After images and music, Google is tackling video generation with Veo

Deal Score0
Deal Score0

Google I/O 2024 generative AI

Google I/O 2024: advances in generative AI.

© Screenshot / Google

On the artistic side, Google's AIs have not yet managed to make a name for themselves. When we talk about image generation, we immediately think of Midjourney or Dall-E, but not of ImageFX. The same goes for music, where Google's MusicFX AI is unlikely to mean anything to most users. But the Mountain View giant is determined to catch up with new AI models.

Advertising, your content continues below

Imagen 3: the best, according to Google

During I/O 2024, Douglas Eck, research director at Google, says that the teams worked hard to improve the quality of renderings, particularly for images, by reformulating the image generation algorithms from 0. Thus was born Imagen 3, which allows a better understanding of prompts, as well as a more “secure” and more easily identifiable rendering thanks to the integration of SynthID, to limit the creation of false information. Imagen 3 would be the best image generative AI offered by Google with even more photorealism, stunning photographic effects and details galore. We will not fail to verify these allegations by comparing a few prompts with the leading figures of the moment. Imagen 3 should also “create” texts in a more natural way, which, surprisingly, is quite complex for a generative AI to produce. You can already submit a few prompts to Imagen 3 on Google Labs.

Advertising, your content continues below

Google I/O 2024: Imagen 3

Google I/O 2024: Imagen 3 would produce even more photorealistic images.

© Screenshot / Google

Music AI Sandbox, to increase your creativity tenfold

On the music side, Douglas Eck announces having worked with artists to improve their creativity. Google therefore announces the arrival of Music AI Sandbox with YouTube, which allows you to create music tracks from a textual description, or to modify the style of a melody in a few seconds.

Veo, the Google-style video generator

Google I/O 2024: presentation of Veo

Google I/O 2024: Veo, videos generated by artificial intelligence.

© Screenshot / Google

The highlight of the show is of course Veo, the video creation tool. Google was particularly expected in this area, and the results seem interesting. Veo is capable of generating 1080p movies longer than a minute from text, images or videos. Veo can also take into account cinematic descriptions for more precise rendering, such as aerial views or a timelapse. This tool is the result of extensive work and is based on previous generative models such as Imagen-Video, Phenaki, WALT, VideoPoet and Lumiere. For the moment, Veo is reserved for a few lucky people and can be used by registering with Google Labs.

Responsible generation?

Just as for images (Imagen) or audio (Lyria), videos (Veo) will also be marked by SynthID to facilitate the search for productions produced by artificial intelligence and limit the risks of misleading use. Text generation (Gemini) will also be marked by this seal, or watermark digital. Good news, Google should make SynthID open source in the coming weeks, so that other AIs can benefit from this treatment via a Responsible Generative AI Toolkit. Google also recalls that it collaborates with the C2PA (Coalition for Content Provenance and Authenticity), which works to identify digital creations made by artificial intelligence.

However, there remains one point to display in order to be fully responsible: the energy cost. Indeed, these requests, these questions asked from a computer or a smartphone, can be charged a few cents, rates which ultimately do not reflect the real energy cost. Because while creating cats playing guitar can be fun, generating an image requires quite a bit of data processing with ultimately thousands of processors running constantly. Warning the user of the energy cost (and therefore of the CO2 emitted) would be a good thing to make creators of all kinds responsible.

Advertising, your content continues below

More Info

We will be happy to hear your thoughts

Leave a reply

Bonplans French