Tavus is moving beyond its original SaaS product and into developer infrastructure. The four-year-old generative AI startup has confirmed a fresh $18 million in funding and is opening its face and voice cloning technology so third parties can build it into their own applications.
The shift matters because Tavus is not only selling personalized video campaigns directly to companies. It is also trying to become a platform layer for AI video, where developers can add digital replicas, lip-syncing, dubbing, and campaign tools inside their own software.
What Tavus raised and who backed it
Tavus confirmed to TechCrunch that it raised $18 million in a Series A round led by Scale Venture Partners. The round follows earlier reports that the company had raised “about $18 million,” but the new confirmation adds the details around the financing.
Scale Venture Partners has previously backed Box, HubSpot, and DocuSign. Other investors in the Tavus round include Sequoia, which led Tavus’ $6.1 million seed round last year, alongside Y Combinator (YC) and HubSpot.
The funding arrives as generative AI video is drawing more attention. Text-based tools like ChatGPT and text-to-image systems such as DALL-E helped define the early wave of generative AI, but video is becoming a major focus for companies and startups alike.
OpenAI recently debuted Sora, a text-to-video model. Google has also been working on similar tooling for several years, and a number of startups have raised significant venture funding around different ideas for generative AI and video.
How Tavus makes personalized AI video
Tavus helps companies create digital “replicas” of individuals through voice and face cloning. Those replicas can then be used to generate personalized videos for sales, marketing, product onboarding, and customer engagement.
In a sales or marketing workflow, a company might use Tavus to send personalized videos to prospects at scale. In a product workflow, a team might create individualized walkthrough videos for new customers. The key idea is that text-based prompts can generate new video output from a previously created replica.
Tavus can also connect with third-party systems such as Salesforce or Mailchimp. That makes it possible to automate parts of the process. For example, when a customer completes an online form asking for product information, the company could send a video instantly, with a sales rep addressing the customer by name and explaining next steps.
The company already has large customers. Tavus co-founder and CEO Hassaan Raza said Salesforce and Facebook’s parent Meta are using the platform to upsell to their B2B customers through personalized demo videos.
From SaaS app to developer APIs
Until now, Tavus has mainly been delivered through a SaaS app. Customers use that product to create AI video templates and define which parts of a video should change for each viewer.
The onboarding process has required an individual, such as a CEO or sales executive, to record a 15-minute video using a script provided by Tavus. Tavus then uses that recording to train the AI. After that, the customer can use a web editor to choose variables such as location, executive name, company, or product.
Those variables can be connected to a CRM system, allowing companies to tailor videos for specific customer segments. Tavus says companies can create hundreds of replicas with different people, as well as different backgrounds for different target markets. The editor can also generate different scripts for different use cases without requiring the original video to be recorded again.
That SaaS product is staying, but Tavus is now adding a developer platform. The first piece is the “replica API,” which allows third parties to create “photo-realistic” digital replicas with text-to-video generation.
What the Phoenix model changes
The replica API uses a proprietary Tavus model called “Phoenix.” According to the source article, Phoenix is based on a deep learning method called neural radiance field (NeRF), which can generate a 3D construct of a person from 2D images in just a couple of minutes.
Raza described the change as a major reduction in training requirements. “It essentially allows you to create entire videos with just two minutes of training data, which is a big leap forward from how we were previously doing the personalization at scale,” Raza told TechCrunch. “And so now all you have to do is record two minutes of training data, and it’ll create a full replica of you. And once you have replica, you can make as many videos as you want — from one, two, or a thousand scripts.”
The replica API captures facial motion across cheeks, nose, eyebrows, and lips. Raza said that matters because natural-looking speech involves more than lip movement. “Moving your entire face drives realism, naturalness and quality — when you talk, your face expresses emotion beyond your lips moving,” he explained.
Tavus is also developing additional APIs. Those include a lip-sync API, a dubbing API, and a video campaign API. The lip-sync API is expected to have a “lower entry cost” and is meant for cases where a “high degree of quality and realism is not necessary.”
The dubbing API uses the lip-sync model and adds multilanguage voice cloning. That means a user who speaks one language could send video campaigns in multiple languages using their own voice. The campaign API combines the replica API with tooling such as hosting, variable mapping, thumbnails, and analytics.
The deepfake problem Tavus must manage
Technology that can create face and voice replicas also raises obvious misuse concerns. The source article frames this as the deepfake dilemma: what prevents someone from uploading an existing video and creating a digital replica without permission?
Raza said Tavus uses checks to reduce that risk. When a user submits two minutes of training footage, they must also submit a specific verbal consent statement. Tavus then aligns that statement to the audio in the training footage to confirm a match.
“We run these checks automatically, and then do a human check for every replica that makes it through the automated checks to ensure safety,” Raza said.
That approach also applies to the API business, according to Raza. Even when developers integrate Tavus technology into other applications, Tavus says it will remain responsible for verification. “We run the same checks, and assume responsibility for verifications with [the] API as well,” he said.
The company’s broader pitch is that AI video replicas can “extend” a person’s own reality rather than simply create visuals from text. Raza said Tavus sees a future where people want digital replicas they control, with their gestures, traits, and personality captured more fully over time.
The new funding is intended to support that direction. Raza said the $18 million will help Tavus continue developing models like Phoenix, sustain growth, and keep hiring across machine learning and engineering teams for its developer and SaaS customers.