What is currently the best open source alternative to OpenAI's DALL-E 2(Text-to-Image Generation)?
2022-04-26 23:49:43 UTC
DALL-E 2 is an advanced text-to-image generator, which seems to work great. The problem is that it's not open sourced, and there's a wait list to get access to the previous version.
I've seen some open source alternatives, but nothing that comes close to it. I wonder if there's a nice open source project out there that I'm missing.
Created971d
971dReplies
1
2022-04-27 05:51:56 UTC
<p>short answer: no</p><p><br></p><p>long answer: DALL-E 2 is derivative of the popular GPT-3 model that openAI invented. Although instead of open sourcing it openAI decided they can make a lot of money by keeping it closed source. Soon they signed a deal with Microsoft and gave them exclusive access to GPT-3.</p><p><br></p><p>GPT-3 is state of the art text model, it is also the base model of DALL-E 2. So without access to GPT-3 its really hard to build a model as good as DALL-E 2.</p><p><br></p><p>There open source alternatives to GPT-3 though. These are:</p><ul><li>BERT: Text model invented by Google | <a href="https://github.com/google-research/bert" target="_blank">https://github.com/google-research/bert</a></li><li>GPT-J: Light version of GPT-3 | <a href="https://github.com/TheProtaganist/gpt-j" target="_blank">https://github.com/TheProtaganist/gpt-j</a></li><li>GPT-Neo: Open source attempt at GPT-3 | <a href="https://github.com/EleutherAI/gpt-neo" target="_blank">https://github.com/EleutherAI/gpt-neo</a></li></ul><p><br></p><p>If we had to build an open source DALL-E 2, we would have to base it off any of the above alternatives. Which currently no one has done.</p>
Post your answer