I need a video upscaler that de-noises old footage and does not create "wax face" smoothing artifacts.

Yes. If your priority is denoising old footage while preserving real facial texture, rather than making faces look airbrushed, I’d shortlist these:

1. Topaz Video — best all-in-one choice

Topaz Video

For old, grainy, compressed footage, Proteus is the model I’d start with. It is specifically designed for low/medium-quality footage and gives you manual control over noise reduction, sharpening, detail recovery, de-halo, etc.docs.topazlabs.com

The important part for your concern: don't crank the enhancement/detail settings. Topaz explicitly provides a Recover Detail control that mixes original source detail back into the result, and its newer Proteus Natural is designed to respect the source rather than heavily modify it.docs.topazlabs.com

My starting recipe: Proteus/Proteus Natural → modest noise reduction → low sharpening/detail enhancement → increase Recover Original Detail → inspect faces at 100%.

2. Neat Video + a separate upscaler — best for preservation

Neat Video

If this is valuable archival footage, this may actually be my preferred workflow.

Neat Video is fundamentally a noise-reduction tool rather than a generative face-enhancement system. Its temporal filter analyzes consecutive frames to distinguish noise from actual detail, and it provides extensive control over the strength of the filtering.www.neatvideo.com

That makes it much less inclined to invent a new-looking face. You can:

Original → Neat Video denoise → gentle upscale/sharpen → output

rather than asking one AI model to simultaneously remove noise, invent detail, sharpen skin and enlarge the image.

3. Topaz's dedicated Denoise — interesting if you want minimal alteration

Topaz also has a Denoise filter based on Nyx High-Fidelity that is specifically intended to remove grain without additional enhancement or sharpening.docs.topazlabs.com

That is potentially excellent for your use case: denoise first, then do a restrained upscale afterward.

What I'd choose

One important warning: I would not start with an aggressive “face recovery” model. Topaz's Iris is specifically trained for face recovery on low/medium-quality footage, but its job is to regenerate fine detail—which is exactly where the "that doesn't quite look like the person anymore" problem can creep in.docs.topazlabs.com

If you tell me what the source is (VHS, 8mm/Hi8, MiniDV, old DVD, 480p web video, film scan, etc.) and its resolution/frame rate, I can give you a very specific anti-wax settings recipe for the best option.

Topaz Video AI is the best commercial option for de-noising old footage without creating "wax face" smoothing artifacts, provided you use its Iris or Proteus models and manually dial down face recovery or add synthetic film grain.

Most AI upscalers suffer from the "wax face" or plastic skin problem because aggressive generative face recovery models (like Topaz's *Recovery * or generic GAN face-swappers) hallucinate smooth skin details over low-resolution features. To combat this, you need software that lets you control the balance between denoising strength and texture retention.

Top Solutions and Settings to Avoid "Wax Faces"

Step-by-Step Workflow to Prevent Plastic Skin

1. Avoid Automatic "Face Enhancement" Sliders:
   Features marketed as "Face Recovery" or "Face Restoration" (powered by models like GFPGAN or CodeFormer under the hood) are the primary culprits behind the plastic, waxy look. If the software forces face generation, set its strength to the absolute minimum (10% to 20%) or turn it off completely.
2. Use Tunable Models (Proteus in Topaz):
   Instead of using "Gaia" or fully automated presets, use Proteus in Topaz Video AI . Proteus lets you explicitly set sliders for Noise, Blur , and Detail Recovery . Keep noise reduction moderate so you don't scrub away natural skin textures and micro-contrast.
3. Reintroduce Film Grain:
   Old footage naturally has grain. When an AI de-noises it aggressively, skin looks flat and synthetic. Always use the Add Grain feature in post-processing or inside the upscaler (set to around 3% to 8%). Adding subtle, authentic grain breaks up digital color banding and hides minor smoothing artifacts.
4. Consider Hybrid Algorithmic Denoising:
   If AI models are too unpredictable for your source material, use professional editing suites like DaVinci Resolve Studio . Its built-in spatial and temporal noise reduction combined with Super Scale upscaling lets you clean up noise mathematically without an AI neural net "guessing" what a face looks like.

