Introducing EMO: Alibaba AI Powers Animated Singing Portraits

Alibaba innovative venture into the realm of artificial intelligence (AI) has led to the creation of EMO: Emote Portrait Alive. This cutting-edge technology harnesses an audio2video diffusion model to craft realistic portrait videos, setting a new standard for the generation of talking head videos with unparalleled expressive accuracy. EMO represents a significant leap forward, overcoming the constraints of traditional methods by capturing nuanced facial expressions directly from audio cues.

Alibaba EMO AI

The introduction of EMO by Alibaba Group’s Institute for Intelligent Computing marks a pivotal moment in the evolution of image and video generation technologies. Utilizing advanced diffusion models and neural network architectures, EMO advances the capabilities of talking head video generation, offering a level of realism and expressiveness previously unattainable.

The quest for creating lifelike and expressive talking head videos has been a longstanding challenge within the realms of computer graphics and AI. Traditional methods often fall short, unable to fully encapsulate the breadth of human expression or produce natural and nuanced facial movements. This prompted researchers at Alibaba Group to develop a solution capable of translating audio cues into realistic facial expressions accurately.


EMO operates through a sophisticated two-stage framework that merges audio and visual data to produce expressive portrait videos. The process begins with Frames Encoding, where ReferenceNet extracts essential features from a reference image and motion frames, setting the stage for the ensuing diffusion process. This process involves a pretrained audio encoder for audio embeddings, integrating facial region masks with multi-frame noise to guide facial imagery creation. The Backbone Network, incorporating Reference-Attention and Audio-Attention mechanisms, ensures the preservation of the character’s identity and the modulation of their movements. Temporal Modules further refine the video by adjusting motion velocity, allowing EMO to create vocal avatar videos that feature expressive facial expressions and head poses for any duration based on the audio input.

Beyond merely generating talking head videos, EMO introduces the innovative concept of vocal avatar generation. With just a single character image and an audio input, EMO can produce vocal avatar videos that showcase expressive facial expressions and head movements. Whether it’s replicating the performance of famous songs or delivering lines in various languages, EMO demonstrates remarkable accuracy and expressiveness. This technology not only supports multilingual and multicultural expressions but also excels in capturing fast-paced rhythms and conveying expressive movements synchronized with the audio. This opens up new possibilities for engaging content creation, such as music videos or performances that require detailed synchronization between music and visual elements.

EMO’s capabilities extend beyond singing avatars. It can animate spoken audio in numerous languages, bringing to life portraits of historical figures, artwork, and even AI-generated characters. This versatility allows for conversations with iconic figures or cross-actor performances, offering new creative avenues for character portrayal across different media and cultural contexts.

The EMO framework signifies a significant advance in portrait video generation, eschewing the need for intermediate 3D models or facial landmarks and ensuring smooth frame transitions and consistent identity preservation. The technology is underpinned by a vast, diverse audio-video dataset, facilitating the training of the EMO model to capture a wide range of human expressions and vocal styles.

Despite its remarkable achievements, EMO is not without limitations. The quality of the input audio and reference image plays a crucial role, and there’s room for improvement in audio-visual synchronization and emotion recognition to enhance the realism and nuance of the generated videos. As technology progresses, further advancements in these areas are anticipated, pushing the boundaries of what’s possible in AI-driven video generation.

In summary, EMO: Emote Portrait Alive by Alibaba stands as a monumental development in the generation of expressive portrait videos, combining advanced AI techniques to achieve unprecedented realism and accuracy. As this technology evolves, it promises to expand the horizons of digital communication, entertainment, and artistic expression, redefining our interaction with digital avatars and the portrayal of characters across various languages and cultural contexts.

author avatar
Staff Report

Keep Up to Date with the Most Important News

By pressing the Subscribe button, you confirm that you have read and are agreeing to our Privacy Policy and Terms of Use