Deep Daze

lastmod 2021-03-31created 2021-03-311 edits

Generating images from text.

Features

• A tool that uses OpenAI's CLIP and Siren to generate an image from a simple string
  • ex) imagine "a house in the forest"
• The default mode allows only up to 77 characters
• With the create_story option, an entire story/song/poem etc. can be generated as changing images

Examples

• Text: mist over green hills

Tips

Since OpenAI's Dall-E is closed source, there are attempts to implement it in the open

• DALL·E: Generating images from text https://news.hada.io/topic?id=3537
• DALL-E in Pytorch https://github.com/lucidrains/dalle-pytorch
• Big Sleep - text-to-image : CLIP + BigGAN, https://github.com/lucidrains/big-sleep

Other projects using OpenAI's CLIP:

• Searching inside YouTube videos with natural language https://news.hada.io/topic?id=3741

See also

• Deep Learning

Favorite site

• Github - deep-daze