Breaking news

Demonstration of technology currently being developed by Adobe. Separate voices, synthesize ponds, and make a woman wearing a mask smile – AV Watch


Adobe held “Adobe MAX Japan 2023” at Tokyo Big Sight on November 16th, and as one of the “Sneaks” to introduce technologies under development there, Adobe will separate background sounds and human speech from miscellaneous sounds. The technology “Project Sound Lift” was unveiled for the first time in the world. It also introduces “Project Fast Fill,” which allows users to type in prompts to composite objects such as ponds into videos.

The technology introduced in Sneaks is currently being developed by Adobe Research, the company’s research institute, and based on the feedback from this sneak peek, we will decide whether to continue development or commercialize it. Whether or not the technology will be adopted in products and the exact timing of its adoption have not yet been determined.

In the demo, after uploading a file to the browser and specifying the sound you want to separate, a file with that sound separated was generated.

This time, Adobe MAX Japan 2023 will introduce a total of 7 Sneaks. One of the technologies unveiled for the first time in the world is “Project Sound Lift,” which “accelerates audio editing using generative AI.” It is possible to separate each sound from audio data that includes human speech.


From the “Project Sound Lift” demo.There are three audio files: original data, only speaking voice, and only environmental sound.

After separation, each piece of audio data can be treated as separate audio data, so editing can be done such as applying effects to only the speaking voice or reducing only the environmental sounds to make the speaking voice easier to hear while retaining the sense of presence in the venue.

As for the function that emphasizes speech, “emphasize speech” has been implemented as a beta version in Premier Pro, but Project Sound Lift is unique in that it can treat speech and environmental sounds as completely separate data. .

We also introduce “Project Fast Fill,” which applies the “generated fill” that is already installed in Photoshop etc. to videos. In the demo, examples were introduced, such as typing “pond, chair” into a video of a park with a wide lawn to combine a pond and a chair, and converting a woman wearing a mask into a smiling face.

Other sneakers introduced included “Project 3D Edge Printing,” which performs 3D printing on the edges of books, and wearable technology “Project Primrose.”


Wearable technology “Project Primrose”.Dresses made using the same technology were also prepared at the venue.

Adobe Firefly Image 2 Model


Keynote speech

The keynote speech, which began at 10am, introduced new features included in Photoshop, Illustrator, Premier Pro, and more. Among these, it was also announced that the second generation model of the generative AI “Adobe Firefly”, “Adobe Firefly Image 2 Model”, will be made generally available.

It is a generative AI developed based on the first model and was announced at Adobe MAX 2023 held in California in October. It was previously available as a beta version, but it will become generally available on the 16th.

With the start of general availability, the accuracy of outputting results from text prompts in multiple languages ​​has been improved. Images can now be generated based on a more accurate understanding of the cultural context.


Black dog image generated with Adobe Firefly Image 2 Model


Image of a black dog generated by the first generation AI

In the keynote speech, images of a black dog generated using the first model and Adobe Firefly Image 2 Model were compared, and it was found that the Adobe Firefly Image 2 Model was able to generate more photorealistic fur, eyes, etc. explained.


In addition, the keynote speech included a demonstration of Premier Pro, which uses AI functions. New features were introduced, including the ability to detect and trim filler words such as “eh” and “ah”, the ability to edit videos from automatically transcribed text, and the “emphasize speech” feature.

The event will also be streamed live on Adobe CreativeStation official YouTube, where you can watch the keynote speech, each session, and Sneaks.

Adobe MAX Japan 2023 Online ~CC Dojo Special~ | Adobe Official

This is the first time in four years that Adobe MAX has been held in real life since 2019. Approximately 3,600 people are expected to attend.


The article is in Japanese

Tags: Demonstration technology developed Adobe Separate voices synthesize ponds woman wearing mask smile Watch


NEXT Rengo confirms wage increase demands from 2024 spring labor union Chairman Yoshino “It is important to pass on labor costs to prices” | Reuters