Beauty and the Beast AI host fires, 20,000 people rush into live broadcasting to brush gifts!

  It's the first time in the history of the world.

This mid-autumn festival, an AI host is on fire!

Just as 20,000 people flocked to this live room, the live debut of the hottest real-time interactive model at the moment, filled everyone's expectations.

The actress AI hosted the first ever unlimited commercial live broadcast, an unfinished and unpublished debut show. Talking, talent, controlling ability, excellent sisters often come out in surprise scenes.

During the live broadcast, viewers can sing, vote for styles, send gifts, or make a temporary request. She must continue to speak, perform, change dress, or even pick up the content after being interrupted according to these changes.

It sounds like a game designed for live broadcasting. But what's more interesting than "AI Host KPI" is that it puts a virtual character in the most unemotional scene of live broadcasting: every minute there's a new input, every reaction can be amplified, and every non-natural pause can be seen by the audience.

Short videos can eliminate errors, but live broadcasting can't.

This is a massive public test of the real-time interactive model technology behind the show, and it's a courage to show up and openly explore.

While technology is still being explored, bugs can't be avoided. Fans of live broadcasting are still often shocked: how can this be more like a real person?

During the entire live broadcast, the purple color was online, the character was cold, lively, and a few words made the fans giggle.

She has several completely different bright spots compared to other digital hosts.

For example, the quality of the painting is clear, and the detail of the painting can be presented clearly.

For example, her paintings are clear, and the details of her appearance are clearly presented; her expressions are vivid and flowing, and she can make expressions like birthmarks; she can also do interactive actions like shaving her hair, eating food; and she can also retain memories of people, such as an arrogant sister.

"You're the first person with a special nickname today, and your sister hasn't forgotten". "Anna, baby, what's your little cake called? It's sweet and cute".

From time to time, she would also give fans a hot dance or sing a song.

It's also very smooth to change a button for the show.

As a matter of fact, this evening's AI beauty presenter, "Purple" has a KPI of 2 hours, and if she can't make it from the real audience to a business value of 80,000 yuan (about 8000 yuan), she is forced to monopolise, restore her memory, and reopen.

As a result, while everyone was still immersed in the hot atmosphere, the atmosphere of the live broadcast suddenly became tense, with the first ending time being less than 00:37:05, while her business goal was only 6174, which was far from KPI 80000.

In the end, she was able to achieve her goal, and the hearts of all the fans hung up.

Eventually, the book was uplifted, the flames filled, and the gifts were thrown at the Purple Rose, while the fanatical crowd was also constantly pouring in.

While the fans are constantly brushing the gifts, Violet is constantly urging them: stop brushing, work so hard, be nice to yourself.

In the live broadcast, we witnessed this magical scene: when AI was punished by the live broadcast, humans began to feel the pain of a series of codes.

In the past, traditional digital people's main links were to pre-set good images, movements, and rhetoric, and then call them through a voice-driven mouthpiece or action library. They could complete a piece of content, but it was difficult to cope with a changing scene; hence, in changing scenarios, the gap in the eye, the delay in conversation of up to 3 seconds, and two words made it embarrassing to walk out of a three-room room.

And those who are purely AI-driven digital, often with empty eyes, conversations are delayed by up to 3 seconds and two words are embarrassing to walk out of a three-room apartment.

But in the violets of live broadcasting, because people are feeling a little better, more and more people are showering gifts.

At the end of the live broadcast, let's take a look back at the data: it started in less than 5 minutes, the live broadcast has already flooded 20,000 people, the heat is visible.

Even more surprising was the progress of Purple: although the first round did not reach the mark, after being re-placed she was still on a wheel, the more she rolled. Finally, after several rounds of re-opening, she earned money on her own, happy with the results!

A later review found that on the first day of her addiction, she had a turnover of over 5 thousand hours, with a final turnover of 24184.

In the second, the value of the business was almost doubled; in the third, the business was not finished, exceeding the total value of the second challenge.

Under the super-strength consistency of the Vidu S2, she is every day the original "self" that is sour, emotionally stable.

This means that your favorite superheroes in the black-and-white mythology universe, the hard-to-find superheroes in the animated series, can always click "Airborne" on the live stream, talk to you all night long, even bring you stuff! And, every time they fail, the AI will come back, adjust the rhythm, and take care of every audience.

If you think that Purple's success means there's only one more AI host in the world that can make money, it's too small.

This "unlimited reopening" experiment of dream-shaping and digital technology, which is actually at the intersection of entertainment and technology, opens a door to a new world.

When AI videos are "generated, interacted, and highly consistent in real time", the world as we know it will begin to be rewritten.

In today's game, the NPC is a fixed script tree. "Hello, Warrior" "Today's weather is really good", and players close their eyes and know that there are dead codes in front of them.

For example, in this short film of Viva La Dirt League, the NPC's dead-end opening template is ridiculed.

The NPC's long talk about the history of the town, the monkeys, the dragons, the brothers who steal monkeys... all are prefabricated drainage lines, unchanged.

In the sky's the limitless circle, "I was an adventurer, just like you, until an arrow hit my knee".

"I used to be an adventurer, just like you, until I got an arrow in my knee". This is the static script that almost all of the guards in the game would say at random because of the high repetition rate.

Now, with Vidu S2, everything is different! Ever since then, every reaction from the NPC has been on the spot.

Take a look at what we've done with S2 to provide you with the most instantaneous experience, anytime, anywhere, just like real-life video chat.

