. detail recreation. Export one short clip to check grain levels before processing the whole file.
3) 360p anime → 1080p (line art and color flatness)
Source: Low-resolution anime rip (360p) with blocky outlines and posterized color bands.
Settings used
- Model: Theia or Gaia (anime-oriented models vary by release)
- Output: 1080p (3×)
- Denoise: Low (to keep line detail)
- Color: Exported with original color profile
Outcome: Cleaner line art and smoother edges with minimal color banding. Some small background textures were hallucinated but matched style. Motion areas avoided stuttering; outlines stayed stable.
Tips: For anime, prefer models trained on animation styles (when available). Avoid aggressive denoise — outlines can blur. If your VVEAI version lacks a dedicated anime model, test Proteus presets emphasizing edge preservation.
4) Smartphone night footage → 1080p (noise reduction and detail restoration)
Source: iPhone low-light clip with heavy luminance noise, soft details, and automatic aggressive sharpening artifacts.
Settings used
- Model: Artemis or Dione (depends on release)
- Output: 1080p (1.5–2×)
- Denoise: Strong
- Sharpening: Moderate
Outcome: Noise was substantially reduced and fine detail such as hair and texture regained definition without obvious over-sharpening. Colors looked cleaner and highlights less blown.
Tips: If smartphone denoise removes too much texture, reintroduce subtle grain in post. For mixed lighting, clip highlights first or do color correction after upscaling.
5) Action camera 720p → 4K (motion preservation and artifact handling)
Source: GoPro-style footage at 720p, lots of motion, compression artifacts from long GOP encoding.
Settings used
- Model: Iris (or a temporal model that handles motion)
- Output: 4K (4×)
- Motion compensation: Enabled
- Denoise: Moderate
Outcome: Motion felt more natural than naive frame-blend upscales, with fewer ghosting artifacts. Textures on clothing and terrain showed plausible high-frequency detail.
Tips: Choose VVEAI models that explicitly use temporal information for action footage. If you see ghosting, reduce temporal radius or try a different motion model.
6) Silent-era black-and-white film → 2K (grain management and artifact reduction)
Source: 1920s black-and-white film scan with scratches, dust, and flicker.
Settings used
- Model: Proteus or a film-focused model when available
- Output: 2K (2–3×)
- Denoise: Careful — remove dust/scratches but preserve film grain
- Manual pre-clean: Use dedicated restoration tools for large scratches
Outcome: Cleaner frames with reduced flicker and preserved film grain; faces and set details gained clarity while maintaining period texture. Automated fixes can’t fully remove deep scratches — manual restoration still best for heavily damaged frames.
Tips: Pre-clean major defects in a restoration tool (e.g., DaVinci Resolve or specialized plugins) before VVEAI. Keep grain to avoid the “CGI” look.
7) Low-bitrate YouTube 480p → 1080p (blocking and color banding fixes)
Source: Streamed, re-encoded YouTube clip at low bitrate with blocking and banding.
Settings used
- Model: Standard/Artemis
- Output: 1080p (2×)
- Denoise: Moderate
- Banding: Reduced by upscaler’s smoothing
Outcome: Block artifacts reduced and surfaces smoothed; banding softened, making gradients and skies look more pleasing. Fine texture is partially synthetic but visually acceptable.
Tips: If video contains logos/text, compare before/after to ensure text is not softened. For archival or legal uses, note that some detail is AI-hallucinated.
8) Documentary interview 1080p → 4K (skin tones and background details)
Source: Modern 1080p interview footage, studio lighting, important background elements.
Settings used
- Model: Artemis HQ or a high-quality temporal model
- Output: 4K (2×)
- Face Refinement: On
- Denoise: Low
Outcome: Skin textures remained natural while background details like bookshelves sharpened, yielding a crisp final image suitable for 4K distribution. Minimal artifacts around motion.
Tips: Use Face Refinement carefully — too strong can over-smooth. Consider exporting two versions (face-refined and non-refined) to pick the better balance.
9) Vintage animation (hand-drawn) → 2K (line stability and color preservation)
Source: 480p hand-drawn animation scan with jittery lines and color inconsistency.
Settings used
- Model: Animation-capable model or Proteus with edge-preserve tuning
- Output: 2K (2–3×)
- Denoise: Low
- Line stabilization: Performed in combination with external tools if needed
Outcome: Lines were steadier and colors more consistent across frames, improving the perceived production quality. Some small artist-intended imperfections were slightly altered.
Tips: Combine VVEAI with manual frame stabilization tools for shaky linework. Preserve color profiles to avoid hue shifts.
10) Gaming capture 720p → 4K (text, HUD, and fine texture enhancement)
Source: Fast-paced gaming footage at 720p showing UI/HUD elements, small text, and screen tearing artifacts.
Settings used
- Model: Iris or a model trained for synthetic/CG content
- Output: 4K (4×)
- Motion compensation: On
- Denoise: Low to keep HUD crisp
Outcome: UI elements and small text became far more readable, textures on character models and environments gained clarity. Very little temporal smearing when the correct motion model was chosen.
Tips: For overlays and HUD, consider doing a two-pass workflow: upscale the gameplay with a temporal model and the HUD/text with a spatial (frame-by-frame) model or mask the HUD to preserve clarity.
Common caveats and limitations
- AI upscalers commonly “hallucinate” detail: new high-frequency information is synthesized, not recovered from source. Forensically accurate restoration is not guaranteed.
- Strong denoising can produce plastic-looking skin or remove desirable texture. Balance is key.
- Motion-heavy scenes risk ghosting; experiment with temporal vs spatial models and motion radius.
- Text and logos sometimes soften; mask and treat them separately if fidelity is essential.
- Processing time and GPU requirements can be substantial; test short clips and check VRAM usage.
Quick workflow to reproduce good results
- Inspect source (interlaced? noise? text?).
- Pre-process: deinterlace, repair large scratches, stabilize if needed.
- Choose model: temporal models for motion; Artemis/face models for people; animation models for cartoons.
- Start with moderate denoise and one test clip at full settings.
- Adjust Grain/Sharpening and rerun test.
- Mask text/HUD for a separate pass if needed.
- Color grade and add subtle film grain in post for natural results.
Conclusion
Topaz Video Enhance AI is a powerful tool that can produce dramatic visual improvements across many types of footage — archival film, animation, smartphone video, gaming capture, and more. The ten examples above illustrate typical outcomes, settings, and practical tips to get the best results while avoiding common pitfalls. With careful previews, selective masking, and the right model choices, you can transform low-resolution or degraded clips into convincing high-resolution results suitable for modern viewing.
Leave a Reply