Preview leaks of OpenAI’s unreleased artificial intelligence image generator quickly gained traction due to its advanced skills that could outperform some of the leading AI image generators currently on the market.
The upcoming generative AI imaging model is currently being tested and most likely be an upgrade of DALL-E 2, OpenAI’s latest well performing AI image creator. While DALL-E 2 is a viable source of creating some awesome pictures, it often fails in comparison to Midjourney as it isn’t good at manipulating text & fails to create the realism you often can recreate with Midjourney.
Surprisingly,many people have been absolutely shocked when testing the new generator: “I promise you have not seen any AI image generations this good. Midjourney cannot compete at this level,” MattVidPro said on his recently uploaded YouTube video.
In his video, he was very impressed by the sneak peak samples saying that the generated images are extremely detailed, cohesive, and people have ‘no idea how powerful this thing is.’
The image samples showcase the AI’s advanced skills that even the tiniest brush strokes of an AI generated painting is visible to the naked eye. Shockingly, explicit images are readily generated without hesitation – nudity, violence, and gore.
“It'll do full frontal nudity, blood, and gore – it can do anything, this is the model everybody wants access to,’” Matt emphasized.
Seemingly, the safety filters were removed to test the generator’s full potential. However, OpenAI is known to water down NSFW content, so it is unlikely to be released on its official public version.
Even images of non-existing products can be AI-generated by this model, containing almost perfect details that exhibit its ability to display artistry, realism, and overall imagery. Not to mention, the correct spelling of texts too! Currently, DALL-E 2 fails to produce writing that is actually readable.
“You guys thought image generation was getting good…. Well get ready. Buckle up because it can get way more specific,” he added.
Compared to current AI image generators, this model does not have a hard time on making realistic, detailed, and properly proportional images. This will definitely be a tough competition to other AI systems due to its consistent performance – as seen in the previews we have so far.
OpenAI proves their drive for continuously improving their models and surpassing its capabilities upon new releases.
According to Matt, the upcoming model is being tested by around 400 people through an invite-only preview inside ChatGPT-4, for the time being.
Users also expressed being very amazed and impressed by the potential of the upcoming AI image generator.
While this model shows the advancing technology of AI, some people have raised concerns about possible misuse. There may come a time where it will be difficult to distinguish real from artificial images, especially as they keep getting better.
We have yet to see more of this unreleased model that will raise the bar high for the rising AI systems. Only time will tell what OpenAI has in store.