news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

the price of 3d digital humans has dropped from tens of thousands of yuan in the past to 199 yuan now.

text｜xu xin

edited by ren xiaoyu

you may not know that you are interacting with a digital person.

many people's impression of digital people is still at the stage of image display. on offline large screens, a real person image introduces products or interacts with the public, which is not a good experience, the performance is also a bit stiff, and the production cost is high.

however, over the past year or so, the emergence of large models has brought more possibilities to the digital human industry. some manufacturers have listed digital humans as the forefront of large model applications, and domestic digital human pioneers are also accelerating product iterations on a quarterly basis.

the technology of digital humans is constantly being upgraded.the expressiveness of digital humans in terms of portrait, voice and language is gradually improving.on the other hand, digital humansproduction costs and thresholds are greatly reduced, and efficiency is rapidly improved。

taking xiling digital human of baidu smart cloud as an example, shen dou, executive vice president of baidu group and president of baidu smart cloud business group, said that based on the support of two industry-leading technologies, baige and qianfan, "customers only need to spend one percent of the cost in the past to easily create their own digital human works in minutes."

technological progress brings inclusive dividends, and more enterprise-level scenarios are unlocking digital humans. shen dou introduced that at present, the xiling digital human platform already has a variety of digital humans with mature images and rich types.covering major industry scenarios such as culture and tourism, e-commerce, and finance. more and more companies are using digital humans to improve efficiency and increase revenue, and gain business value.

as enterprises apply digital humans more deeply, the demand patterns of different enterprises for digital humans are also stratified, and manufacturers are also updating their business models and strategies around digital human products. the xiling digital human team of baidu smart cloud said that this year they will focus onpromoting the use of digital humans on public cloud platforms, and the signing of projects between standardized saas products and industry customers is expected to form a virtuous circle.

01

new products are released quarterly to continuously lower the threshold for digital human implementation

digital humans are becoming one of the hottest scenarios for the implementation of large models.

since the second half of last year, manufacturers have been accelerating their exploration of how to implement large models in the industry, and digital humans have been regarded as pioneering scenarios by many manufacturers. this year, when many platform manufacturers showcased cases combining large models with vertical industries, digital humans were listed as key applications.

baidu, which was the first company in china to release a large model and explore industry applications, has also invested a lot of resources in this field. at the baidu cloud intelligence conference held today, xiling digital human ushered in the 4.0 upgrade, which is also the third release and update of this product this year.

this update focuses onimproved capabilities, lowered production barriers, and optimized efficiency and coststhree dimensions.

in terms of capability upgrade, shen dou introduced that xiling 4.0 solves the problem of stiff movements of traditional 2d digital humans, and can realize the characters in differenthigh consistency in angle, shape, and expression, even the facial micro-expressions are very realistic and natural.

to this end, the xiling team carried out special development. zhang yuxiang, general manager of baidu intelligent cloud digital human product department, introduced that they have created a uniquelip-matching algorithm, so that the digital human's lip shape is more in line with the content when speaking. in order to improve the naturalness of dialogue interaction, they introduced a listening state design andthe front small model intelligently inserts guide words in the gaps of conversation, significantly improving the immediacy and interactivity of replies.

in terms of production threshold, xiling digital human 4.0 also further simplifies the requirements for user input materials. taking 3d digital human as an example, users nowjust enter a simple text description to quickly generate3d digital human images and videos with different makeup and industry characteristics.

in terms of production efficiency and cost optimization, the time required to generate digital humans has now been reduced to minutes. the industry has observed that the june update,the time it takes to generate xiling 2d digital humans has been reduced from 3 to 7 days to hours.

at the cost level, the price of digital humans for enterprises is also falling. when xiling digital human was updated in june, the price threshold of 3d hyper-realistic digital humans was reduced from 100,000 to 10,000 yuan. after today's upgrade, the price of 3d hyper-realistic digital humans has dropped to 10,000 yuan.the price continued to drop from 10,000 yuan to 199 yuan, reaching the lowest price in the industry. this is undoubtedly another bombshell for the digital human industry.