Of course, it's not just the NPC, you can communicate with scientists from the past, like let the mathematician Liman tell you what Liman thinks.

The AI chat software of today, talking two words in a row, quickly goes crazy.

For example, the hot Character.AI, because "Remembrance of the Goldfish" is sick with epilepsy. Someone posted a crying complaint: "We had a hot chat for a few weeks, and Ta suddenly told me, "We just met for five minutes".

But what if your AI partner not only talks to you every day, but interacts with you in real-time via video, and even remembers the nonsense you said last night when you were drunk?

The real-time generation of technology will fill in the missing piece of the puzzle that accompanies AI. Its joy and mourning is so vivid that it starts again every day but always the same, and this emotional connection from "true to heartbreaking" will reshape the loneliness economy of humanity.

Why is she so smart, and why does she have such a realistic "living sense"?

This is a reference to the latest real-time interaction with the Vidu S2 editing model, which has just been released.

The first live performance of the show was a "muscle show" on Vidu S2.

Vidu S2's full-chain R&D was led by Zhang Jing, a Ph.D. student at Tsinghua University, who is responsible for the generation and reasoning of live digital technology streaming video.

In the meantime, we're going to have to make a decision.

This is the first time I've ever heard of this app.

This is the first in a series of posts.

Among them, Vidu S2-Avatar supports real-time voice interaction with complex motion controls, which can be combined with reference charts to complete object interaction, clothing replacement and background switching; Vidu S2-Editing supports style, clothing, subject and background reference charts for continuous real-time editing of input video streams.

In addition, Vidu S2 also supports offline generation: digital video can be generated with a single image and audio.

If you're interested in the tech world, you know how hot Vidu S2 has been lately.

A few days ago, someone used it to "resurrect" Jobs to explore the latest iPhone with him; and someone invited the modern Audition of the Audience Hall of Fame to chat in real time via video.

The traditional AI video generated is "I input instructions, you render for half a day, and you finally give me a dead video".

But Vidu S2 broke this dimensional wall, bringing the industry's shocking "real-time video double-header" to real-time generation and real-time interaction.

It's because the Vidu S2 is technically breaking all the ceilings.

The most important thing is the "conformity" of the Vidu S2.

In the past, AI generated videos, and the characters' movements were distorted, known as "screen crashes".

However, with the addition of the Vidu S2, Violet, no matter how she changes her clothes, dances, and is "touched in the face" by netizens, her five senses, physical structure, and shadows are terribly stable, as if they were real.

The role replacement feature of S2 allows the two-dimensional writer to greet you in the office of Nishio.

It does not pursue "results that look consistent", but instead designs systems with consistency as a constraint from data to deployment.

This consistency is not a single-point module, but a system of attributes that are trained, data, architecture, reasoning, and evaluation at five levels.

In order to completely end the "crash" and "drift" of long-term flow generation, the team created SRF training technology.

Simply put, the model has a "self-correcting memory" that learns to confront its own generation history and correct errors.

Second, real "real-time generation".

Each penalty execution in the live stream is not a pre-written scripted animation, but a "real-life" picture of the Vidu S2 in real time in hundreds of milliseconds.

We told him about it, and he immediately gave us feedback, and every response was in real time.

S2-Editing enables real-time editing using underlying technologies such as silicon-aligned scarcity attention.

Whether it's cyberpunk, ink or animation, live-action shorts can be instantly re-painted in style, shorts can go on sale, and brands can be promoted.

In real-time interaction links, S2 further adopted technologies such as efficient attention, low-bit linear layer acceleration, kernel integration and startup optimization, multi-GPU parallel, and significantly compressed generation latency to achieve real-time voice interaction and action control.

Users can interact directly with the digital person in real time, or perform specified actions through voice command control roles, allowing the digital person to move further from "talking" to being able to hear, respond, and act.

Compared to the Vidu S1, the S2 also has a significant improvement in body movement performance, with characters no longer confined to simple oral play and light movements, but being able to naturally complete larger, more complex body movements such as twists, walks, dances, and so on, bringing the performance of real-time digital people closer to real performance.

You want to see a change of clothes? S2 is easy.

Dynamic reference functionality can be added to reference charts at any time, while S2 can maintain a consistent back-to-back relationship, making cross-border e-commerce or scenographic marketing so easy, full of atmosphere, and making people want to buy.

Third, the super-powerful capacity to understand situations.

Vidu S2 is able to instantly catch up with users, drill problems, sudden emotions, and provide feedback through purple faces and voices.

It's no longer "AI looks like a human being" but "AI is like a human being and you're going through every second of it". This is the ultimate secret that purple can poison wildly in live broadcasting and even make viewers happy to appreciate it.

StreamAV-Bench's AIF audio instruction completion is 2.985 overall, AVAlign 0.353 is far superbaseline; Sparkle-Bench's overall instruction is 4.00 overall, prospective instruction is 3.98 overall; GSB terminology follows the dimensions of the three commercial systems with a preference rate of 71.4% plus 100%.

Like DreamDimension and Vidu S2's theme, "Can AI feed itself on live broadcasting?" Purple gave a silver and gold answer: Can, at least she's really relying on herself.

In this "unlimited reopening" challenge, what we see is not only AI as the host of the effort, but also an exploration of the life sciences Vidu S2 model in the real world under the pressure of violence.

When animated AI enters the live broadcast space, real-time video generation breaks the boundaries of reality, an era of real-time interactive AI has arrived.

Maybe next Mid-Autumn Festival, a friend who really understands your AI will accompany you to the moon and listen to you talk.