Source: New Wisdom Element

Image Source: Generated by Wujie AI

Adobe's AIGC image generation platform Firefly recently underwent a major update. Adobe directly named it Firefly 2.

First of all, the image model quality of Firefly Image 2 has made a significant leap.

The new model image on the right can even show the wrinkles of a marshmallow.

The new model on the right has much better details of bird feathers and depth of field compared to the previous generation model.

The Firefly Image 2 model can generate images of significantly higher quality, especially when rendering realistic human bodies, involving high-frequency details such as leaves, skin textures, hair, hands, and facial features. Images generated using the Firefly Image 2 model have higher resolution and more vibrant colors and color contrast.

In addition, a new function for generating vector graphics has been added, which can directly generate vector graphics through prompts.

Adobe has also added several new features that can greatly improve the user experience:

Match generation: Images with styles similar to the user's uploaded sample images can be generated.

Image settings: Detailed settings for generated images, including depth of field control, motion blur, field of view adjustment, and generation.

Prompt optimization: For user prompts, the system can automatically generate prompt suggestions to help users better express their needs and obtain more ideal image outputs.

More powerful image models, comprehensively improving image generation quality

Adobe claims that Firefly 2 has greatly improved the quality of image output, especially in aspects such as leaves, skin textures, and facial features, compared to the previous generation model, providing more realistic and detailed details.

Especially for portrait photos, the rendering quality of human bodies has been enhanced by improving skin, hair, eyes, hands, and body structure, providing better colors and more accurate dynamic effects.

The skin of the character generated by the new model on the right is more realistic and natural, and the depth of field effect in the background is also more pronounced.

In the left image, the lighting and shadow effects of the new model's character are more realistic and consistent, and the background details are more natural. In the middle image, the cloud details are more natural and rich, and the lighting and shadow are softer. In the right image, the character details are more realistic, the hand depiction is more natural, the marshmallow effect is more realistic, and the details are richer.

From the content of the images, it is obvious that the new model has a more accurate understanding and depiction of what a CRT monitor and television are. The previous generation engine basically only depicted a display screen, while the new generation engine draws the details of the monitor more reasonably.

From the perspective of image style, the images generated by the new model have more natural lighting and shadow, are more realistic, and have more consistent color tones.

Like several other image generation AIs, Firefly also allows users to freely choose the model version for image generation. Although version 2 is theoretically better than version 1, there is no problem at all if users want to use version 1 for image generation.

Tested 3 new functions, giving users the ability to finely control images

The change in image models is a major upgrade of Firefly's overall image output quality.

However, the most important aspect for users of image generation AI products is the ability to control the image generation process. Adobe's decades of accumulation in these areas are obviously not something that a newcomer in image products like Midjourney, DALL-E 3, can compare with.

For the updated match generation, image settings, and prompt suggestion functions, we also conducted some practical tests.

The biggest feeling after using it is that the combination of PS and AI for image generation is really cool!

First, by registering through the official website of Firefly, you can use all these functions.

Entrance address: https://firefly.adobe.com/inspire/images

Match generation function:

Similar to the "style converter" function of other image generation AIs, the match generation function can generate images with highly similar styles and content based on a given image by the user, which can be understood as Adobe's version of "image to image" function.

For example, if we want to generate an image of eyes in the style of a sketch pencil, we only need to provide a sketch with the same style, and then enter a prompt, such as "a person's eyes."

Then, Firefly 2 will create four images in the style of a sketch pencil, perfectly matching the style we expected.

Using a "neon theme" image with the prompt "puppy" to generate a "neon puppy."

Generate a papercut-style image of a cat based on the original image, combined with the prompt word "cat."

In addition, you can also adjust the image style according to personal preferences, transforming it into a vivid three-dimensional style, giving the image more vivid and captivating qualities.

This feature allows you to easily customize and create unique images according to the requirements of different projects or personal creativity.

Image Settings

In addition, Firefly also provides the ability to edit generated images, essentially integrating the capabilities of PS and AI for image generation. For users with high demands for image details, this provides a very convenient workflow.

For example, you can use the image of the papercut cat generated by the system to edit the material at any time, including adding text, graphics, or other elements.

This flexibility allows you to further personalize and improve the generated content according to your needs, creating images that match your unique creativity and meet your specific requirements.

This feature provides users with more choices and control in terms of creativity and customization.

Furthermore, further modifications can be made to visual intensity, color tones, lighting, and composition. For example, based on the generated cat image, adjustments can be made to color tones, lighting, and other effects in the lower right corner.

If you want to generate a golden cat, you can make the selection in the color and tone options of Firefly 2.

You can also composite more elements in the creation, such as scenes shot through a window.

Prompt Word Suggestions

As the biggest bottleneck for human input in AI image generation, Adobe has also provided its solution in this update.

If the user's input prompt is too short or not clear enough to describe the user's needs, the system provides the function of prompt word suggestions.

For example, when we input a short prompt like "a woman is on top of the mountain during sunset," the system will automatically generate five different sentences, presenting multiple ways to express similar situations for us to choose from and further refine our creative needs.

This feature allows us to more flexibly customize and expand our creative ideas, increasing the possibilities and diversity of creativity.

These aspects collectively demonstrate the enhanced performance of Adobe's release of Firefly 2. It not only improves image quality and introduces the function of generating vector graphics but also adds multiple features, significantly improving the user experience and providing outstanding tools for creators and designers.

Reference: https://news.adobe.com/news/news-details/2023/Adobe-Releases-Next-Generation-of-Firefly-Models/default.aspx

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。