The Baidu MEUX team explored a new digital human social experience through the Dudou APP. They combine large language models, hyper-realistic images and emotion perception technology to create a virtual social world full of temperature and vitality. Here, each digital human has a unique worldview and personality, and is able to communicate deeply with users. This article will introduce in detail the thinking behind this innovative design, the implementation of technology and the improvement of user experience, and discuss how to make virtual life truly enter reality and become a caring partner in people’s lives.
1. Preamble
With the continuous iterative upgrading of AI technology, digital humans have long gone beyond simple graphical interfaces or voice assistants. From the initial mechanical and cold text interaction to a deep experience that can resonate with users, from the same mechanical answer to an intelligent life form with a unique “soul”, AI is constantly breaking through the inherent boundaries of tool attributes. In this technological revolution, a core proposition has been repeatedly discussed: how to shape digital humans so that they “truly enter the hearts of users?”
The Dudou APP is an innovative exploration of this development trend. By integrating technologies such as large language model capabilities, hyper-realistic images, and lip drives, a digital human social world with “warmth, story, and vitality” has been carefully created. In this world, each digital human is endowed with an independent worldview, and the communication between users and digital humans is no longer limited to superficial questions and answers, but can truly feel the emotional connection across the virtual and reality. Digital humans not only have the ability to think and express, but also perceive users’ emotional changes through chat content and adjust their response methods. They will even take the initiative to ask in late-night conversations: “How are you doing today, is there anything you can share?” This “realistic” interactive experience proposes a new development direction for the field of digital human chat.
2. Explore the three major challenges faced by digital human chat interaction
1. From “functional needs” to “emotional companionship”
In traditional AI conversation products, users’ essential needs for “chat” have changed, from the “functional requirements” that pursue connection efficiency to the “emotional value” needs that focus on emotional communication. I hope to get real understanding and companionship in chatting with AI, not just the delivery of information.
How can product managers do a good job in B-end digitalization?
All walks of life have taken advantage of the ride-hail of digital transformation and achieved the rapid development of the industry. Since B-end products are products that provide services for enterprises, how should enterprises ride the digital ride?
View details >
2. The construction of realism and trust
In the process of AI digital human chat, how to cross the image “uncanny valley” and give AI a personalized image, voice, and natural and smooth expression ability like a human is a key issue that needs to be solved urgently. Only by making digital humans truly “come alive” can we establish a deep emotional connection with users, win their trust, and make users willing to immerse themselves in communication and interaction.
3. The contradictory dilemma of “emotional desire” and “social fear”
At present, some people are troubled by social fear and do not know how to effectively socialize or chat, so they are more inclined to interact online. However, existing social software is difficult to provide “burden-free deep companionship”, and traditional AI dialogue products have limited satisfaction at the emotional level, and they still feel lonely and confused in virtual socializing.
3. Design goals and strategies: build a realistic digital human innovation experience
1. Personalization shaping: let digital humans have a “specific soul” and “realistic skin”
“Specific Soul” – Create a unique character
Each digital human builds an independent worldview structure, covering background settings such as age and region, as well as basic character settings (such as age, occupation), personality traits (such as lively and cheerful, INTJ personality, etc.), variable traits (such as tone of voice), and hidden personality (triggered under specific conditions). Compared with traditional virtual characters, which often fall into the “labeling trap” (such as “tsundere girl” and “gentle uncle”), it effectively avoids the interaction fatigue caused by “a thousand people”, makes each digital person unique, and brings users a more life-like dialogue and communication experience.
“Realistic skin”: Let digital humans “look like real people” AI to create a super-realistic image
In order to create a super-realistic digital human, a large number of real-life images are pre-trained SD models and image detail control lora, generated and finely tuned in the later stage, creating a very realistic digital human image. These images, whether it is physical features or facial expressions, are highly restored to real humans, allowing users to get a strong sense of substitution visually, and by building a Comfyui multi-workflow, carefully produce character atlases to create an exclusive circle of friends for each digital human.
2. HOOK “addictive” model innovates digital human experience
The HOOK model, also known as the addiction model, is a concept proposed by Neil Eyal and Ryan Hoover, the authors of “Addiction”, which summarizes the process of users developing habits about products into four steps: Trigger, Action, Reward, and Investment. It is applied to products to give users a better experience.
Trigger: Innovation in behavior and interaction
Dating App Realistic Interaction Framework:
With a hyper-realistic image, a reasonable page framework is needed to better display character information and improve distribution. The main scene “Discovery” of the Dudou APP adopts the interaction mode of the Dating app to carry a super-realistic digital human, strengthen the display of character images, blur the boundary between real and virtual, and more naturally allow users to find the characters they are interested in, making the dialogue experience more face-to-face, and also enhancing the sense of interaction between users and digital human characters. In this form of interaction, users seem to be communicating with real friends, which enhances the fun and immersion of the interaction.
