A badly rusted 1963 Saab 96 sits forgotten inside an old building. Its red paint has faded, corrosion has eaten deeply into the body, and the car appears to have spent decades untouched. The video then follows what looks like a complete restoration: dismantling, rust repair, body preparation, mechanical work, painting and final reassembly.
There is one fundamental problem with the story: this restoration never happened.
The YouTube video “1963 Saab 96 Restoration | Forgotten Winter Rally Car Saved After 40 Years” is a remarkably convincing example of a rapidly growing type of automotive content: AI-generated restoration videos designed to resemble documentary footage of real projects.
For viewers familiar with early Saabs, the illusion starts to collapse quite quickly. The car changes subtly between scenes, some body details are inconsistent, several components do not belong to a 1963 Saab 96, and the supposed engine is particularly revealing. At the same time, the overall restoration sequence is surprisingly believable, which is exactly what makes this type of content worth examining.
AI Has Learned How a Restoration Should Look
The video follows the structure used by countless genuine restoration channels. First comes the discovery of a severely neglected car, followed by dismantling, inspection, rust removal, body repairs, sanding, filler, preparation, paint and reassembly.
Even the fictional restorer is presented convincingly. He wears broadly consistent work clothes, handles tools in plausible surroundings, photographs components and appears to document the process carefully. The workshop scenes are visually coherent enough that a casual viewer could easily accept them as genuine footage.
That becomes particularly apparent when compared with a real project. SaabPlanet previously followed a Saab 96 restoration that consumed around 800 hours and £46,000 before the finished car was sold for roughly half its restoration cost. The real project involved precisely the problems that make restorations unpredictable: deciding how far to strip the body, dealing with previously hidden corrosion, rebuilding mechanical components and watching costs escalate as additional work emerged.
The AI video has learned to reproduce the appearance and sequence of this process without having to solve any of those physical problems.
The Engine Gives the Game Away
One of the clearest problems appears when the drivetrain is shown outside the car.
The early history of the Saab 96 is particularly important here. The model arrived in 1960 as an evolution of the Saab 93 and initially retained Saab’s distinctive two-stroke engineering.
A genuine standard 1963 Saab 96 would normally be powered by Saab’s 841cc water-cooled three-cylinder two-stroke engine, driving the front wheels. The detailed Saab 96 history at SaabMuseum.com records the 841cc three-cylinder two-stroke as the defining engine of this early generation. A surviving 1963 US-market example documented by Hemmings provides an even more useful reference: its original 841cc Solex-carbureted three-cylinder produced 42 hp and was paired with a column-shifted three-speed gearbox.
The machinery shown in the AI video is something entirely different. It looks old, dirty and mechanically plausible, but it does not resemble the correct Saab two-stroke powertrain.
That discrepancy neatly summarizes one of the limitations of generative video: AI can create something that looks like an old engine without understanding which old engine belongs in a particular historic car.
For a Saab enthusiast, that is enough to raise immediate suspicion. For someone encountering an early Saab for the first time, the same sequence can easily pass as authentic.
The ‘Winter Rally Car’ Story Is Cleverly Chosen
The fictional backstory is particularly effective because it is rooted in genuine Saab history.
Calling the car a forgotten winter rally car sounds entirely plausible. Saab 96 competition history is extensive, and SaabPlanet’s retrospective on the model covers the period when Erik Carlsson won the Monte Carlo Rally in 1962 and 1963 and the RAC Rally in 1960, 1961 and 1962.
There are still genuine survivors from that era. The Saab Heritage Car Museum USA preserves the actual three-cylinder two-stroke Saab 96 that Carlsson and Stuart Turner used to win the 1960 RAC Rally. The car itself was rediscovered decades later and its identity reconstructed from its history, making it an interesting real-world counterpoint to the invented provenance in the AI video.
SaabPlanet has also covered a Saab 96 Monte Carlo 850 built in connection with Saab’s competition program, while the Saab Car Museum Supporters organization documents how extensively two-stroke Saabs competed in period racing through its Group 3 Saab 96 project.
The AI story therefore works because it mixes fiction with authentic historical context. A neglected early Saab rally car rediscovered after several decades sounds like something that genuinely could happen.
The Car Does Not Always Remain the Same Car
The bodywork provides another clue. The opening ‘barn find’ image is highly convincing at first glance. The silhouette is recognizably Saab 96, while the rust, faded paint and damaged panels are rendered very effectively.
Closer inspection reveals problems. Grille details, lights, bumpers, body openings and smaller trim elements are not consistently accurate, and some appear to change as the supposed restoration progresses.
This matters more than a simple visual mistake. A real restoration has unavoidable physical continuity. Damage visible when the car arrives must either remain or be physically repaired. Panels cannot subtly change shape between stages, and components removed during dismantling have to correspond with those subsequently restored and reinstalled.
A genuine Saab 96 restoration demonstrates exactly that continuity. In another documented racing Saab 96 restoration previously featured on SaabPlanet, mechanical changes can be traced individually: the engine and gearbox were removed, seals replaced, gearbox overhauled, carburetion changed and other specific components addressed. Every intervention has a mechanical consequence and a history.
Generative video only has to make the next scene look convincing.
Why the Video Is Still Interesting to Watch
Despite all these inaccuracies, the AI restoration remains surprisingly watchable. The general workflow is convincing, the visual progression is satisfying, and the transformation from rusted wreck to finished car follows the structure familiar to anyone who watches restoration content.
