Imagen 4现已正式发布。
Imagen 4 is now generally available

原始链接: https://developers.googleblog.com/en/announcing-imagen-4-fast-and-imagen-4-family-generally-available-in-the-gemini-api/

谷歌发布了其最先进的文本到图像模型 **Imagen 4**,现已通过 Gemini API 和 Google AI Studio 广泛可用。此次发布推出了一系列模型——**Imagen 4 Fast、Imagen 4 和 Imagen 4 Ultra**,在质量、速度和成本之间提供平衡。 **Imagen 4 Fast** 擅长快速、大批量图像生成,每张图像 0.02 美元。**Imagen 4** 是一款多功能旗舰模型,具有改进的文本渲染效果,而 **Imagen 4 Ultra** 则提供最高的细节和提示词遵循度。 Imagen 4 和 Ultra 现在都支持高达 **2K 分辨率**,以呈现令人惊叹的详细视觉效果。所有图像均带有 SynthID 水印,以支持负责任的 AI 实践。用户可以探索示例,包括使用 Imagen 4 Fast 生成的风景和漫画,并访问文档和教程以开始创作。

## Imagen 4现已全面可用:反应不一 谷歌的Imagen 4现已广泛开放访问,但初步反应冷淡。许多用户对其图像生成质量感到失望,尤其是在遵循提示方面。展示的示例——一个四格漫画——包含大量错误,未能遵循有关角色动作甚至基本细节的指示。 一些评论员指出,早期版本(Imagen 3)为照片级图像产生更理想的结果,而Imagen 4则倾向于“卡通化”风格。与OpenAI的GPT-Image-1和尤其是Veo 3等竞争对手的比较表明,谷歌在质量和一致性方面落后。有人推测谷歌优先考虑Imagen 4的速度而非质量。 虽然一些用户通过迭代提示获得了成功,但另一些人质疑其价值主张,考虑到成本(每张图像2美分)以及实现所需结果需要多次尝试。人们也对公告和可用之间漫长的延迟表示担忧,这是对谷歌AI发布的常见批评。尽管声称“Ultra”版本在遵循方面有所改进,但早期测试并未显示出显著的改进。
相关文章

原文

We're excited to announce that Imagen 4, our most advanced text-to-image model, is now generally available in the Gemini API and Google AI Studio. This release marks a significant step forward in text-to-image generation quality, with substantial improvements in text rendering over our previous models.


The Imagen 4 family: A model for your creative needs

In addition, we're thrilled to launch Imagen 4 Fast, our new model built for speed, which is now available alongside the powerful Imagen 4 and Imagen 4 Ultra. The complete Imagen 4 family gives you a perfect tool for your creative needs, allowing you to balance between quality, speed, and cost.

  • [New] - Imagen 4 Fast: Ideal for rapid image generation and high-volume tasks, this model offers incredible speed at an accessible price point of $0.02 per output image.
  • Imagen 4: Our flagship model can be your go-to for a wide variety of high-quality image generation tasks, showing significant improvements in areas like text rendering.
  • Imagen 4 Ultra: When your creative vision demands the highest level of detail and strict adherence to your prompts, Imagen 4 Ultra delivers highly-aligned results.


Higher resolution for greater detail

Pushing creative boundaries further, both Imagen 4 and Imagen 4 Ultra now support the generation of images with up to 2K resolution. This allows for the creation of stunningly detailed and crisp visuals, perfect for things like marketing assets to intricate artistic compositions.


See Imagen 4 Fast in action

To give you a glimpse of Imagen 4's capabilities, here are some examples of what you can create. The prompts below, created using Imagen 4 Fast, showcase the model's versatility across various styles and content.

Imagen 4 Fast demo - landscape

Landscape/nature image: A breathtaking landscape of a mountain range at dawn, with a crystal-clear lake in the foreground reflecting the snow-capped peaks.

Imagen 4 Fast demo - four panel comic strip

Create a four panel comic strip in a retro style. The first panel should show a friendly cat sitting next to a Chromebook that is pulled up to the website https://ai.dev comic caption: Imagen 4 is now Generally Available! The second panel should show a dog saying “And we’re introducing Imagen 4 FAST which offers low-latency images at just $0.02 per image” panel three should show the cat saying “2K image upscaling is available too!” Panel 4 should show the cat and dog high-fiving with the caption “Try Imagen 4 in AI Studio now!”

Imagen 4 Fast demo - retro sci-fi movie poster

A retro science fiction movie poster with an airbrushed art style. The poster features a detailed spaceship, flying towards the right through a vibrant nebula in a star-filled deep space. The ship's two engines emit bright blue glowing trails. The title at the top of the poster reads "SUPER GALACTICA: THE LAST NEBULA" in a bold, beveled, metallic chrome font with a drop shadow. Below it, the subtitle "STARFALLS REVENGE" is written in a simpler, clean white font. The entire image has a vintage, weathered look, with a distressed, off-white border. At the very bottom, in a small font, is the text: "This poster was created by AI as was this disclaimer :)".

Start building with Imagen

As part of our commitment to responsible AI, all images generated by the Imagen 4 family are imperceptibly watermarked with SynthID. Ready to start creating? Dive into our official documentation and cookbooks to begin.

We can't wait to see what you build with Imagen 4 through the Gemini API and Google AI Studio

联系我们 contact @ memedata.com