If you can share what resolution your source footage is (e.g., 480i VHS, 8mm film scan, or old DV tape) and whether you have a dedicated Nvidia/AMD GPU for local processing, I can recommend the exact configuration or open-source local scripts to use.

Avoiding the dreaded "wax face" (plastic, over-smoothed, uncanny-valley skin textures) while upscaling and de-noising old footage requires specific software where you can dial back aggressive face recovery models and manually reintroduce or preserve organic film grain.community.topazlabs.com+2

The top desktop software solutions that handle this balance best, along with tips on how to configure them, include:

• Topaz Video AI : Widely considered the industry standard for archival and old footage restoration. To avoid wax faces here, turn off or lower the dedicated "Face Recovery" slider (which often triggers the plastic look) and instead use models like Gaia (which preserves natural details better than generative models). Crucially, use Topaz’s Add Grain feature in post-processing to lay a subtle, realistic layer of texture back over the frame, masking any residual digital smoothing. You can check out details or download it directly from Topaz Labs.docs.topazlabs.com+2
• VapourSynth / RIFE / Custom Avisynth Scripts : If you have technical know-how, open-source script environments give you 100% control. Combining classic spatial-temporal de-noisers (like KNLMeansCL or hqdn3d ) with a conservative upscale model lets you clean noise without an AI guessing what a face looks like. It is completely free, though it requires a steep learning curve.
• Nero AI Video Upscaler Pro : Features specialized recovery and realistic models that focus on micro-details rather than aggressive hallucination. It is more automated than Topaz, but you'll still want to keep sharpness settings modest to prevent edge-haloes and plastic skin. You can review system specs and features on Nero AI Video Upscaler Pro.[1]
• DVDFab AI Upscaler : Good for quick batch processing, though like most plug-and-play AI tools, you should run test clips on a short 5-second loop first to ensure it doesn't over-process close-up shots.[1]

Pro-Tips to Prevent the "Wax Face" Effect:

1. Denoise separately from upscaling: If your tool allows a two-pass approach, do a light temporal denoise first, then upscale with a minimal sharpening footprint.
2. Add a touch of grain back in: Real film or tape has grain. AI hates grain and tries to erase it into mush. Forcing a tiny amount of synthetic grain back over the final render tricks the human eye into seeing organic sharpness instead of plastic.community.topazlabs.com+2
3. Lower Face Recovery strength: If the software has a dedicated facial AI enhancement toggle, drop it to 10%–20% or disable it entirely for distant/medium shots where it tends to misinterpret low-res features.community.topazlabs.com+1

To help narrow down the best choice, tell me:

• What is the source format (e.g., 8mm film, VHS tape, MiniDV, old DVD)?
• Do you prefer a one-click desktop app or are you comfortable with open-source/advanced command-line tools?

Avoiding the dreaded "wax face" (or plasticky, over-smoothed AI hallucination) on old footage comes down to choosing software that lets you dial down AI strength, disable aggressive face recovery models , or use grain-retention settings . When AI models try too hard to "guess" facial features on low-res or grainy sources, they flatten skin textures into smooth plastic.

The top software options for managing and avoiding this artifacting on old footage include:

