Uncensored and 'Crazy': A Look Inside OpenAI's Secret Image Generator

The unreleased generative imaging tool doesn't shy away from nudity and gore, and it caught the attention of MidJourney.

Images created with AI

OpenAI has been privately testing a new version of its generative artificial intelligence (AI) imaging model for the past few months, and early samples leaked by YouTube user MattVidPro show the model outperforming previous image generators.

Matt said in a preview posted on YouTube: "Very exciting, it blows the minds of anything we've seen before, it's crazy. Midjourney can't compete at this level, I don't even think Midjourney 6th edition can Compete at this level."

Don't expect to try it anytime soon, though. Access is extremely limited.

The unreleased model is likely an upgraded version of the DALL-E 2 and is being tested via an invited preview within ChatGPT-4. Only about 400 people around the world have access to this new OpenAI image generator, Matt said.

Although image samples are limited, the advanced skills of artificial intelligence are demonstrated. It produces sharp images by mimicking the lighting and reflections of real photographs, and the model reproduces painterly details down to visible brushstrokes. It also perfectly recreates brand names like "Snickers" and logos of well-known brands like Subway on the generated products, and achieves pretty good spelling in rendered text.

An example of an image generated by OpenAI's text-to-image. Source: Matt VidPro

While current image generators struggle to achieve coherent hands, these examples show realistic and properly proportioned hands with more convincing backgrounds than competing AI systems.

OpenAI apparently removed its safety filters to test the model's full potential. Users say it doesn't hesitate to generate violence and nudity. However, given OpenAI's stance on NSFW content, it's highly unlikely that an official public version will be released according to such standards.

Matt observed, “It’s OpenAI, so you can bet they’re going to shut it down like they did with everything before release.”

Some experts have criticized OpenAI for "simplifying" its models to avoid potential controversy. Some studies have even shown that OpenAI-trained ChatGPT has a strong political bias in its output.

Still, the consistent quality shown in the samples is a leap forward, and it underscores OpenAI's ongoing efforts to improve the capabilities of generative AI. The company may reveal more of its progress later this year, especially if the field of image recognition and generation contributes to the robustness of its star product: a multimodal GPT that can understand text, images, and drafts in a single prompt— 4.

For now, the technology is still limited to a closed beta with a very small number of users.

As models continue to improve, the lines between the artificial and the real become more blurred. While this excites many, concerns about misuse will persist, and building this technology responsibly remains a pressing challenge.