Today, OpenAI released the DALLE-2 Beta. DALLE-2 is an AI-powered content generation tool that can help you create unique, high quality images based on textual input. The AI is powered by GPT-3 and created by Open AI.
OpenAI is a research company founded in late 2015 with the goal of advancing artificial intelligence in the interests of humanity as a whole. The company is backed by some of the biggest names in tech, including investor and Tesla CEO Elon Musk and co-founder of PayPal Peter Thiel. OpenAI has made significant advances in artificial intelligence since its inception, most notably creating the world's first successful general artificial intelligence AI picture bot, DALLE-2.
The History of DALL-E
DALLE-2 was initially released by OpenAI in early 2021 as a way to generate images from textual input. The AI is powered by GPT-3, which is the world's most powerful artificial intelligence platform. The bot was first trained on a dataset of 12 billion parameters, and then fine-tuned on a smaller dataset of 3.5 billion parameters. Vox goes over a brief overview of the product pretty well here.
What Can DALL-E do?
The AI is nothing short of amazing, it can take any type of textual input and turn it into a corresponding image. For example, you could simply enter "a dog" and DALLE-2 will generate a random image of a dog. You can also get WAY more specific with your requests, like asking for "a black dog walking through the streets of manhattan while it's raining, photorealistic style" or "sunset off a snowy rocky mountain with a pretty sky in 4k" The possibilities are endless.
How to Use DALL-E
Once you have access to the software, visit the OpenAI Labs page. You have two options: enter a detailed prompt description, or upload a picture and have DALL-E do the magic work in the background.
Think of the generation input as a google search for any picture. If you can imagine the picture, DALL-E can generate it! For example, you could try something extremely detailed like: "A close up of a black cat's face with green eyes, looking directly at the camera." If you did, you'd get something like this:
You could try something more general like: "Cats." The output will be different each time you run it, and sometimes the results are better than others. The more details you provide, the closer the output image will be to what you're thinking.
We'd recommend as many details as you can think of to get the best results and if you're ever stuck, OpenAI has included a number of example inputs on their website to help get you started!
If you want to upload an image, the process is a bit different. For example, let's use this picture of a beach and the ocean. You can edit the image or generate variations of the same image. We'll go ahead and "erase" some of the picture in the editor and add a fireman in the sky. You'll get 4 images just like before and some of them come out decent!
How Much Does DALL-E 2 Cost?
OpenAI announced you can purchase 115 prompt credits for $15. Each entry will provide 4 pictures of the same prompt. This comes out to about 13 cents per prompt (or 3 cents per image). Previous users of the beta received 100 additional credits on top of 50 free monthly credits. Some users on reddit have expressed their frustration with the pricing, but it's still unclear how this will affect long term demand/use. OpenAI said they'll gather user feedback and "explore other options that will align with users’ creative processes."
Can I Use DALL-E Commercially?
As of the beta release, users can use DALLE-2 commercially. Users have full rights to use generated images both now and previously generated during the research preview phase in any way they like, including for sale and distribution. Images can be used on product packaging, in marketing collateral, or on websites and social media. Some users stated they will use their art for book illustrations, album covers, and logos.
Safety Concerns Regarding AI
There have been some concerns raised about the safety of using AI-generated images. However, OpenAI has addressed these concerns by releasing a strict set of content policy guidelines. You can't create images with realistic or political faces, or images that are sexually explicit, violent, or derogatory. You also can't use the AI to create images that are meant to deceive or mislead people.
OpenAI has a human content moderation team in conjunction with automatic monitoring tools to ensure that these guidelines are being followed.
Bias in DALL-E
Artificial intelligence prompts questions regarding bias and how artificial intelligence may perpetuate or create new forms of bias. OpenAI has openly stated they are aware of these issues and are working to mitigate bias in their AI. After implementing an update across their models, they resulted in a 12x increase in a diverse set of images.
How Do I Get Access?
Over the next few weeks, the company is extending invites to around 1 million people who sign up on their website. We're not sure how the rollout goes, but it seems like they're letting people in gradually.
How Long Does it Take to Get DALLE-2 Beta Access?
It truly varies based on how many people have signed up for the Beta Release. From what we've seen, it can take anywhere from a few weeks to upwards of months to gain access. It took us about 3 months to gain access, while we've heard others getting it after about a month.
What Do I Need to Use DALLE-2?
Once you have access to the Beta Release, all you need is a computer with an internet connection. No special hardware or software is required.
Subsidized Access Opportunities
Additionally, the company wants to expand to as many users and doesn't want financial barriers to get in the way. So, they're subsidizing credit costs for some artists who can't afford it. If you want to apply, fill out this form to be notified of more information as it comes out.
DALLE-2 is an AI-powered content generation tool that can help you create unique, high quality images based on textual input. The AI is powered by GPT-3 and created by OpenAI. You can use DALL-E commercially with full rights to use generated images in any way you like now that the beta has been released. We hope you gain access soon if you don't already have it so you can start creating some awesome content!
Are you excited about the release of DALLE-2? Will you be using it commercially or for personal use? Let us know your thoughts in the comments below!