Make digital human distribution “more vivid” – achieve dynamic presentation:
It uses a variety of AI tools such as Tusheng Video to generate dynamic videos of digital human images. This innovative method abandons the cumbersome process of traditional video and video shooting, and can efficiently produce rich and diverse plot content, allowing digital humans to show more vivid and natural postures and expressions in dynamic displays;
Dehomogenization allows digital humans to distribute “more vivid” – enriching character plots:
Through the combination of multiple AI tools, generate theme scripts, storyboards, videos, and BGMs to create short videos of personal character plots. In the process of choosing a character, users can not only have an in-depth understanding of the character’s background story, but also experience the fun of watching short videos, which further enhances the vividness of the character. Through this series of emotional and personalized design measures, AI is deeply endowed with personality, significantly strengthening the emotional bond between users and agents, thereby improving user retention.
【Action】Create an immersive multi-AI dialogue framework
Innovative chat form – improve the realism of interaction:
Using video material training technology, the whole body of the digital human in the chat state is realized, and the lip opening and closing and voice content are matched to the millimeter level, and natural body language is automatically triggered in combination with user dialogue scenes. This innovation breaks through the limitations of the high cost of traditional offline model video shooting, bringing users a more realistic emotional chat experience compared to traditional digital humans.
Break down the barriers to human networking – eliminate communication concerns:
In the Dudou APP, users do not need to worry about “saying the wrong thing” or “being evaluated”, and the simulated digital human has an image and voice that suits the user’s preferences, always listening patiently and responding positively. This communication environment allows users to let go of their psychological burden, express themselves to their heart’s content, and truly realize that they can chat, can chat, and want to chat, effectively breaking down many barriers in real social interaction.
Group chat interaction form – rich interactive scenes:
In order to meet users’ needs for multi-role and multi-scene interaction, the APP introduces a group chat function. Users can interact with multiple digital human characters in the same group, which greatly improves the playability of the product and lowers the threshold for use. In terms of dialogue framework design, continuing the immersive dialogue experience, users can easily switch roles by clicking on the screen through the group chat effect of background character switching, which brings a strong sense of presence to users while ensuring the smoothness of interaction. In addition, the character avatar is displayed in the sidebar, and a glare animation effect is added to indicate the character status, effectively improving the visibility of the conversation progress and enhancing the user’s sense of control over the chat process.
Hosted chat formats—lower the entry barrier:
In order to further improve the user experience, Dudou APP innovatively introduces the “hosting” function. In the chat scenario, users can turn on the AI chat mode, freeing their hands and letting AI input content for them. This feature not only reduces the user’s input cost, but also provides users with more diverse chat ideas and topics, making the chat process easier and more convenient.
【Reward】Multi-dimensional emotional feedback incentives
Intimacy System – Unlock more interactive content:
The increase in the amount of time users interact with digital humans can unlock a higher level of intimacy, and the heart is used as a super symbol in visual design, precisely because it has emotional uniqueness, recognition and expansion.
Character atlas – create a more three-dimensional sense of realism:
Build a Comfyui multi-workflow, carefully produce character atlases, and create an exclusive circle of friends (exclusive photo atlas) for each digital person to create a more three-dimensional realistic character.
【 Investment】Encourage users to continuously invest in improving product activity
Establish a commercialization framework system for the Doudou APP, obtain “Dudou” (tokens) through daily tasks, which is used to unlock new characters outside the daily restricted characters, continue the super symbol heart to guide users to click on the pop-up window, and encourage users to continuously unlock the atlas through dialogue through visual expression, improve the overall product activity and stickiness, and increase the possibility of users using the product again.
4. Emotional design improves user retention
Multi-scene companionship:
Provide all-round care: Companionship is an important embodiment of the freshness of digital humans. The digital humans in the Dudou APP can flexibly switch identities in various scenarios such as chat decompression, language learning, workplace simulation, and psychological counseling according to user needs, becoming an “all-round partner” online 24 hours a day. Unlike the traditional model of passively waiting for users to initiate conversations, digital humans will actively look for topics, keenly sense user emotions and adjust responses, so that users can truly feel that companionship is everywhere. This active companionship provides users with higher emotional value, improves user activity, and fully proves that emotional design is an effective way to enhance the emotional connection between users and agents.
Emotional Health Protection:
Create a good environment: In order to ensure the emotional health of users in virtual social networking, the APP has set up multiple risk control systems, which monitor the input layer, processing layer, output layer and other links, and trigger the guidance mechanism. Through these measures, a positive, healthy and safe virtual companionship environment is built, so that users can communicate with confidence and worry-free.
5. Conclusion
Our design goal is not to create a simple “perfect tool”, but to explore the possibilities of AI digital human chat and socialization, creating close friends, think tanks and confidants to accompany you. Through interaction with digital humans, users can not only obtain the value of practical tool attributes, but also reap the emotional comfort brought by companionship. When technology can give the code a temperature and allow AI to truly understand the subtleties of human emotions, such as “whether to silence or hug when sad”, we need to re-examine and think: what exactly is the definition of new companionship in the era of AI?
In the world of Dudou APP, every conversation is a journey of surprises, and every interaction creates a unique story. Whether the user needs a listen from a friend, a mentor’s guidance, or a virtual life of his own, there is always a digital person waiting to meet the user and greet softly: “How are you doing today, I’ve been waiting for you.” ”
