DALL·E3 is coming! Not only is it coming, it will also be integrated into ChatGPT.In other words, in addition to the commonplace conversations, writing code, and solving math problems,ChatGPT, which integrates the latest DALL·E3, finally has a picture function this time.
ChatGPT+DALL·E, this wave is a strong alliance. One is the uncrowned king in the large language model, and the other is also the leader in the Vincentian graph model. The real effect will inevitably be 1+1>2.
This sudden official announcement is believed to have filled a big hole in the multi-modal ChatGPT that was widely rumored after GPT-4 came out at the beginning of this year.
However, OpenAI has only announced this news now.The specific launch time is October. For Plus and Enterprise Edition users, a separate DALL·E3 will also be launched this fall.
How powerful is this thing? Although we can't get started yet, judging from the examples released by OpenAI, it is still quite explosive.
Among them, some enthusiastic netizens called Midjourney directly and fed it the sample prompt words of DALL·E3 so that they could compete directly.
And the result can only be said:Midjourney is in danger.
The first is a very classic avocado medical meme. The prompt is: an avocado is sitting on a therapist's chair, saying "I feel so empty inside." There is a hole the size of a small crater in the middle of the avocado. Therapist, spoon, doodle notes.
Although at first glance they appear to be two different styles. But if we carefully compare the prompt words, it is obvious thatMidjourney ignores the therapist, spoon, and graffiti notes. The text in the dialog box is also written randomly and does not follow the requirements.
They were then asked to generate a picture of a translucent heart and asked to have a specific quote engraved underneath the heart.
Tip: This is an illustration of a human heart made of translucent glass, standing on a pedestal in a stormy ocean. Sunlight penetrates the clouds and illuminates the soul, revealing the tiny universe within. The quote “Discover the universe within you” is inscribed in bold letters on the base.
There is no doubt that DALL·E3 once again defeated Midjourney this time.In addition to not engraving the characters as required, Midjourney also failed to show any details such as the stormy ocean and the inner microcosm.
Here’s another photo of a lychee-inspired spherical chair, with details that call for a white bumpy exterior and a soft interior that contrasts with the tropical wallpaper behind it.
This brings all elements of the picture generated by them to life.But Midjourney seems to have misunderstood the difference between tropical wallpaper and tropical rainforest.
Of course, misunderstanding prompt words and taking them out of context are equivalent to the chronic diseases of the previous Vincentian diagram model.
Just give birth to a crab like a hermit crab...
Asking it to generate a 2D anthropomorphic forest band resulted in a 3D...
As for these old problems, according to OpenAI's own statement and the examples given, this situation basically does not exist in the new DALL·E3.
In addition to solving old problems, DALL·E3 has also upgraded the texture of the original second-generation version.
For example, let them draw a scene of a basketball player dunking, with the element being an explosion in the starry sky.
Originally, the pictures generated by DALL·E2 already met the requirements. Unexpectedly, the upgraded DALL·E3 was more realistic, with details such as muscle lines and the colors of the universe displayed one by one. It was indeed a blow to dimensionality reduction.
Left: DALL·E2, right: DALL·E3
Overall,With the support of ChatGPT, DALL·E3's language understanding ability is directly maxed out, and it's almost impossible to win.
The upgraded version of ChatGPT will not only not lose key information points, but even if you only type a few keywords here and there, it can help you automatically complete the description and then let DALL·E3 generate the picture.
OpenAI has grasped the essence of the "cultural desert" of contemporary netizens (dog head).
Of course, the integration of DALL·E3 and ChatGPT is not just as simple as being able to understand human speech better, they will also produce some wonderful sparks.
For example, the upgraded version of ChatGPT also has context understanding capabilities in drawing, and can even be used directly as a productivity tool.
To see how powerful it is specifically, the official website of OpenAI provides a demonstration video. To be honest, after watching it, Shichao was worried about the job of an illustrator.
First, let ChatGPT generate a super sunflower hedgehog. It will give you four pictures. After you choose the one you like the most, you can proceed to the next step of the conversation.
Then name the hedgehog Larry, and let ChatGPT generate a few more photos of it.
Next, let’s increase the difficulty and create a scene to show Larry’s home.
This one can directly show the strength of DALL·E3+ChatGPT. Not only does Larry's appearance remain the same (this may have changed for other AIs), but the mailbox at the door also has the name "LARRY" written on it.
In addition, describing Larry's characteristics, using pictures to show his love, and even making Larry's peripherals and designing a few stickers are all easy for ChatGPT.
Finally, let it organize a bedtime story and an ending pose. ChatGPT is also at your fingertips. u1s1 I was really shocked by this silky dialogue...
Seeing this, Shichao suddenly got a new inspiration. Afterwards, the article we wrote could be thrown directly to ChatGPT and let it extract the keywords to make the cover.
If you have any special requirements, you can tell ChatGPT directly. You can also throw reference pictures to it. Fishing skills +1+1...
Closer to home, the above-mentioned effects are only unilaterally demonstrated by OpenAI. How it will actually be used will not be known until it goes online in October. It is estimated that there will be a wave of hidden techniques developed by netizens by then. Just wait and see.
In addition, when it comes to AI painting, there is still a common topic that cannot be avoided: copyright issues.
OpenAI still maintains its previous position.Just like the second version, pictures generated with DALL·E3 can be used without permission and can be used commercially.
However, having learned too much from the past, OpenAI is a little clever this time, saying that artists can choose to refuse their works to be fed to DALL·E, as long as they fill out a form.
Although this somewhat means "not to refuse is to acquiesce", but compared to before, at least the artists are no longer so passive...
DALL·E also has countermeasures against the previous AI paintings on the Internet that invaded the privacy of public figures and other biased issues:In actual use, ChatGPT will directly reject requests with the name of a public figure in the prompt.
In other words, we probably won’t see fun pictures like this in ChatGPT...
And they also formed a "red team" to evaluate and reduce risks that may arise from the model at various stages.
Finally, OpenAI also stated on its official website that they are working on a tool to identify AI drawings, which can determine which pictures were generated by DALL·E3.
(I just hope it won’t be like the previous AI text recognition tool, because it was useless and died halfway...
In general, ChatGPT with Vincent graph function is enough to stir up a wave of enthusiasm in the AI circle, and this wave is the integrated upgraded version of DALL·E3, which is hard not to be exciting. Anyway, Shichao can't wait to try it out.
But some people are happy and some are worried. This wave of OpenAI has once again left its peers by a long way. After the October update, it is estimated that many AI startups will be crushed under the wheels of ChatGPT...