Our great sponsors
-
WorkOS
The modern identity platform for B2B SaaS. The APIs are flexible and easy-to-use, supporting authentication, user identity, and complex enterprise features like SSO and SCIM provisioning.
Dall-e was originally built by open-AI (closed source) and then recreated in a OS project Dalle-Mini (/Mega): https://github.com/borisdayma/dalle-mini
Dall-e Flow, another popular repo, uses Dall-e Mega (~9 GB, CLIP 3 GB, GLID-3 XL 6 GB and SwinIR 3 GB- all at once to optimize the workflow, which is why it requires 21 GB. This could be reduced if you free up resources, but will take more time (and work to do)
Spinning up your own AWS instances is def. possible and cheap, but might be a bit difficult with all the overhead. There are companies (like jina.ai that i mentioned previously) working on solutions to make it easy
Related posts
- How to Personalize Stable Diffusion for ALL the Things
- image generation API similar to Dall-E or Dall-E 2
- Who needs prompt2prompt anyway? SD 1.5 inpainting model with clipseg prompt for "hair" and various prompts for different hair colors
- Sequential token weighting invented by Birch-san@Github allows you to bypass the 77 token limit and use any amount of tokens you want, also allows you to sequentially alter an image
- I made a discord bot for artsy ML stuff - just finished integrating SD