What is the best AI video upscaler for old film in 2026?

The best AI video upscaler for old film depends on what kind of 'old film' you are working with, but as of August 2026 the strongest overall pick for home users restoring 8mm, Super 8, VHS, and DVD-era footage is Topaz Video AI (now frequently branded Topaz Video Enhance AI), with VideoProc Converter AI as the best all-in-one alternative for people who also need trimming, conversion, and basic editing, and open-source pipelines built on Real-ESRGAN or waifu2x-style models as the free option for technically confident users. If your source is genuine celluloid — scanned 8mm, 16mm, or 35mm reels — no single tool does everything: you want one tool for grain-aware denoising and detail reconstruction, and ideally a separate pass for scratch and dust removal before upscaling. This guide breaks down exactly which tool fits which scenario, why old film behaves differently from modern digital video when it hits an upscaler, how to run a restoration workflow step by step, and where the current tools still fail.

The Direct Answer: Which Tool Wins for Old Film

Also worth reading: Topaz Video AI vs RTX Video: which AI video upscaler should you actually use in 2026? · SeedVR2 vs desktop upscaler comparison: which is better for upscaling video to 4K in 2026? · What is the best AI video upscaler in 2026 for transforming low-resolution footage into pristine 4K?

For most people asking this question, the answer is Topaz Video AI. It has been the reference point in nearly every serious comparison of AI video enhancers since around 2021, and by 2026 its Proteus and Artemis model families remain the most reliable at reconstructing plausible detail on low-resolution analog sources without inventing obvious artifacts. It handles interlaced VHS captures, soft DVD rips, and grainy film scans natively, offers frame interpolation to convert choppy 16fps or 18fps silent-film footage to smooth 24p or 60p, and exports up to 4K and beyond. Pricing sits at roughly $299 for a perpetual license with one year of updates, which is expensive but reasonable if you have more than a handful of reels to restore.

VideoProc Converter AI is the better choice if you want an all-in-one toolkit rather than a dedicated upscaler. Fstoppers tested it as a full media toolkit and found its AI enhancement module respectable on old photos and video alike, though it does not match Topaz on the hardest, grainiest film scans. At roughly $25–$40 per year (or a lifetime tier around $60–$80), it costs a fraction of Topaz and bundles conversion, compression, and editing, making it the pragmatic pick for casual family-archive projects.

If budget is zero, open-source tools are genuinely viable in 2026. A pipeline using ffmpeg for deinterlacing, Real-ESRGAN (or the video-oriented Real-CUGAN) for upscaling, and RIFE for frame interpolation can produce results within striking distance of commercial software. HackerNoon documented a hobbyist restoring black-and-white photos to 4K color entirely with open-source AI, and the same approach extends to video. The tradeoff is time: expect hours of GPU processing per hour of footage, plus a steeper learning curve than any GUI product.

Why Old Film Is Harder Than Modern Footage

AI upscalers are trained mostly on modern digital video, so they make assumptions that break down on analog material. Old film has four problems that modern 1080p phone footage does not: heavy organic grain, physical damage (scratches, dust, hair, splice marks), unstable framing from projector gate weave, and non-standard frame rates. Silent-era films were often shot at 16 to 20 frames per second; home movie reels typically run at 16 or 18fps; early television transfers may be interlaced at 25 or 29.97fps with field-order quirks.

A naive upscale simply enlarges all of these defects. Grain becomes giant colored blobs, scratches become glowing scars across the frame, and gate weave turns into visible wobble at 4K. Worse, some aggressive models interpret film grain as noise to be removed and then hallucinate plastic-looking skin textures to fill the gap — a failure mode widely mocked in coverage of AI-processed classic cartoons, including Engadget's reporting on the oddly smoothed Super Mario Bros. Super Show re-release. That project became a cautionary tale: aggressive AI processing stripped the texture of hand-drawn animation and produced uncanny, smeared frames.

The practical lesson is that good old-film restoration is a sequence of targeted operations, not one button. Denoise first (gently), stabilize second, deinterlace third if needed, then upscale last with a conservative model setting. Tools that force you through this order explicitly — or expose per-stage controls — produce far better results than one-click enhancers.

Comparison Table: Leading Options in 2026

Other names worth knowing: Adobe has previewed AI upscaling features aimed at old, fuzzy footage since October 2023 (Jess Weatherbed's reporting covered Adobe's research previews), and its Firefly-based video tools continue to expand, but as of mid-2026 there is no standalone Adobe product that matches Topaz for archival film work. Free web-based enhancers reviewed by outlets like North Penn Now and ePHOTOzine are fine for short clips but impose length limits, watermarks, or upload privacy concerns that matter when your source is a one-of-a-kind family reel.

Step-by-Step Workflow for Restoring Old Film to 4K

Start with the best possible capture, because no upscaler can recover information that was never digitized properly. For physical reels, a frame-by-frame scan (a telecine or scanner service charging roughly $0.10–$0.30 per foot, or a DIY unit like a Wolverine or a modified projector setup) beats a projected-screen recording every time. For VHS, capture with a time-base corrector or a decent USB capture device at full resolution before touching AI tools. Pocket-lint's writer revived a DVD collection with near-perfect results precisely because the source rips were clean and lossless.