the industry has observed that this year, xiling digital human has been rapidly iterating and updating on a quarterly basis, focusing on several major problems that have long plagued the digital human industry. this version 4.0 update also continues baidu xiling digital human“high availability, high cost performance" is intended to further reduce the threshold and cost of using digital people.

in fact, this is also the common direction of the industry. idc told digital intelligence frontier that the popularization of ai digital human technology is becoming the focus of market attention.how to lower the usage and cost threshold of products through relevant technologies will become one of the key factors in future competition。

as an old player in the domestic digital human field, the xiling team of baidu smart cloud believes that digital humans should be able to replace real people and surpass them in some areas. zhang yuxiang, general manager of the digital human product department of baidu smart cloud, explained that only when digital humans outperform real people in performance can they unlock more industry scenarios and be used more widely.

he introduced that thanks to baidu's continuous deep development in the field of digital humans over the past six years, it has accumulated massive, high-quality data. at present, xiling digital humans can do things that real people cannot do in multiple dimensions such as portrait, voice and language ability.

for example, taking portraits as an example, real people rely on the lighting and makeup in the current environment, but based on the xiling digital human platform, when restoring real people,you can improve your image in the video, similar to the face-slimming function of live broadcast。

in terms of sound, ordinary people have a lot of pauses and lags in their daily spoken language, but digital humans canbe fluent and natural, with a steady tone and a sense of rhythmin terms of language expression ability, the multi-language switching ability of digital humans has also broken through the ability limitations of real people and can easily unlock multiple languages.

"2d digital humans can break the limitations of real people in terms of time, space and abilities, and can replace real people on screen. their performance surpasses that of real people in all aspects. this is the direction that we will all work towards together in the field of 2d digital humans," said zhang yuxiang.

02

the way enterprises use digital humans is changing

after more than a year of development, large model technology has empowered digital humans and brought new possibilities to the digital human market.

first, as digital human capabilities are upgraded,digital humans are unlocking more application scenarios。

"in addition to portraits and voices, the language capabilities that big models bring to digital humans have opened up more possibilities for us." the xiling team led by zhang yuxiang has more than five years of experience in the digital human field and has observed the application of thousands of corporate clients. he has seen that, with the empowerment of big models, digital humans have been widely used in many scenarios that were previously unimaginable.

typical islive broadcast scene, with the support of large model capabilities, you cangenerate a live broadcast script for the digital human, which can explain the product content in real time, and can also complete real-time questions and answers about product information, 24/7, stable and efficient. for example, with the support of multilingual capabilities, digital humans can flexibly switch languages ​​and publish a set of content to global media and customers, adding convenience to cross-border e-commerce and foreign trade businesses. "after a breakthrough in one capability point, it can open up more possibilities," said zhang yuxiang.

secondly, as technology advances and barriers to entry are lowered, digital humans are entering more industries.different companies have different needs for digital human capabilities。

"digital human technology has different application requirements in various industries," said zhang yuxiang. he and his team received a wide variety of customer feedback. for example, a person in the media industry was interested in whether the digital human platform couldclone your own voice with high definition, and output high definition videoin the educational scenario, can the digital human teacher be based on the student's learning ability and previous knowledge mastery?give different answers and explanations。

this also calls for digital human technology service providers to systematically sort out various capabilities and decouple different capability modules to adapt to and meet the diverse needs of the market. some pioneering companies also follow the market and deposit digital human capabilities into open platforms to achieve flexible component-based calls.

take baidu smart cloud's xiling digital human as an example. in july this year, xiling digital human's open platform was launched.split standardized capabilities into flexible components for industry users to callfor example, customized cloning of portraits, customized cloning of voices, interactive dialogue scenarios, rendering capabilities on various terminals, production and live broadcast of digital human videos, and other capabilities can all be easily called upon.

the open platform capabilities have also been warmly welcomed by the market.”after the launch, hundreds of customers have tried it out every week, and the application scenarios for digital humans have far exceeded expectations."zhang yuxiang believes that this reflects the diverse and booming demand in the digital human enterprise application market, and also means that the application of digital humans in enterprises is gradually deepening.

as a result, the service model of the digital human market has also evolved and updated. a few years ago, the digital human services in the industry were mainly large-scale customized projects. with the improvement of the scale replication capability of digital human technology,platforms begin to transform digital human capabilities into standardized saas productsas more and more industries use it, the needs of enterprises are differentiated, and more flexible component-based calling methods are added.

andthe three forms of baidu xiling digital human services currently provided to industry customers are the out-of-the-box saas platform, the efficient and easy-to-integrate component platform, and industry-level solutions tailored for leading customers.