• 1. Topaz Video AI : Widely considered the industry standard for consumer/prosumer AI upscaling, but it is notorious for wax faces if left on automatic ("Autopilot"). To avoid this, use models like Proteus or Iris rather than older legacy models, and manually lower the Face Recovery slider (or turn it off completely if the faces look synthetic). Proteus lets you manually tweak parameters for noise, blockiness, and detail recovery, keeping a much more natural, film-like texture. You can check it out on the Topaz Labs Official Website.youtube.com+1
• 1. DaVinci Resolve Studio (Neural Engine) : If you want a more hands-on, traditional grading/restoration approach without heavy generative AI "hallucinating" features, Resolve's Super Scale combined with its spatial/temporal noise reduction is phenomenal. If you use its Face Refinement tool (Studio version only), make sure to keep the skin smoothing and texture sliders very subtle. It won't invent fake pixels out of thin air like generative models do, which entirely prevents the wax-face look, keeping the organic sensor noise or film grain intact. Learn more on the Blackmagic Design DaVinci Resolve Page.youtube.com+2
• 1. VapourSynth / Real-ESRGAN (Open Source / Custom) : For advanced users willing to use script-based restoration, VapourSynth allows you to combine classic algorithms (like BM3D for natural denoising) with custom, milder AI upscaling models. Because you control every single node in the pipeline, you can apply light denoising and a gentle upscale model without a black-box AI guessing a face incorrectly.

Tips to Eliminate Wax Face Regardless of Tool:

• Add back a touch of grain/noise: Many restorers upscale cleanly and then add back 3% to 5% subtle synthetic film grain over the final output. This tricks the human eye into ignoring minor digital smoothing and breaks up flat, plasticky skin gradients.
• Never use 100% AI strength: Always blend the enhanced/upscaled result with the original source (around 20–30% original mix if the software allows) or pull the detail/recovery sliders down to the 30–50% sweet spot.

If you want to narrow this down, let me know:

• What is the resolution and source format of your old footage (e.g., 480i miniDV, 8mm film transfer, old VHS)?
• Are you looking for a one-click desktop app or do you have a GPU/technical setup for advanced tools?

For your specific requirement — denoise old footage while preserving real skin texture and avoiding “wax face” smoothing — I would look at tools that let you control restoration strength rather than ones that apply aggressive AI beautification.

Best overall:topazlabs.com

This is probably the closest match for old VHS, MiniDV, home movies, archival clips, and compressed footage. It has separate models for different problems instead of one “make it pretty” filter.Topaz Labs Docs

Recommended workflow:

1. Denoise first

• Try Nyx for noisy footage, especially VHS, MiniDV, dark footage, or film scans. It is designed around noise reduction while retaining detail.Topaz Labs Docs
• Keep the strength moderate. Over-cleaning is what usually creates plastic skin.

2. Upscale second

• Use Proteus (Manual) for most old footage. It gives you control over detail recovery and sharpening rather than forcing a heavy AI look.Topaz Labs Docs
• Start conservative: - Detail recovery: low to medium
• Sharpening: low
• Noise reduction: enough to remove crawling noise, not all texture
• Add a small amount of grain if the result looks sterile

3. Be careful with face models

• Iris can help damaged faces, but it is also the model most likely to “invent” facial detail if pushed too hard. Use it only when faces are genuinely too degraded.Topaz Labs Docs

Best for maximum naturalism: Blackmagic Designblackmagicdesign.com

If your priority is “make it look like the original camera captured it better,” Resolve can be a safer choice:

• Use temporal noise reduction lightly.
• Use Super Scale conservatively.
• Avoid heavy sharpening.

It tends to preserve the character of footage better, though it may recover less detail from extremely damaged sources than Topaz.Topaz Labs

A restoration workflow I’d use

For old family footage:

1. Deinterlace properly (many VHS/DVD/MiniDV sources are interlaced).
2. Remove chroma noise and compression artifacts.
3. Apply gentle temporal denoise.
4. Upscale 2× first (480p → 1080p is often more natural than jumping straight to 4K).
5. Add back a little grain if needed.

Avoid:

• Maximum AI enhancement sliders
• “Face recovery” on every shot
• Strong sharpening
• 4× upscale on already-soft footage

Those are the settings most likely to produce the waxy mannequin look.

If you tell me the source (VHS, Hi8, MiniDV, DVD, old phone video, film scan, etc.) and target (1080p or 4K), I can suggest a more specific model/settings chain.