Step one is cleanup before upscaling. Run a gentle denoise/despeckle pass to knock down dust specks and heavy grain while preserving texture. In Topaz, this means using the Iris or Proteus model with moderate settings rather than maximum strength. Step two is stabilization if the footage suffers from gate weave — Topaz includes a stabilizer module, or you can use Deshaker in VirtualDub or ffmpeg's vidstab filter. Step three is deinterlacing for broadcast-era sources, done once, correctly, with the right field order.

Step four is the actual upscale. Target 4K (3840×2160) rather than 8K unless you have a specific display reason; doubling resolution twice from standard-definition sources already stretches the model's ability to synthesize believable detail. Keep sharpening and detail-recovery sliders conservative — around 30–50% strength rather than 100% — because overdriven settings are what produce the waxy, AI-slop look. Step five is optional frame interpolation to bring 16–18fps footage to 24fps for a cinematic feel, or 60fps only if you accept the soap-opera effect many viewers dislike on narrative film. Finally, export at high bitrate (H.264 or H.265 at 20–40 Mbps for 4K) and always keep your original capture untouched as a master.

Common Mistakes That Ruin Old-Film Restorations

The most frequent error is cranking every slider to maximum. Users see a preview at 100% detail recovery, love the apparent sharpness on a still frame, then discover motion smearing, flickering grain, and hallucinated faces across the full clip. Conservative settings applied consistently almost always look better in motion. Related to this is skipping the pre-cleanup pass: upscaling a dusty, scratched scan bakes those defects into every enlarged pixel, making them harder to remove afterward.

The second cluster of mistakes involves frame rate. Interpolating silent film from 16fps straight to 60fps destroys the deliberate cadence of the era and produces the artificial smoothness audiences associate with cheap TV conversions. If you interpolate at all, 24fps is the defensible target for cinematic material. Similarly, applying modern colorization or HDR-style tone mapping to black-and-white footage is a creative choice, not a restoration — label it as such if you share the result publicly, because mislabeled 'restored' clips have generated real backlash online.

Third, do not trust one-click web upscalers with irreplaceable footage. Upload services compress your video server-side, may retain copies, and give you limited control over model parameters. For a unique family reel, process locally. Fourth, do not delete intermediates: save your denoised, stabilized intermediate file separately from the final upscale so you can re-run the upscale with different settings later without redoing cleanup. Storage is cheaper than reprocessing.

When to Act and What It Costs

There is a real urgency argument for physical media. Magnetic tape (VHS, Video8, Hi8) chemically degrades and suffers print-through loss every year; estimates commonly cited in archiving circles give tapes 10–30 years of remaining playable life depending on storage conditions, and many home tapes from the 1980s are already past their best. Celluloid fares differently — nitrate stock is dangerously unstable, acetate suffers vinegar syndrome, but polyester-based film stored cool and dry can last centuries. Either way, digitization is the clock-sensitive step; AI upscaling can be done anytime afterward, repeatedly, as tools improve.

Budget-wise, plan for three tiers. The free tier: open-source tools plus your existing computer, costing nothing but electricity and patience — realistic if you have a GPU with 6GB+ VRAM. The mid tier: VideoProc Converter AI at $25–$80 covers most family-archive needs with minimal fuss. The prosumer tier: Topaz Video AI at ~$299, justified when you have dozens of hours of footage or commercial restoration work. Add scanning costs if starting from physical reels: professional 2K scans of 8mm run roughly $0.15–$0.35 per foot, meaning a typical 50-foot reel (about 3–4 minutes) costs $8–$18 per reel before any AI processing.

Timing matters less than consistency: because AI models update several times a year, avoid batch-processing everything the week you buy a tool. Restore a test reel first, refine your settings, then commit to the full archive. Your original captures remain valid inputs forever, so improvements in 2027 models can be applied retroactively.

Where These Tools Still Fall Short

Honest assessment requires acknowledging limits. Current upscalers cannot recover true detail that was never captured — they synthesize plausible texture based on training data, which means faces in distant shots may be subtly reshaped toward generic averages, text remains unreliable, and period-specific objects can drift in appearance. For documentary or evidentiary purposes, AI-upscaled footage should never be presented as enhanced ground truth. The Engadget coverage of AI-mangled cartoons illustrates the cultural cost of careless application: when studios apply aggressive models to beloved classics, fans notice, and the backlash is deserved.

Grain remains a philosophical battleground. Removing it entirely makes footage look sterile and modern; keeping it fully makes 4K gains marginal. The sweet spot is partial grain management, which requires manual tuning that automated presets rarely get right. Scratch and dust removal inside general-purpose upscalers is still weak compared to dedicated restoration software used in professional film labs. And processing speed, while vastly improved since 2023, still means a feature-length film takes anywhere from under an hour on a top-tier RTX GPU to overnight on integrated graphics.

None of this changes the bottom line: for anyone with aging home movies, VHS tapes, or DVDs, 2026 is the right moment to digitize and upscale. The tools are mature enough to produce genuinely watchable 4K results, cheap enough for hobbyists, and improving fast enough that anything you preserve now will only look better when reprocessed later.

Final Recommendation

Choose Topaz Video AI if restoration quality is the priority and you will process more than ten hours of footage; choose VideoProc Converter AI if you want one affordable program that handles enhancement alongside everyday video tasks; choose the open-source route if you enjoy tinkering and refuse to pay subscription-era prices. Whichever path you take, follow the discipline that separates good restorations from AI-slop: clean before you upscale, stay conservative with detail sliders, respect original frame rates, and never discard your untouched masters.