"component-based cooperation is more suitable for industry users, who can integrate digital human capabilities into their own systems and applications through these easily integrated components. currently, the mainstream calling mode in the industry is mainly component-based, and the scenarios that saas can cover are more general scenarios." zhang yuxiang introduced the difference.

he believes that digital humans are the performance layer, and to be able to use them well in an industry they need to be combined with the industry's vertical fields, involving industry know-how and in-depth scenario knowledge.

for example, in an educational scenario, when a teacher is teaching online, there may be a digital human entrance. if there is something you don’t understand, you can just click on it, and the digital human teacher can communicate and interact one-on-one based on the knowledge points and the students’ situation.

to achieve this, we need to call on the capabilities of the digital human open platform, build it together with partners in the education industry, and connect the digital human capabilities with the company's existing course system and student management system. this again involves industrial division of labor, and we need to build it together with partners in the education industry to truly use the digital human capabilities in the scene.

03

digital people are becoming digital employees in all industries

as digital human capabilities continue to upgrade and application thresholds continue to decrease, the way companies obtain digital human services has become more flexible, and baidu smart cloud's xiling digital human is also accelerating its application in more scenarios.

"the original digital human project cycle was very long, but now it only takes one or two days from trial to actual operation. if the company has stronger programming skills, it can get started and see results in half a day." zhang yuxiang observed that many companies canxiling makes it easier to see the effects and business value of digital humans。

idc china research manager cheng yin also told digital intelligence frontline that at present, enterprises are using ai digital peoplemainly for the purpose of innovating business and helping enterprises improve efficiency and increase revenuethe most obvious areas of digital human value are in live streaming, digital human customer service, virtual anchors and other scenarios, where roi is easier to calculate. the value brought by other scenarios is difficult to calculate, which is one of the challenges facing the implementation of the technology.

zhang yuxiang believes that we should look at the value of digital people more comprehensively.roi indicators are more inclined to be used for the measurement of digital human effect indicators in some delivery and advertising marketing scenarios.the key lies in whether this technology is actually used in the enterprise.

currently, inscenarios such as delivery and advertising marketingin the video, baidu smart cloud's xiling digital human is playing the role of a shopping guide, enhancing the attractiveness and interactivity of the content, significantly accelerating the creation process of marketing videos and reducing costs.the production cycle of 2.5 days was reduced to 0.5 days, which has won valuable market opportunities for businesses. at the same time, digital humans can greatly reduce the cost of filming. in first-tier cities, the daily cost of real actors is at least 1,500 to 2,000 yuan.

the materials uploaded by users can also be used to batch generate multiple videos, so the cost per video becomes lower.the production cost of advertising materials has been reduced to about one-third.”

in addition to advertising and marketing scenarios,digital employees played by digital humans are also widely used in the financial industry.. digital intelligence frontline learned that many leading banks are using baidu xiling digital humansettled in the business hall, efficiently taking over many tasks that traditionally rely on offline branch salespersons, greatly improving business processing efficiency and customer experience. offline outlets no longer need to be equipped with more salespersons. considering the number of outlets across the country, the amount of cost savings is very considerable.

