Artificial Intelligence
Out of context: Reply #168
- Started
- Last post
- 1,324 Responses
- neverscared0
https://www.unlimiteddreamco.xyz…
Created by Katherine Crowson, VQGAN+CLIP is a powerful text-to-image generation tool. Enter a prompt and it’ll create an image that matches the text.
VQGAN and CLIP are two separate neural networks that work together to create an image. CLIP is an image classifier, able to tell how well an image matches a text prompt, while VQGAN is an image generator, able to create images that look like other images. During the generation process, VQGAN creates an image and CLIP determines how well that image matches the prompt. Over many iterations, the result gets closer to the prompt until CLIP is satisfied that the prompt and the image are the same.
The system also needs a dataset – this is what the networks use to understand the prompt and create the images. Some datasets, such as ImageNet, are trained on millions of images and enable VQGAN to generate pretty much anything you ask.