Tell me what to draw!
DrawGPT is a an AI art project that uses the GPT-3 large language model by OpenAI to generate new images from text prompts.
This AI was not trained on any images, only written words.
Using Javacript and HTML programming the AI is able to express itself visually through words by writing code to draw and create unique images. This is the same AI powering the very popular ChatGPT large language model AI.
This uses words to draw pictures. How does this even know what an image is? It only knows words.
Its just very strange compared to many of the other ways artificial intelligence is being used to generated images right now.
This was never trained on any art, or even on any images, everything this AI knows about the world it read in a book.
This makes it a novel and unique way to use AI to create art. It is a bit regressive compared to modern image diffuser models but that is also what makes it fun. It also provides some truly hilarious outputs because the AI has no sense of scale, color, shape, or really anything visual so it makes hilarious things. It also makes beautiful things too, go figure.
DrawGPT can draw anything you can think of and more! Just type your text prompt in to the textbox and see what the AI gives you! Seriously, you can get GPT to draw just about anything for you that you can type in the box.
DrawGPT is runs on an AI that has never see an image in its life!
GPT-3 only knows words. That makes its understanding of the visual world extremely unusual. Typically this AI is not able to draw images.
This method of drawing images using raw code is not a great way to draw complex images like a photograph detailed digital illustation. This makes the images that DrawGPT outputs very simple.
You can see all the commands that DrawGPT uses to draw images in the Commands section.
Yes and no. Same same but different.
ChatGPT runs on the same model that this project uses, but it is a different instance of the model. This means that the AI is not precisly the same but it is the same AI large language model that ChatGPT is using.
What is the difference? ChatGPT is specifically wired up to be conversational and track a conversation thread across multiple user prompts.
In comparison DrawGPT does not remember things from prompt to prompt, each image is a unique image that does not reference any of the images or prompts previously supplied.
Only programmers, and we know they don't count!
Just kidding, I am a programmer.
You can read more about the training sets that OpenAI uses on their website but basically this AI is trained on a huge corpus of text that was cut off around mid-2021.
Once more for the violent weirdos in the back:
This not an image AI it only uses written words.
You can do what you want it's your party.
We humbly ask that you backlink to DrawGPT if you do use our images in any promotion or commercial ways, but it is not required.
Yes! You can use the images for commercial purposes! And so can DrawGPT.
Images are made public during submission.
Depending on the situation the prompts themselves are stored internally for research purposes but given that people are morons and put things in there like personal information it is unlikely DrawGPT would ever be able to release the raw prompts publicly.
Employees at OpenAI and DrawGPT have access to any prompts you submit.
DO NOT SUBMIT PERSONAL INFORMATION.
This works because add in a lot of tokens to the prompt you supply until it forces GPT-3 to output code in a way that I want.
You can copy the code, create a 500x500 canvas, and run the code on your own to regenerate the image.