The Pros and Cons of AI Restoring and Colorizing Old Films

There’s a strange magic in watching century-old footage. The grainy, flickering images of a bygone era—silent figures in bowler hats rushing down a city street, or soldiers waving from a departing train—feel distant, almost unreal. For decades, this was our only window into the past. But today, artificial intelligence is rewriting that history, frame by frame. AI-powered tools are now taking this deteriorated black-and-white footage and transforming it into smooth, sharp, and strikingly colorful video. The results can be breathtaking, making events from 1910 look like they were filmed yesterday. But this technological marvel is at the center of a fierce debate. Are we restoring history, or are we creating a synthetic, stylized version of it?

The Case For: Rescuing History from Obscurity

The argument in favor of AI restoration is powerful, and it centers on one key concept: accessibility. For many modern viewers, especially younger generations accustomed to 4K resolution and vibrant color, monochrome footage is a significant barrier. It feels abstract, slow, and “boring.” AI enhancement shatters that barrier, acting as a translator for a modern audience.

Bridging the Empathy Gap

Perhaps the most profound impact of AI restoration is emotional. When Peter Jackson’s team used AI to restore and colorize World War I footage for the documentary They Shall Not Grow Old, the effect was transformative. The grainy, jerky, black-and-white figures of soldiers suddenly became recognizable young men. Viewers could see the color of their eyes, the texture of their wool uniforms, the mud on their boots, and the nervous smiles on their faces. The historical distance collapsed. These were no longer just “historical figures”; they were people. By making the past look like the present, AI fosters a level of empathy and understanding that black-and-white footage often struggles to achieve.

Rescuing What Was Lost

AI’s capability extends far beyond just adding color. The real workhorse of this technology is restoration. Old film stock is notoriously fragile. It suffers from scratches, dust, jitter, warping, and missing frames. Manual, frame-by-frame restoration is an incredibly expensive and time-consuming process, reserved for only the most famous and commercially viable films. AI changes the economics of restoration completely. Neural networks trained on vast datasets of modern, high-definition images can intelligently “in-fill” missing information.

• Noise and Scratch Removal: An AI can analyze a sequence of frames, distinguish between the actual image and random film grain or a scratch, and remove the artifact without softening the underlying detail.
• Stabilization: It can smooth out the jitter characteristic of hand-cranked cameras, providing a stable, watchable image.
• Frame Rate Interpolation: Much old footage was shot at variable frame rates (like 16 or 18 frames per second), making movement look comically sped-up. AI can generate new, synthetic frames to insert between the original ones, smoothing the motion to a natural 24 or 30 fps.

This allows archives, libraries, and museums to restore vast quantities of footage—local newsreels, home movies, scientific records—that would otherwise have been left to decay, unseen and inaccessible.

The Case Against: Painting Over the Past

Despite the “wow” factor, many historians, film purists, and archivists are deeply concerned. They argue that this “restoration” is actually an act of corruption, imposing a modern, digital aesthetic onto a historical artifact and, in the process, destroying the original’s context and intent.

The Problem of “Guesswork”

Colorization is not an exact science; it is an interpretation. An AI doesn’t “know” the color of a woman’s dress in 1925. It makes a highly educated guess based on its training data. But what is that training data? Mostly modern photographs. The AI might color a street scene with the bright, saturated colors of a modern city, not the muted, specific palette of 1920s dyes and paints. It might mistake a khaki uniform for olive green. This “best guess” is then presented as fact, creating a version of history that is, quite simply, wrong. This process steamrolls over historical nuance. The original black-and-white footage is ambiguous, leaving room for interpretation. The colorized version is definitive and, potentially, definitively inaccurate.

The primary danger of AI restoration lies in replacement, not supplementation. If the new, colorized, and “smoothed” version becomes the default, the original artifact is effectively erased from public access. We risk a future where audiences believe the AI’s interpretation is the history, forgetting that the original print—with all its grain, flaws, and lack of color—is the true primary source. Artistic intent is not something an algorithm can or should “fix.”

Authorial Intent and the “Plastic” Look

This leads to the most critical artistic argument: authorial intent. Filmmakers in the black-and-white era were masters of their medium. They used shadow, light, and contrast to create mood and meaning. Think of the stark shadows in film noir or the ethereal glow of a silent film star. These were deliberate artistic choices, not limitations. When AI “enhances” these films, it often does two things that violate this intent:

1. It adds color: This fundamentally changes the film’s mood. A dark, moody thriller can suddenly look like a daytime soap opera, destroying the carefully crafted atmosphere.
2. It removes film grain: Many AI tools use aggressive “denoising” to create a “clean” image. This process often scrubs away fine detail, like skin texture, hair, and fabric weaves, leaving actors looking like waxy, unnatural “plastic” figures.

This “smoothing” effect is a common complaint. The AI, in its quest for a “perfect” image, removes the very texture of the medium itself. It treats the film grain, which is part of the art, as an error to be corrected.

A Tool, Not an Oracle

Ultimately, AI is a powerful tool, but it is not an artist or a historian. The problem isn’t the technology itself, but its automated, one-click application. The best restoration projects, like They Shall Not Grow Old, used AI as a starting point. After the AI did its pass, human artists and historians spent months correcting its mistakes, researching uniform colors, and ensuring the final product was as historically accurate as possible. The danger comes from low-effort, automated enhancements that prioritize a “clean” look over authenticity. As this technology becomes more accessible, we may see a flood of “fake” history—old films and photos scrubbed of their character and painted with an algorithm’s best guess. The key, then, is to treat these new versions as supplements, not replacements. The original, untouched monochrome print must always be preserved and respected as the definitive historical document.