5 ms·
in terms of ai, this uses latent diffusion models: smaller model, does not use clip, and it was trained on noisy data.
by dvrp 5y ago
in terms of ai, this uses latent diffusion models: smaller model, does not use clip, and it was trained on noisy data.
- legaloslotr 5y agoWhy invent your own models when DALL-E exists? Is it supposed to be better/more-accurate?
- dvrp 5y agoWe are using already existing models based on open-source research. We have other models that OpenAI does not offer (e.g. super resolution) but they’re not in the app. The reality though is that DALL-E 2 is still not open for access; open-source models are. If you know how to get API access—besides waitlist of course—please let me know!