Adobe's new version of AI painting explosion, 2k resolution can be played online, netizens: the effect is stronger than DALL·E 3

Firefly’s new version of Vincent Graphics Tool will directly hit the Adobe Spring Festival Gala!

At Adobe’s annual MAX conference, they launched a series of eye-catching new features. The most eye-catching one is "Wensheng Image 2.0" Firefly Image 2

not only generates better image quality, the default generated resolution has increased to 2048×2048, and every cat hair can be seen clearly :

Pictures

A large number of new features have been added, including the "Picture to Make Picture" function, which can imitate similar styles and generate various settings Adjust effects such as depth of field and motion blur

The most important thing is that, like DALL·E 3, it is friendly to newcomers: if the prompt word "writing is stuck", Adobe Firefly can also give direct suggestions.

Picture

Some netizens couldn’t wait to try it out and exclaimed: “It’s really great!”

Picture

Some netizens even believe that the generation effect is better than OpenAI’s DALL·E 3

Picture

So, what is the actual generation effect of Adobe, and how to operate it?

How do you compare Midjourney and DALL·E3?

Currently, many netizens have provided comparative results of several mainstream AI tools.

For example, Kris Kashtanova, who works at Adobe, released several mainstream AI painting tools Midjourney, SDXL (Stable Single portrait comparison of Diffusion XL), DALL·E 3 and Firefly Image 2.

Picture

Among them, Midjourney and SDXL (Stable Diffusion XL) lost in terms of command compliance, because the prompt word required "a thumbs up" ”;

Then comparing the DALL·E 3 and Firefly Image 2, the portraits generated by Firefly Image 2 are closer to “photo clarity”, while those generated by DALL·E 3 are more like portraits.

Some netizens also specially evaluated the effects of these models on generating single female portraits under the same prompt words.

He found that compared to Midjourney and DALL·E 3, which tend to generate "young and beautiful" women by default, Firefly Image 2 generates portraits with more flexible age groups:

Picture

However, if you deliberately add "young and beautiful" to the prompt word, Firefly Image 2 can also generate a young girl:

Picture

In addition to personal photos, some netizens also conducted family photo tests

Regarding the issue of hand details, Midjourney and Firefly will still find finger problems when zooming in, but DALL·E 3 It has been able to solve the hand generation problem well.

But in terms of portrait fidelity, Midjourney and Firefly are better than DALL·E 3 (this is also a highlight recognized by many netizens):

Picture

However, some netizens pointed out that if Firefly Image Model 2 is deliberately asked to generate human hands, the details and fidelity seem to be pretty good

Picture

In this regard, some netizens evaluated and summarized the four major features of Adobe Firefly Image 2:

• Photography quality
• Color and dynamic range
• Can recognize more landmarks and cultural symbols
• Can better generate portrait details

Pictures

According to Adobe, Firefly users have generated 3 billion images using these tools, with 1 billion generated in the last month alone Image

Regarding model details, Alexandru Costin, Adobe’s vice president (responsible for generative AI and Sensei), said in an interview with TechCrunch:

Although Firefly is composed of multiple models Collection, but compared with Image 1, the overall model volume of Image 2 has increased by 3 times, and the training data has also increased by 2 times.

However, the above are only the evaluation results of netizens.

How to start using the latest Adobe Firefly Image 2, if you want to try it yourself?

How to play Firefly Image 2?

First, open the Adobe Firefly official website, and then log in to your account

Picture

Please click "Text to generate image", and then Enter the prompt word below (since last month, software and web pages have been able to directly enter Chinese)

Picture

to enter the trial interface. Firefly will give 4 reference samples at once, each with different style effects:

Picture

Look from the functional interface on the right , this time Firefly Image 2 has added many new features.

In addition to adjusting the model version and aspect ratio:

Picture

You can modify the image style to make it more Close to a photo or artistic portrait. At the same time, you can also manually adjust the visual intensity

Picture

In addition, you can also upload images yourself, or select directly from the reference image library, Firefly Image 2 is required to imitate (the degree of imitation is customized):

Picture

The generated effect can vary due to factors such as material and movement, and also You can adjust it through professional photo settings such as aperture, shutter speed and angle of view.

Picture

If you want to download the image, the official website will also pop up a prompt interface. Tell you that they will "add watermark" to the image to prevent others from knowing that it was generated by AI:

Picture

After a round of observation, it can be confirmed DALL·E 3 already has "help writing prompt words" and Midjourney's detailed parameter adjustment functions

Adobe announced that in addition to the key updates of Firefly Image 2, it will also apply Firefly to Photoshop, Illustrator, Premiere Pro, etc. Among a series of suites

Have you tried Adobe’s latest series of AI tools? How do you feel about it?

Reference link:
[1]Official website: https://firefly.adobe.com/
[2]https://blog.adobe.com/en/publish/2023/10 /10/next-gen-of-creativity-powered-by-ai
[3]https://techcrunch.com/2023/10/10/adobe-firefly-can-now-generate-more-realistic- images/

The above is the detailed content of Adobe's new version of AI painting explosion, 2k resolution can be played online, netizens: the effect is stronger than DALL·E 3. For more information, please follow other related articles on the PHP Chinese website!