Soul to Launch Interactive AI Audio-Video Product Loxinity, Featuring Loxi Speech and Loxi Avatar

Dujia has learned exclusively that Soul will launch "Loxinity (Luoxi Wuxian)," a speech synthesis product built specifically for interactive scenarios and designed to feel highly human. Unlike the TTS products on the market aimed at dubbing and short videos, this model targets real-time interactive scenarios from the ground up and focuses on a strong sense of being a real person. It supports two modes—single-person dubbing and multi-person dialogue—comes with tools such as voice cloning and voice design, and also offers API access. Loxinity encompasses the speech synthesis model Loxi Speech and the video generation model Loxi Avatar.

At present, compared with traditional speech models that only pursue standard pronunciation, Loxinity places more emphasis on paralinguistic details such as pauses and emotional ups and downs in dialogue, which it uses to support continuous voice interaction in social scenarios.

The product's mission is: to make AI interaction more natural—using voice, vision, and multimodal technology to shorten the distance between people and intelligence. It is quite impressive, and it is not ruled out that this product will become an important attempt by Soul in AI voice scenarios beyond its social business.

Soul is packaging AI voice, real-time video calls, and digital humans together and bringing them to market

Dujia's hands-on test of Loxinity, an AI product under Soul, found that this product is based on the speech synthesis model Loxi Speech and the video generation model Loxi Avatar. Moreover, it is not just a single speech synthesis tool, but a set of integrated AI audio-video solutions for real-time interaction, with its core divided into three major sections: speech synthesis, real-time video calls, and digital humans.

The speech synthesis module supports single-person dubbing and multi-person dialogue, comes with voice cloning and voice design functions, and can also add paralinguistic, pinyin, and emotion markers. Its optimization focus is on dialogue scenarios, with emphasis on restoring the pauses and tonal ups and downs of real human speech. It also offers API access and adopts two billing models: credits and subscription packages.

The digital human and real-time video call section is another key focus of this product. Users can upload images to customize a digital human image, input a text script or upload audio, and generate a corresponding digital human video.

Overall, Loxinity avoids the crowded general large-model track and focuses specifically on interactive audio-video AI. Many similar digital human tools suffer from problems such as a strong read-aloud feel in the voice and disconnection between audio and video, while Loxinity chooses to natively connect speech and digital human driving, letting voice capabilities serve continuous dialogue.

Of course, uncertainties remain. For this type of real-time interactive product, the core bottlenecks lie in the stability of long conversations and end-to-end latency. After leaving Soul's internal social scenarios and being delivered externally to B-end customers, concurrency capacity, usage costs, and the naturalness of digital human micro-expressions all still need more implemented projects for verification. How much recognition this AI interaction product, which comes from a social platform, can ultimately gain in the market remains to be seen.

From in-app social interaction to external output: Soul bets on interactive AI

Young people's core demand for AI is not text Q&A, but immersive interaction that carries emotion and is close to real people. Even before Loxinity was implemented, Soul had already made multiple rounds of implementation attempts within the app.

In 2022, Soul launched the AI chatbot "AI Goudan," which, as a first-generation AI companion character, focused on everyday conversation and proactive emotional care, no longer limited to Q&A-style interaction. During the same period, it launched the NAWA virtual avatar engine, allowing users to create their own virtual avatars, put AI conversations into virtual avatar carriers, and build virtual social scenarios. In 2023, Soul officially released its self-developed emotion-oriented vertical large model SoulX, positioned as an emotion-perception interaction foundation. It no longer followed the trend of building general large models, but focused on strengthening AI's ability to understand emotions, maintain personas, and retain long-context memory.

Relying on the SoulX foundation, a series of AI social products were launched one after another. The platform launched AI painting and AI dimensional singers for users to create virtual avatars and AI voices; it launched an end-to-end full-duplex speech model and implemented an AI host in group chat parties, supporting real-time interjections and natural interruptions, replicating the conversational rhythm of real multi-person casual chat and breaking away from mechanical interaction. In-app scenarios such as Werewolf and virtual companions are all connected to this speech capability, enabling long-turn continuous voice conversations. These in-app scenarios have all provided settings for continuously iterating and polishing this "human-feel" interaction technology.

The general large-model track has long been crowded with giants, and Soul did not confront them head-on; instead, it chose the emotional interaction track. The massive real-human conversation samples accumulated over years of social business are its core moat for polishing paralinguistics and natural conversational rhythm.

But the challenges are equally clear: interaction capabilities verified as feasible in in-app scenarios, when delivered externally to B-end customers, concurrency load, real-time latency, and project costs are all hard thresholds. The commercialization potential of this set of AI capabilities still needs market implementation cases to test.#Soul#Soul launches AI product#Soul launches Loxinity