Content
newsletter Newsletter

DALL-E gets “Outpainting” that can be used to enhance existing artwork. More than one million people already use OpenAI’s image AI.

OpenAI adds a new feature to DALL-E 2 for all users, which has been in test use for some time: With Outpainting, users can expand a photo or painting with simple text input.

For example, a picture of the Golden Gate Bridge could acquire a new narrative through the voice input “A UFO is hovering over the Golden Gate Bridge”. Outpainting takes into account shadows, reflections, and textures of the original painting to maintain its context, according to OpenAI.

Extending Historical Paintings

Outpainting can be particularly interesting when combined with classical paintings, as AI artist Denis Shiryaev demonstrated in early June. Shiryaev used outpainting on famous portraits and paintings by artists such as Leonardo da Vinci, Vasily Vereshchagin and Johannes Vermeer.

Ad
Support our independent, free-access reporting. Any contribution helps and secures our future. Support now:
Bank transfer
Mona Lisa stood in a river during her portrait to cool her feet – did you know that? | Image: DALL-E 2

OpenAI brings an outpainting example of Johannes Vermeer’s “Girl with the Pearl Earring”. The portrait becomes a drawing of the girl in a room. It is interesting that OpenAI cuts off the extended painting above the girl’s hand.

No hands, no problems. | Image: OpenAI / DALL-E 2

Shiryaev also expanded Vermeer’s well-known painting, but had issues with a distorted drawing of the hands. It is possible that this problem persists and that OpenAI intentionally omitted the hands here.

Outpainting is now available to all DALL-E users in the desktop browser.

DALL-E 2 has more than one million users

OpenAI also announces more than one million users for DALL-E 2. This makes DALL-E 2 probably OpenAI’s most successful commercial product to date, in any case the one that has generated the most attention.

According to OpenAI co-founder Sam Altman, DALL-E took only about 2.5 months to reach one million registrations (now its users). The well-known text AI GPT-3 reached this mark only after about two years.

Recommendation

It will be interesting to see whether and how OpenAI reacts to the changing market for image generation: For a few days, the open-source alternative Stable Diffusion has been available, which can be run locally free of charge and is not subject to any restrictions on text prompts. However, the handling is still a bit more complicated at the moment.

In any case, OpenAI cut the price of the text AI GPT-3 by two-thirds a few days ago. Here, too, more alternatives have come onto the market in recent months. The price reduction could also be an indication of an imminent launch of GPT-4.

On the other hand: Image generation with DALL-E 2 is already very cheap and is critically questioned by artists and designers.

Ad
Support our independent, free-access reporting. Any contribution helps and secures our future. Support now:
Bank transfer
Join our communit
Join the DECODER community on Discord, Reddit or Twitter - we can't wait to meet you.
Support our independent, free-access reporting. Any contribution helps and secures our future. Support now:
Bank transfer
Sources
Online journalist Matthias is the co-founder and publisher of THE DECODER. He believes that artificial intelligence will fundamentally change the relationship between humans and computers.
Join our communit
Join the DECODER community on Discord, Reddit or Twitter - we can't wait to meet you.