A start-up called Incantor is positioning itself at the intersection of intellectual property protection, generative AI, and content creation—a murky corner of the content coral reef which will in some way eventually disturb all types of weird and wonderful creatures inhabiting the media value chain.
“Current AI tools are flagrantly contemptuous,” begrudged Incantor CEO Lauren Oliver, over video call to Faultline this week.
With many of the mainstream genAI platforms from OpenAI, Google, and DeepSeek facing backlash for scraping copyrighted material from the web, Incantor claims to have built a proprietary “Light Fractal Model” trained only on licensed data, with built-in attribution tracking. The idea is to cultivate a creative marketplace, where every media asset contributed (voice, image, script, and eventually video) will be tracked, credited, and compensated by a 10% share of any resulting revenue streams.
Incorporated in January 2025, the 8-person Incantor is about as raw as start-ups come. The AI model, a result of Oliver’s own literary PTSD as a best-selling American novelist, has the potential to be successfully disruptive, but to reach that stage requires scale (as well as some “duct tape and wishes” required for many of the 50% of US tech start-ups to survive beyond year 5).
Oliver claims, with a grin, that Incantor is already at a stage where it is needing to scale—which could present teething problems for such a nascent operation, when trying to run before it can walk. However, the rise of lawsuits between Hollywood writers and genAI giants, described as an apocalyptic threat to creative livelihoods, is all the evidence needed to justify the bootstrapping of Incantor at personal expense.
Currently, Incantor is a text in, voice out model. We are shown a quick demo of a synthetic voice based on a real actor’s IP. The text-to-voice model plays the statement “Oh my god, what do you mean she’s dead?”—with the voice output able to be pulled around an emotion wheel to convey 16 different emotions (and growing). The result is an impressively expressive range from just two data sources.
This immediately reminded us of the NotebookLM experiment we ran with the Faultline podcast last year. Let’s just say it didn’t catch on, and was comically cheesy in parts, but this is music to the ears of Incantor with what the company has planned for an automated pipeline to transform media into adaptive podcasts.
Oliver is optimistic that Incantor’s proprietary AI-generated outputs, including a couple of trialed podcasts, are a cut above what typical transformer/diffusion models can produce.
Naturally, the conversation turns to video. We learn that Incantor has partnered with an unnamed animation studio to train a lightweight AI model that can animate characters without relying on traditional rigging. Instead, the model was trained directly on performance data, allowing it to generate animations autonomously, with only a day’s worth of manual refinement needed in the animation pipeline.
A separate AI-driven lip manipulation demo was also teased, but unfortunately an NDA blocks that door from opening, for now.
Where Incantor could fall over is if the attribution model fails. If content IP is leaked outside of the platform and manipulated, then there is no way to prove the provenance and therefore protection of the owned IP.
This is why Incantor is likely to follow more of a metadata-based model similar to that of the C2PA (Coalition for Content Provenance and Authenticity), which works fine in controlled environments, such as a TV studio or a trusted newswire, where media passes through known systems. But once content enters the open internet, metadata becomes a liability, and if your authenticity model collapses the second someone re-uploads a file, then your entire backbone collapses.
Oliver notes that Incantor is exploring tokenizing attribution, though the company co-founder dismisses going down the blockchain route—bemoaning the distributed ledger approach for not scaling. This contrasts with claims from blockchain advocates that the technology will scale once (if) the decentralized web 3.0 revolution takes off, rather than the centralized internet that standards bodies imagine.
Returning to the Light Fractal Model itself, these algorithms based on fractal mathematics—geometric shapes that form infinite patterns similar to neural activity in the human brain—claim to reduce data usage by between 10x and 100x.
That in turn returns power consumption reductions orders of magnitude under what traditional incumbent AI companies are generating, while at the same time claiming to achieve performance that is orders of magnitude over what these household LLMs are outputting. Oliver did not have any energy benchmarks to hand, but we have been promised a follow-up with hard supporting data.
If Incantor can live up to promises of beating the big boys over quality, price, and power consumption, while maintaining attribution and ownership of IP with royalties to boot, then we are onto a winner.
Does it sound too good to be true? Possibly. But that is primarily because the company cannot currently share a public customer case study.
To grasp the SaaS-based model of the creative-driven marketplace, Incantor provides the following example:
“Let’s say John and Jane Doe have both contributed voice timbers to create a Synthetic Voice #1.”
“That Synthetic Voiceover is then used by one of our creative partners to serve as the narrator for one of our channels, which begins to earn ad revenue in its second year.”
“We will create a pool of Contributing Sources, and their relative contributions to the finished IP, in which John and Jane Doe will be represented. When we begin earning Revenue, they will begin earning Royalties. Additionally, their contributions will be credited at the end of every piece of media that makes use of their data.”
Once Incantor achieves the kind of scale that its business model inherently relies on—with a larger pool of contributing source IP creating a higher chance of generating royalty-earning revenues—then we will reassess whether the AI-scorched Great Barrier Reef of media ownership is showing signs of life.