news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

company name: beijing yunke technology co., ltd.

financing round: angel+ round

product/service: aiuni - ai-generated 4d animation platform

founder: hu yating (chuangyebang star camp 25th)

year of birth: 2000

educational background: department of computer science, peking university

author: ma wenpei

editor: liu hengtao

image source: aiuni

hu yating, born in 2000, chose to start a business right after graduating from university. with her beautiful appearance and fashionable clothes, hu yating completely refreshed our imagination of the image of entrepreneurs.

hu yating, who graduated from the department of computer science at peking university, has worked as an algorithm engineer at google, alibaba, and bytedance. she also participated in the informatics olympiad and won the national gold medal and the best female athlete award. during her internship at a large company, she saw that internet traffic had reached its peak, but 3d was a content upgrade, so she decided to start a business after graduating from university.

aiuni, founded by hu yating, is an ip platform that generates 4d animations from ai. in june this year, aiuniai's 3d generation model unique3d was open sourced on platforms such as github and huggingface. it quickly topped the list of popular models on huggingface and was nominated as the "best image-generated 3d model", winning praise from many developers.

unique3d can generate high-fidelity and diversely textured 3d meshes from single-view images, which takes about 30 seconds on a 4090 graphics card. many netizens shared their generation results on social media and praised the model for reaching a very high level in fidelity, consistency, and efficiency. in just a few months, unique3d has reached a generation volume of millions.

this summer, hu yating joined the 25th star camp of chuangyebang - the star camp new ai star acceleration program.

after the acceleration, she will take the stage of 2024demo china on the 19th and 20th of this month, and present her thoughts on the products and commercial applications she has created to well-known investors, industry experts, and industrial partners in various fields in an advanced manner.

targeting the 3d market

as a key means of mapping the real world into the internet world, 3d technology has penetrated into various fields. from game development, film production to product design, e-commerce rendering, and architectural planning, the application scenarios of 3d technology almost cover the entire internet industry.

the cost of 3d modeling varies depending on the application scenario. the more sophisticated the model, the more complicated the production process, the longer the production cycle, and the higher the cost. the creation cost of each 3d model is at least several thousand yuan or even tens of thousands of yuan.

currently, the entertainment industries such as animation, film, and games are the main application areas of 3d modeling. in the game industry, with the continuous development of 3d engines, 3d games have gradually become the mainstream of the market, and the demand for 3d modeling is extremely high; in the animation and film industry, using 3d technology, the production of grand scenes only requires green screen technology and important actors, which reduces the personnel costs of the production party.

according to data released by toubao, in 2021, china's 3d modeling market reached 10.34 billion yuan, and it is expected that the market size will reach 19.57 billion yuan by 2026. the size of the global 3d animation market in 2021 is estimated to be around 164 billion us dollars, and the market is expected to grow at a compound annual growth rate (cagr) of 11.5% to reach around 310 billion us dollars by 2026.

the aiuni team is targeting this market.

in addition to hu yating, the company's technical partner wu kailu was someone hu yating met when she participated in the informatics olympiad. he graduated from tsinghua university's yao class and conducted research on 3d generation and nerf at the institute of cross-disciplinary information sciences of tsinghua university. he published many papers during his undergraduate studies and innovatively proposed fsd (text-to-3d flow fractional distillation) and memsr (efficient training super-resolution model). he once participated in the national training team for the informatics olympiad and met her during the competition. the company's operating partner ren jinshan was once the top scorer in liberal arts. he graduated from peking university's guanghua school of management and holds a master's degree in art theory from the university of chicago.

aiuni has successfully obtained three rounds of financing including angel+ round.

create vertical models to form technical advantages

in the process of making products, hu yating found that since almost all animations and games are centered around characters, the most valuable of all types of 3d assets are characters. in addition, many users create secondary creations of classic characters or have original characters they want to realize, and the demand for original creation is relatively large. hu yating believes that this part is more suitable for aigc to generate.

"these users want to generate 3d characters conveniently and cheaply, but the cost of 3d modeling is high, and in most cases it can only be used in b-side studios such as animation, film and television."

when ai is used to lower the threshold for 3d modeling to the extreme, users who were previously unable to create 3d content will have the ability to create and can independently produce new 3d works.

