Large models are popular with digital people: one sentence can be customized in 5 minutes, and you can hold it while dancing, hosting and delivering goods

You can create a 3D digital person who can work directly in as little as 5 minutes.

This is the latest shock that large models have brought to the field of digital humans.

Just like this, one sentence describes the demand:

The generated digital people can directly enter the live broadcast room and become the anchor.

It’s easy to dance in a girl group dance.

# During the entire production process, just say whatever comes to mind. The large model can automatically disassemble the requirements, and you can get designs and modify ideas instantly.

△2x speed

No longer afraid that the boss/Party A’s ideas are too novel.

Such Vincent digital human technology comes from the latest release of Baidu Intelligent Cloud. It’s time to say it or not, but it’s time to cut down the barriers to digital people’s use in one fell swoop.

After hearing about such a magical tool, we immediately secured the qualification for internal testing as usual. Let’s take a sneak peek at more details~

5 minutes per sentence, 3D digital people will be on duty directly

From Chatbot to Vincent Pictures, to Vincent Videos, it goes without saying that the changes in interaction methods brought about by large models are needless to say.

Now, on Baidu Intelligent Cloud Xi Ling platform, based on Wenxin Yiyan 4.0, digital human customization can also be realized through natural language dialogue.

For example, how many steps are needed to generate a brand spokesperson?

First, enter the prompt word "Generate a Baidu Smart Cloud brand spokesperson" and upload the logo image at the same time.

The big model will automatically start thinking step by step from multiple dimensions such as face shape, hairstyle, makeup, clothing, accessories, etc.:

Automatically create a digital person that meets the requirements.

△8x speed

If you need to adjust details, you can do it just by "speaking".

In just 5-10 minutes, a 360° high-quality digital human with no blind spots is basically formed.

After pinching the face, the next step is to attach expressions to the digital person so that he can move. It also only requires one click and wait for 1-2 minutes.

Compared with the customization cycle of high-precision 3D digital people in the past, which took several days or even months, this minute-level efficiency , it can indeed be called "subversion".

It is worth mentioning that under the premise that the efficiency has been greatly improved, the detail quality of such Vincent Digital Man still maintains a high level.

Expression details:

Action quality:

Combined with Baidu Intelligent Cloud's long-term accumulation in the field of digital people, it is easy to broadcast news and deliver goods live.

Comprehensive AI-based digital human technology

In addition to the intuitive improvement of efficiency and implementation capabilities, behind the Wensheng digital human solution launched by Baidu Intelligent Cloud this time, Many technical details are also worth talking about.

As mentioned above, its technical base is Wenxinyiyan 4.0.

The large model capabilities that play a key role include:

• Automatically dismantle the tasks and subtasks to be done
• Display the thinking process, be well-founded, and make the entire generation process "white box"
• realizes short-term memory based on content extraction, which can Continuously adjust the digital human image through dialogue

In this way, the large model becomes a digital human modeling assistant that can understand the psychology of human Party A and can imitate humans ideas, dig into every detail of digital human customization, and make the process controllable.

At the same time, the large model also demonstrates the ability to call tools behind the scenes.

For example, the "knowledge base" covering 6000 dimensions of face shape and facial features details is called to adjust the digital human face as a whole.

In addition to large model technology, Baidu Smart Cloud has also added new AI rendering technology to the Xi Ling platform, supporting AI drive and AI cloth simulation, making digital people's expressions and body movements more natural, and the texture of clothing fabrics More real. Includes:

• Dynamic wrinkle maps to make textures more realistic.

• Minute-level 4D automatic binding allows eyes, lips and other parts to be perfectly closed, and supports expression style switching.
• Real-time simulation of limb muscle extrusion and collision.
• ……

Officials also revealed that next, Baidu Intelligent Cloud plans to implement comprehensive AI for characters, behaviors, scenes, lighting, and lens elements.

Digital people, entering the era of large models and applying a new paradigm

If last year everyone was still discussing basic models in full swing, then this year Sora has Since then, the changes in application paradigms brought about by large models have become a new hot topic in the technology circle.

On top of the changes in interaction methods, what has attracted the most attention is actually efficiency improvement:

Outputting ideas and generating what is needed, large models are allowing more and more people to Many tasks that originally required a lot of time, manpower, and money have become simple, efficient, and available to everyone.

Now, the latest technological progress of Baidu Intelligent Cloud in the field of 3D digital people is a representative of the expansion of this possibility beyond the more familiar image and video fields.

It is foreseeable that more digital personnel, who were used in large enterprises and institutions in the past, are driven by the new paradigm. , it is becoming possible to enter "ordinary people's homes".

Previously, data from Tsinghua University's "Virtual Digital Human Research Report Version 2.0" showed that from the perspective of the layout of leading companies, digital human products and services for the B-side are the main component of the market, accounting for 79% .

As large model technology subverts the application model of digital humans, not only small and medium-sized enterprises no longer have to be afraid of 6-digit 3D high-precision digital humans, but C-side applications will also be expanded.

This also means that the application and commercialization of digital humans has turned a new page.

The above is the detailed content of Large models are popular with digital people: one sentence can be customized in 5 minutes, and you can hold it while dancing, hosting and delivering goods. For more information, please follow other related articles on the PHP Chinese website!