- Регистрация
- 17 Февраль 2018
- Сообщения
- 28 199
- Лучшие ответы
- 0
- Баллы
- 2 093
Offline
Google has begun rolling out private access to its Veo and Imagen 3 generative AI models. Starting today, customers of the company’s Vertex AI Google Cloud package can begin using Veo to generate videos from text prompts and images. Then, as of next week, Google will make Imagen 3, its latest text-to-image framework, available to those same users.
With Veo’s rollout, Google says it’s the first hyperscale cloud provider to offer an image-to-video model. To that point, OpenAI’s Sora model is still only available to select artists, academics and researchers — though that could change quickly with the company teasing 12 days of product demos starting December 5.
Of Veo, Google says the model creates 1080p footage “that’s consistent and coherent” and can run “beyond a minute.” The tool is also capable of working with both text prompts and images. In the latter case, it’s possible to use either AI-generated or human-made pictures as the starting point for a video.
Looking at the sample footage Google shared, it’s evident Veo, like all AI models, can struggle with cause and effect. For example, in the clip of the roasting marshmallows, the treats don’t yellow and char as they’re exposed to the heat of a campfire flame. Artifacting is also an issue, as is apparent if you look closely at the hands in the concert footage.
Google
As for Imagen 3, Google says the model generates “the most realistic and highest quality images from simple text prompts, surpassing previous versions of Imagen in detail, lighting, and artifact reduction.” Here again, however, you don’t have to look too closely to see Google has more work to do.
In the first example of a group of friends sitting on the trunk of a car, the original prompt includes mention of “flash photography,” but the subjects are clearly backlit. One could argue that a flash was used to create intense backlighting, but if the idea behind the prompt was to create something representative of flash photography from the 1960s, this image isn’t it.
Still, Google is keen to get more of its enterprise customers using generative AI. Citing its own research, the tech giant says among companies using generative AI in production, 86 percent report an increase in revenue. However, a recent Appen survey found return on investment from AI projects fell by 4.6 percentage points from 2023 to 2024.
This article originally appeared on Engadget at https://www.engadget.com/ai/googles...ble-in-private-preview-160055983.html?src=rss
With Veo’s rollout, Google says it’s the first hyperscale cloud provider to offer an image-to-video model. To that point, OpenAI’s Sora model is still only available to select artists, academics and researchers — though that could change quickly with the company teasing 12 days of product demos starting December 5.
Of Veo, Google says the model creates 1080p footage “that’s consistent and coherent” and can run “beyond a minute.” The tool is also capable of working with both text prompts and images. In the latter case, it’s possible to use either AI-generated or human-made pictures as the starting point for a video.
Looking at the sample footage Google shared, it’s evident Veo, like all AI models, can struggle with cause and effect. For example, in the clip of the roasting marshmallows, the treats don’t yellow and char as they’re exposed to the heat of a campfire flame. Artifacting is also an issue, as is apparent if you look closely at the hands in the concert footage.
As for Imagen 3, Google says the model generates “the most realistic and highest quality images from simple text prompts, surpassing previous versions of Imagen in detail, lighting, and artifact reduction.” Here again, however, you don’t have to look too closely to see Google has more work to do.
In the first example of a group of friends sitting on the trunk of a car, the original prompt includes mention of “flash photography,” but the subjects are clearly backlit. One could argue that a flash was used to create intense backlighting, but if the idea behind the prompt was to create something representative of flash photography from the 1960s, this image isn’t it.
Still, Google is keen to get more of its enterprise customers using generative AI. Citing its own research, the tech giant says among companies using generative AI in production, 86 percent report an increase in revenue. However, a recent Appen survey found return on investment from AI projects fell by 4.6 percentage points from 2023 to 2024.
This article originally appeared on Engadget at https://www.engadget.com/ai/googles...ble-in-private-preview-160055983.html?src=rss