"the digital employee operation platform driven by digital human technology in the banking scenario can truly provide digital employee operation capabilities and greatly liberate employees' energy." zhang yuxiang said that digital humans have been truly used in this scenario.xiling digital’s coverage rate in 18 leading banks has reached 50%"the product repurchase rate is high. many customers have reached the third, fourth or even fifth stage, and we iterate the product every year."

there are also some scenarios that are not suitable for roi measurement, a typical example being the cultural tourism sector. some regions are using local historical celebrities created by baidu digital humans to reproduce them in the form of generated ips, interact with tourists at cultural and tourism attractions, and play the role of electronic guides. however, the industry believes that the value it generates should not be measured simply in numbers. it can bring a richer tourism experience and allow historical culture to be passed on in a more public and interactive way. in the future, as the capabilities of digital humans continue to evolve, the role it plays will be further presented and released.

it can be said that as digital humans play the role of digital employees in more and more enterprise-level scenarios, more and more scenarios can clearly calculate roi and business value, and the digital human application market is gradually opening up. idc predicts thatby 2026, the scale of china's ai digital human market will reach 10.24 billion yuan。

as the market moves from its infancy to maturity, product teams like baidu smart cloud xiling have begun to develop a systematic approach.

zhang yuxiang introduced that previously, the revenue from public cloud products accounted for a small proportion of xiling digital's overall revenue.the majority of projects are from cooperation with leading government and enterprise companies. now, they are beginning to pay more attention to revenue growth on public clouds.。

"previously, our capabilities were all concentrated in projects, but this year we will fully commercialize them. we will achieve leading capabilities and technical levels, and then follow up with functional scenario coverage of application products," said zhang yuxiang.

here, different product models are expected to form a virtuous linkage - the income accumulated from past projects is supporting the development of standardized public cloud products, and the capabilities brought by the development of public cloud can better promote the signing of projects.

04

how to become a pioneer in industry implementation

at present, baidu smart's yunxiling digital human, as a typical application scenario of large-scale model landing industry, has been applied in major industry scenarios such as culture and tourism, e-commerce, and finance, and its application breadth and depth are moving forward.baidu's large-scale model technology enters the industrial field in miniature。

over the past year, large models have been accelerating the transformation of the industry from technological change. data shows that from january to august this year, the number of domestic large model projects has reached five times the number of the whole year of 2023, and the amount of the bid has reached twice the whole year of last year. the role of the leading large model manufacturers is still very prominent.baidu ranked first in four key indicators: the number of projects won by large models, the amount of winning bids, the industries covered, and the number of central state-owned enterprises covered.。

the rapid advancement of industrial implementation is inseparable from the support of a new generation of infrastructure. focusing on the industrial implementation of large models, baidu smart cloud is forming a full-stack infrastructure foundation.

computing powerwith the huge training demand of large models, the required cluster size is getting larger and larger. how to achieve efficient and stable management of gpus to reduce the training and reasoning costs of large models has attracted much attention in the industry. at today's cloud intelligence conference, baidu smart cloud launched the baige 4.0 version upgrade. the upgraded baige, focusing on the computing power requirements of the entire journey of implementing large models, provides enterprises with "more, faster, more stable, and more economical" ai infrastructure in four aspects: cluster creation, development experiments, model training, and model reasoning.

the explosion of large model applications is inseparable from the convenient and efficient large model tool chain and application development platform. to meet the needs of enterprises in the large model industry, today,qianfan's large model platform has also been comprehensively upgraded in terms of model development layer, model service layer, and application development layer.

the upgraded qianfan 3.0 is further lowering the threshold for enterprise-level application development, while providing richer large and small models to cover more industry scenarios, and providing a more complete large model tool chain to help enterprises achieve one-stop large and small model development services.

at the baidu cloud intelligence conference held today, shen dou, executive vice president of baidu group and president of baidu intelligent cloud business group,xiling digital human, intelligent customer service "keyue" and wenxin quick code have undergone major upgrades, which is aimed at enterprises, is also the ai ​​product model room built by baidu based on infrastructure. he believes that only by personally walking the path that users have to take can we design products that understand users better.

the introduction of big models into thousands of industries is a huge systematic project. model manufacturers, application companies and model ecosystem service companies are continuously working on various aspects such as computing power infrastructure, algorithm training and optimization, industry scenarios, data preparation and governance, and on-site deployment to accelerate industrial applications.

the series of major product upgrades and updates at this cloud intelligence conference is undoubtedly an important footnote in this wave.