Custom video editing at the heart of Selendis’s Archon model

Developed by Selendis with Steve Aoki, Archon turns raw footage into video edits and adapts its choices to creators’ preferences.

The choice of a shot, the duration of a sequence, or the precise timing of a cut are as much technical decisions as they are personal preferences. Two editors can work on the same footage and produce very different videos. It is this subjective dimension that Selendis seeks to integrate into Archon, an artificial intelligence model designed to turn raw footage into customized edits while progressively learning each creator’s habits.

Developed by a lab that includes DJ and producer Steve Aoki, the system is presented as capable of producing a first cut from raw sequences without requiring task-specific training beforehand for each new job. The company claims processing in just a few minutes, within a collaborative environment where several people can work on the same project. The November 3, 2026 announcement is accompanied by the opening of a waitlist on the Selendis website. Learning the editor’s taste rather than reproducing a generic style

Archon’s announced distinction lies in the way creative preferences are taken into account. Rather than systematically asking the user to describe their style through text instructions, Selendis intends to leverage the decisions made during previous edits.

Three sources of information feed this personalization: the actions performed in the collaborative editor, the catalog of videos already produced by the creator, and the audiovisual references they provide. The combination of these elements is meant to help the system identify certain constants in shot selection, their arrangement, and the pacing of the sequences.

This approach seeks to bring automated editing closer to the functioning of a long-term creative collaboration. An editor who works regularly with a brand or filmmaker gradually learns their preferences, sometimes without them being explicitly stated. The tool aims to reproduce part of this adaptation by retaining the choices made across projects.

Selendis says that each edited video helps improve the next results for the creator or their team. However, the technical details of this learning process are not explained. The company does not specify whether preferences are stored in a user profile, whether they alter the model’s parameters, or whether they are reintroduced as context in subsequent projects. An initial edit produced directly from the raw footage

The other highlighted feature concerns the ability to produce an edit without going through a series of detailed instructions. The term zero-shot, used in the announcement, here refers to generating an initial result from the provided content, without additional training examples specific to the requested task.

The presentation video notably shows an editing interface with a media library, a preview of the result, and a timeline organized into multiple tracks. Vertical sequences intended for social networks appear there, with shots taken from performances and events.

The system thus stands apart from tools that only automate isolated tasks, such as subtitling, cropping, or silence removal. The stated ambition is to handle part of the editorial decisions needed to build a complete video from a set of takes.

The demonstration does not, however, make it possible to determine precisely which operations were carried out without human intervention. No public protocol yet documents the quality of the edits, processing times depending on the volume of footage, or the number of corrections needed to obtain a usable version.

The designation of foundation model, used by Selendis, is also not accompanied by a description of the architecture, the training corpus, or the system’s generalization capabilities. Steve Aoki associated with an approach drawn from music production

Steve Aoki’s involvement is one of the project’s distinguishing elements. In the video accompanying the launch, the artist appears alongside Gavin Wong, both presented as co-founders. The DJ notably mentions the significant costs associated with content production and the need to speed up its preparation for social platforms.

The demonstration draws on sequences from his musical activities, including travel, events, and stage performances. This context illustrates a use case in which teams have many hours of footage but must regularly produce short formats suited to different distribution channels.

The involvement of an artist whose work requires continuous audiovisual production gives the project a concrete testbed. It does not, by itself, constitute technical validation of the claimed performance.

Editing also remains a matter of selection and interpretation. Identifying relevant shots, building a progression, or preserving an artistic intent does not rely solely on repeating prior decisions. Preferences may evolve depending on the project, its audience, or the distribution context. Archon’s ability to handle these variations remains to be documented. Still limited availability

For now, access to Selendis goes through a waitlist. The company offers early registration and announces three months of free use for selected people as part of its social media launch.

No full pricing chart, storage limit, or specifications regarding accepted video formats are published on the site. The available information also does not further specify the import options for existing catalogs, the confidentiality of transmitted content, or export possibilities to other editing software.

Selendis says it is part of the NVIDIA Inception program, intended to support young technology companies. However, this does not reveal the infrastructure used for Archon or the resources required for it to operate.

The platform is currently presented as a collaborative video editor powered by a personalization model. Access remains restricted to the first users admitted from the waitlist, with no public general launch date.