Could AI Turn Your Holiday Videos Into a Film? New Research Tests the Possibility

Returning from a holiday with hundreds of video clips is easy, but transforming those recordings into a film that someone might actually enjoy watching can require hours of reviewing, selecting and arranging footage. The problem is familiar to anyone whose phone contains recordings of beautiful scenery, memorable meals and interesting places that have never been edited into anything more substantial.

Researchers have developed an artificial intelligence system called Clips2Story that attempts to address this challenge by analysing collections of unordered video clips and identifying how they could be arranged into a coherent narrative. Released on 4 November 2026, the research explores whether multimodal AI and large language models can assist with organising existing footage rather than simply generating new scenes from written instructions.

The potential applications extend beyond holiday recordings to independent filmmaking, online entertainment and documentary production, although the technology also raises important questions about editorial judgement. Creating a sequence that appears convincing is not necessarily the same as representing events accurately, particularly when footage was recorded in different locations or without an intended narrative.

How does Clips2Story turn separate recordings into a story?

Clips2Story was developed by Junnuo Wang, Weihan Xu and Hao-Wen Dong as a training-free approach to video storyboarding and editing. Rather than training an entirely new model specifically for the task, the researchers combined existing multimodal analysis, retrieval techniques and language-model capabilities to examine video material and propose a narrative.

The process begins when a user provides a collection of clips alongside a keyword describing the subject they want the finished video to explore. The system analyses the recordings and generates textual descriptions of their contents, allowing relevant material to be identified without requiring someone to watch and categorise every clip manually.

Those descriptions are subsequently processed by a large language model, which selects footage, proposes an order and develops connecting narration where appropriate. The resulting storyboard can be edited before the system assembles the selected material and accompanying visuals into an output video, giving the user an opportunity to review the proposed structure.

This approach differs from conventional video summarisation because it does not depend on the original recordings forming one continuous sequence. Instead, the system attempts to construct a narrative from clips that may have been recorded separately, making the problem closer to organising a collection of material than shortening an existing film.

To evaluate the approach, the researchers created a dataset using 100 long-form videos covering five genres, dividing the original recordings into shorter segments and shuffling them to remove their chronological order. The system was then assessed on whether it could retrieve relevant material and produce a coherent sequence based on the supplied subject.

Human evaluators preferred Clips2Story over a comparison approach on measures including narrative coherence, relevance to the requested keyword and overall realism. These findings suggest that the proposed method performed well within the research conditions, although they do not demonstrate that it can independently produce professional-quality documentaries or reliably handle every type of unstructured footage.

The design of the evaluation is particularly important because the test material originated from videos that had already been assembled into longer productions. Shuffling clips creates a useful technical challenge, but it does not entirely reproduce the difficulties of working with unrelated recordings captured across several days by different people.

The study also does not establish a specific reduction in editing time for professional users or demonstrate that the system is available as a finished consumer product. Further testing would be needed to determine how its performance translates into everyday editing tasks and whether creators find the proposed storyboards genuinely useful.

Could this influence the future of filmmaking?

The potential audience for easier video production is substantial, although the popularity of watching online video should not be confused with evidence of demand for a particular editing tool. Ofcom's Online Nation 2025 report found that YouTube reached 94% of UK online adults in May 2025, equivalent to approximately 46.3 million people, while average daily viewing reached 51 minutes, compared with 47 minutes the previous year.

Those figures demonstrate how important online video has become within British media consumption, creating an environment in which creators compete for attention through increasingly sophisticated content. They do not reveal how many viewers produce their own videos, but they provide context for the growing importance of tools that assist with video production.

For someone recording a weekend in Edinburgh, a system based on Clips2Story could potentially identify footage of historic buildings, museums, streets and transport before proposing a narrative about the city's heritage. Instead of beginning with an empty editing timeline, the creator would have a suggested structure that could be reviewed, rearranged and developed.

Independent filmmakers might find similar assistance useful when working with interviews, location footage and other material gathered across multiple recording sessions. Identifying relevant clips can consume considerable time, particularly when a production has limited resources and no dedicated team responsible for organising footage.

However, documentary and journalistic work introduces a more demanding requirement because the arrangement of footage can influence how audiences understand real events. Two clips may appear to belong together visually while having been recorded at different times, and placing them in a particular order could create a misleading impression.

This distinction between narrative coherence and factual accuracy is fundamental to evaluating AI-assisted editing. A system may produce a sequence that viewers find logical and engaging without independently establishing whether the relationship between scenes is historically or factually correct.

The research does not demonstrate that Clips2Story can verify chronology, resolve conflicting accounts or identify when an apparently effective sequence misrepresents events. Human editorial oversight would therefore remain essential in factual productions, particularly where accuracy and context are central to the audience's understanding.

Creative judgement presents another challenge because experienced editors do more than identify relevant footage. They make decisions about emotional impact, visual contrast, sound, silence and the duration of individual scenes, sometimes choosing unexpected moments precisely because they interrupt an otherwise predictable narrative.

An automated system designed to maximise coherence may not recognise the artistic value of ambiguity or an unusual transition. This does not make the technology ineffective, but it suggests that its most useful role may involve assisting with preparation and organisation rather than independently determining the final creative direction.

The research remains a preprint, meaning its findings should be considered as an early technical contribution rather than an independently established standard for professional editing. Studies involving working editors, genuine production deadlines and diverse collections of footage would provide stronger evidence about practical benefits.

Nevertheless, Clips2Story demonstrates an interesting alternative to the current emphasis on generating entirely artificial video. By concentrating on footage that people have already recorded, it explores how AI might help creators make better use of existing material while preserving opportunities for human review and revision.

For ordinary users, that could eventually mean spending less time searching through forgotten recordings and more time producing videos worth sharing. For professional creators, the more significant possibility is technology that reduces repetitive organisational work without taking responsibility for the editorial decisions that make a story accurate, distinctive and engaging.

Research and sources: Clips2Story research (2026) | Ofcom Online Nation 2025