For casual viewers, that may be enough. They receive much of the immediate satisfaction produced by a genuine restoration video: visible progress, mechanical activity and a dramatic before-and-after transformation.
For enthusiasts, the attraction can be completely different. The video becomes a game of identifying errors: spotting the wrong engine, changing panels, incorrect trim and components that do not belong on an early Saab 96.
That creates a curious situation in which the same video can entertain two audiences for opposite reasons. One viewer watches the restoration; another watches the AI trying to maintain the illusion of a restoration.
The Real Problem Begins When Viewers Believe It
The larger issue is therefore not simply that AI can generate fictional automotive restorations. The problem begins when viewers do not realize that the footage is synthetic.
A knowledgeable Saab owner may recognize the mistakes almost immediately, but someone seeing an early Saab for the first time has no reliable reference point. That person may genuinely believe that the video documents the recovery and restoration of a real 1963 Saab 96.
Photorealistic video gives incorrect information a particularly persuasive form of visual authority. Instead of merely reading an incorrect engine specification somewhere online, the viewer can now watch a realistic-looking mechanic remove that incorrect engine from what appears to be a real Saab.
The fictional component has acquired a workshop, tools, a mechanic and an entire visual history.
Why Some Enthusiasts React So Negatively
The objections visible among knowledgeable viewers are understandable because classic-car communities place considerable value on provenance and authenticity.
For many enthusiasts, the attraction of restoration content comes from following an actual object through real problems. Rust repair takes time, parts have to be sourced, seized components resist removal, budgets change and additional damage frequently appears only after dismantling.
The contrast with a genuine project is enormous. The £46,000 Saab 96 restoration documented by SaabPlanet required six months and approximately 800 hours of actual work. Those figures matter because somebody really spent those hours repairing a physical car.
When a viewer discovers after watching an apparently similar timeline that there was no forgotten Saab, no difficult restoration and no finished car, the impressive visuals can suddenly lose much of their emotional value.
AI-generated restoration content is not necessarily worthless because of that distinction, but it needs to be understood for what it is: visualization rather than documentation.
YouTube Already Requires Disclosure of This Kind of Content
The distinction is also relevant under YouTube’s own rules. According to YouTube’s current GenAI disclosure policy, creators must disclose content that uses AI to meaningfully generate or alter photorealistic material, including a realistic scene depicting something that did not actually occur. YouTube distinguishes this from minor production assistance such as AI-generated outlines, captions, image enhancement or other changes that do not materially alter what happened.
The platform strengthened that approach in May 2026, announcing that disclosure labels for photorealistic AI-generated or meaningfully altered long-form videos would appear directly below the player rather than being buried deeper in the description. YouTube also began rolling out internal signals intended to help identify AI-generated material automatically.
YouTube even provides a separate ‘How this content was made’ explanation, where viewers can encounter a ‘Made with AI’ disclosure when material has been meaningfully altered or synthetically generated.
For something such as a fictional Saab restoration, that context is important because the video deliberately adopts the visual conventions of documentary footage.
Saab 96 Is a Particularly Tough Test for AI
The Saab 96 is actually an excellent car for exposing the weaknesses of this technology. Its early three-cylinder two-stroke drivetrain, front-wheel-drive layout and highly recognizable period details leave less room for generic interpretation than many conventional 1960s cars.
The Saab Car Museum places the arrival of the 95 and 96 in 1960 and the later V4 transition in 1966, while the more detailed SaabMuseum.com model history documents the original 841cc three-cylinder two-stroke configuration. These are not obscure details: they are central to what an early Saab 96 actually is.
Saab enthusiasts also have an unusually rich body of surviving cars, restoration knowledge and competition history against which synthetic images can be checked. Even Saab’s rally heritage survives in real machinery rather than merely archive photographs.
That makes the mistakes relatively easy to expose today. The more interesting question is what happens when the technology becomes better at maintaining the same grille, engine, body panels and mechanical layout from the first generated scene to the last.
Interesting to Watch, Provided We Know What We Are Watching
This fictional 1963 Saab 96 restoration is therefore worth watching, but not because another rare Saab has been rescued.
It provides a surprisingly effective demonstration of how quickly generative video is learning the grammar of automotive content. AI “understands” that an abandoned classic should look dirty and corroded, that dismantling comes before body repair, that primer precedes paint and that a restoration story needs a dramatic final transformation.
What it still struggles with is the part Saab enthusiasts are most likely to examine closely: the actual Saab.
With clear disclosure, that can itself become entertaining. Enthusiasts can identify mistakes, compare the generated car with genuine examples and watch the technology attempt to reconstruct a machine whose history and engineering are extremely well documented.
Without that context, however, a less experienced viewer can easily leave believing that a nonexistent rally car was genuinely discovered and restored.
And that may be the most significant change represented by videos like this: seeing a convincing restoration on YouTube is no longer, by itself, evidence that a restoration ever took place.
interesting To watch & the day is coming, where we won’t know the difference between what’s real and what’s AI. Read a couple books on AI and there’s some scary things down the road potentially and some fantastic developments and discoveries ahead. Like all new technology it’ll be how it’s used.
But I do I want that cherry picker!!!
I don’t know if they make a zero gravity cherry picker that automatically adjusts to weight. I know factories have them, but they’re usually stationary for repetitive assembly line work