Google's new Gemini Omni is first exposed. The video version of the "banana" is here, and the professor's formula derivation on the blackboard is all correct.

The Google I/O Conference is about to kick off, and the native Gemini has been exposed in advance!

Now, the entire internet is flooded with this video —

A professor is giving a lecture on stage while casually deriving formulas on the blackboard. The texture and smoothness are truly amazing.

Actually, this video is generated by the "brand - new video model" Gemini Omni, with top - notch coherence and consistency.

Some netizens exclaimed, "The video version of Nano Banana is here!"

Some people also said that "seeing is believing" no longer holds true.

The native Gemini Omni is exposed for the first time

Just yesterday, a screenshot of the homepage of the Gemini mobile app leaked, and an entrance for a brand - new video model, Gemini Omni, went live.

In the interface, it clearly reads —

Come and meet our brand - new video generation model. Remix your videos, edit them directly in the conversation, and try out templates.

Apparently, Google has introduced Gemini Omni in a brand - new form!

This might be an all - modality Gemini that supports input and output of text, images, audio, and video simultaneously.

Different from Veo, Omni will be deeply integrated into Gemini like Nano Banana, with better prompt understanding and reasoning abilities.

Especially in real - time video editing, it can replace objects and remove watermarks with a single click.

Meanwhile, the ID of the Omni model has also been exposed —

fbard_eac_video_generation_omni /bard/v3smm - lora - prod.goat - cr - rev6 - xm171555416 - at - 1200

It can be seen that the videos generated by Omni support a duration of 10 seconds and a resolution of 1280x720.

What really drove the entire internet crazy are several demos released in advance for testing.

The professor deriving formulas on the blackboard shocks the whole internet

The most astonishing one is the video at the beginning where "a professor derives trigonometric identities on the blackboard".

In the video, the professor holds a piece of chalk and writes out the mathematical proof step by step on the blackboard, while verbally explaining the current derivation step.

People in the know would be extremely surprised. How difficult is it to write the mathematical formulas correctly in an AI - generated video?

Text consistency has always been the "Achilles' heel" of video generation models.

Previously, the text generated by Sora often looked like words at first glance, but upon closer inspection, they were all "gibberish", not to mention a complete mathematical derivation.

In this Omni demo, the formulas are correct, the derivation is coherent, and the handwriting is natural.

Even more incredibly, this amazing demo was generated with just one prompt —

A professor writes out a mathematical proof for trigonometric identities on a traditional chalkboard, explaining the step he is currently on in the equation.

Many people were completely blown away after watching it!

It has to be said that AI video generation has crossed the "uncanny valley" and officially entered the hyper - realistic era.

Some netizens who got access to the gray - scale test also made a batch of similar videos, and they are all excellent.

The real killer feature: real - time editing

The power of Gemini Omni lies not only in generation. This time, there has also been a significant leap in "real - time editing".

In the leaked demonstration, Omni demonstrated astonishing editing capabilities —

• One - click watermark removal: Omni can remove watermarks through direct conversation, and there are no flaws in the picture;
• Object replacement: Just by speaking, objects in the video can be accurately replaced, and the lighting, shadows, and occlusion relationships are all automatically adapted;

For example, if you upload a video previously generated by Sora, Gemini Omni can directly remove the watermark.

Some people said that just the watermark removal feature alone is enough to make this tool a game - changer for creators.

Moreover, Gemini Omni also supports stylized output.

In the following anime - style video, the blue flame special effects and the lines of the fighting actions in each frame look like they were hand - drawn by a professional animator.

Video screenshot

However, early tests show that the quota of Gemini Omni is consumed very quickly.

Google makes a comeback, while Sora 2 shuts down

The timing of the leak of Gemini Omni is "precisely positioned".

Just two weeks ago — on April 26th, OpenAI's Sora App officially stopped service.

This AI video generator that once drove the world crazy has come to an end in its short and dramatic life.

Looking back at the cause of Sora's demise, it's simply a business tragedy:

The money - burning rate was astonishing. Allegedly, the inference cost of Sora was as high as 1 million to 15 million US dollars per day. Video generation is much more expensive than text and image generation, and this cost has never come down.

The most crucial thing is that OpenAI couldn't retain users. The peak number of active users was about 1 million, but it had dropped below 500,000 before the service shutdown, and the 30 - day retention rate was less than 8%.

The in - app revenue throughout its entire life cycle was only about 2.1 million US dollars, not even enough to cover one day's computing power cost.

On March 24th, the official Sora account on X posted that famous farewell message, "We're saying goodbye to the Sora app".

The API will be completely shut down on September 24th, marking the end of an era.