Openai is integrated directly to the image generation capabilities of Surah Generation in the Chat GPT, which begins today – this feature has been named as “Pictures in Chat GPT”. While the Surah was the first Accessible through a separate websiteConsumers can now use it to create photos inside Chat GPT.
Surah was announced as an AI -powered video generator, but this preliminary release is fully focused on image creation and will be available in Chat GPT Plus, Pro, Team, and Free Subscription Tires. Spokeswoman Taya Christian said the free level of use is like Del E. StuffyBut he added that he had “had no special number to share” and “these could change over time based on demand.” FIRST Chat GPT Normal QuestionnaireFree users were able to produce three images daily with “Dell · E3. As far as Dale E’s fate is concerned, Christian said” fans “will still have access to” customs GPT. “
“This model is a step change from the previous models,” said Gabriel Goh, the research lead. StuffyHe added that the team has used a model GPT-4O “Ominymodel”-or a model that can produce any kind of data such as text, image, audio, and video-the basis of this repetition of its repetition.
The improvement that Goh has made includes “bound”, which means to what extent the AI image generator maintains the right relationship between attributes and items. For example, a model that is poorly binding can get a gesture for a red triangle in addition to the blue star and makes a red star and can not make a triangle. Cowah said that most image models struggle with it, when colors and shapes are often found to offer multiple items – usually around 5 to 8. He says that the new image of Surah can be properly tied to the attributes for 15 to 20 items, which represents significant improvement in accuracy and reliable.
Users will also feel improved in text rendering, which makes it easier to produce integrated text without typos on an image (in existing tools, you will often see this text. Is very easily messed up) Goh said correcting the text was an important challenge. If small titles or text elements have types or errors, the whole image can be unusable.
“It was like a repetition process that took many months to get right,” said Guh. Although not perfect, he said the team reached the point where the quality of the text is permanently usable (where it causes the error is a really small text). “It has only improved for several months.”
This system uses an automatic point of view-to prepare images with the right and top to bottom sequence, instead of being written in the text-instead of the techniques used by the more image generators (such as Dall-E), rather than the technique that configure the entire icon together. GOH has speculated that it may be a technical difference that Surah has better text rendering and binding capabilities.
In a briefing before the feature launch, the team showed numerous examples that showcase the system’s abilities, including scientific arigram such as Newton’s printed experience such as labeled components, multi -panel humor with permanent characters and text bubbles, with proper texts and text bubbles. He also highlighted practical applications such as stickers, restaurant menos, and transparent background images for logo.
Chat GPT Multi -Model Product Lead Jackie Shannon explained, “If I go to pull an image, I do this with my skill limit … but also with all the knowledge I have developed.” “The model brings global knowledge to equality, so when you ask for a picture of Newton’s proliferation experience, you don’t have to tell what it is to bring back the picture.”
The new system takes longer to produce images than before, though the Openi suggests that this is a valuable trade. “Although we certainly have the capacity to improve delayed … The quality, ability, knowledge of the world, the world’s knowledge, actually adds for the extra seconds that they will just wait,” Shannon said.
When asked about safety measures – Taylor Swift’s notorious nude fax, using the Microsoft model, the ability to present Zai’s Kamil Harris with a gun, and Google Gemini’s ability to remove watermarks, emphasized that the system has strongly emphasized that the system has strongly emphasized that the system has been strong. Shannon said the device prevents the removal of the watermark, prevents the breed of sex deep faxes, and denies CSAM generation requests.
Open AI’s new image generation system does not contain visual water marks or indicators that show the images are AI-influx. However, Shannon explained that “all of our created images will be included Standard C2PA metad data There will also be some internal tooling to be able to view the pictures as Openi has created to mark this image. “
Shannon added, “Finally, no system is perfect for this kind of thing, but we are permanently improving our safety measures and we think of it as a starting point.” “One thing that is true about all the pictures made from Chat GPT is that the user owns them and are free to use them within the limits of our use policies.”