"most of the video content we see now is real-life. in the future, animations and special effects based on 3d models will be made into short videos, and the amount and creativity of videos will be greatly increased," said hu yating.

compared with other images, character generation is more difficult. hu yating said: "because 3d models are more professional scenes, and aiuni will do character generation, animation generation and video synthesis in the future, and provide services to digital content creators. these have put forward higher requirements on model accuracy. because it involves fine dimensions such as human bones, clothing accessories, etc., a lot of details and data are required, and geometric data processing is a huge challenge. at the same time, in order to support standard model poses (i.e. a pose), material and motion data also need more standardized processing.

aiuni's solution is to achieve sota-level accuracy through algorithm architecture innovation. compared with previous methods such as score distillation sampling (sds), aiuni solves the problems of long-term optimization, poor geometric quality, and inconsistency in model generation. at the same time, the team also optimized the problem that the multi-view diffusion model method is limited by local inconsistency and generation resolution, making it difficult to produce fine textures and complex geometric details. for the first time, the resolution was increased from 256 to 2k/8k to meet users' requirements for model accuracy and quality.

unique3d's paper introduces that this solution can generate better 3d results through a multi-view diffusion model and a corresponding normal diffusion model, a multi-level upgrade process, and an instant and consistent mesh reconstruction algorithm isomer. in the experiment, aiuni's model was compared with instantmesh, crm and openlrm, and was able to generate more accurate geometry and detailed textures, which was significantly better than other models.

"unique3d is both generative and generalizable, and can actually be implemented in the rendering engine to create content that is very valuable to users." hu yating believes that compared to competitors, the advantage of unique3d is that it can combine cg graphics and ai differentiable 3d rendering very well.

compared with the generalized large models of large manufacturers, hu yating believes that aiuni has more modes and is more vertical. "in fact, we have multiple vertical models of different modes, such as character setting models, 3d models, automatic binding or generation of action data and rendering synthesis models, etc. we will string them together vertically. and this is not a matter of pure computing power or data. it requires some innovation in algorithm architecture."

large video models have strong generalization capabilities, but they are difficult to control in terms of character consistency, action controllability, generation training costs, and inference efficiency. in comparison, aiuni has greater advantages in these aspects.

from 3d to 4d, for global ip creators

on the aiuni.ai website, there are currently two functions that have been launched, which can generate 3d worlds and 3d models respectively. hu yating revealed that the next functions that aiuni will launch will focus on the animation video mode - generating 3d characters, generating character animations, replacing real-life videos with 3d characters, and synthesizing character animations into real-life videos. this is also the direction that the aiuni team is working on, that is, dynamic 3d content (i.e. 4d).

the aiuni team believes that the development trend of aigc multimodality is from generating 2d to generating 3d models and finally generating 4d content. generating 2d images from text is just the beginning. giving 2d images a spatial dimension will give you a 3d model, which is also the basic carrier for future spatial computing. they hope that in the future this function can give 3d models a time dimension, making it a high-frequency, interactive 4d content.

"after generating the 3d character model, we can extend it to 4d, such as changing the model's movements or allowing the model to interact with users. we can also have the ability to render videos and interact with voice. any model can continue to create dynamic content." hu yating said that to complete the entire workflow from natural speech to pictures, and then to 3d models and dynamic videos, the most important thing is the combination of ai technology and graphics cg art.

aiuni first launched a beta version of 3d model generation in april and distributed invitation codes in some channels. in october this year, the company plans to launch a new public beta version for ip creators of character animation.

"a user may not have used professional 3d tools before, and may just be an acgn enthusiast of games or animations. through our platform, he can also create his favorite ip characters for original or secondary content, and export animated videos on the platform to disseminate them in creator communities and new media platforms." hu yating said that the public beta of the new version will support character generation and animation generation, and is suitable for scenarios such as spoken videos and dance videos.

in terms of business model, aiuni plans to launch different charging models for professional creators and ip enthusiasts. 3d professional users need 3d raw data, and the platform can improve their productivity, so they will have a strong willingness to pay; for new media ip creators, token fees are mainly charged based on generation time and rendering accuracy. for other types of users, the creator economy is more diverse. for example, some users want to add personalized voice models for characters, or 3d print models for making other ip derivatives. these are potential payment points.

hu yating said that in the future, aiuni will focus on new media scenarios for ip creation and gradually launch more multi-modal algorithms that combine ai technology with cg art. aiuniai has 70% overseas users, and the platform will operate for global creators. hu yating said that 3d/4d modalities are still new things, and chinese people are doing more, and the company has no direct competitors overseas.

"our mission is to create an ip platform for ai-generated 4d animations for digital content creators of new media. 3d generation is our first step. we have unlimited creative space in this new modality. we also believe that aigc combined with content upgrading will give the new generation of ip creation endless vitality." hu yating said.

the industrial paradigm revolution brought about by ai technology has enabled many young entrepreneurs like hu yating to realize their dreams.