Research OpenAI Vision Proprietary

OpenAI DALL-E

The original image-generating AI — multimodal GPT-3 that shocked the world

Released 2021-01-05 · ~12B

Overview

Neural network that creates images from text captions

Specifications

DeveloperOpenAI
Release date2021-01-05
Model typeVision
Parameters~12B
ArchitectureTransformer (discrete VAE + GPT-3)
LicenseProprietary
Input modalitiestext
Output modalitiesimage

Benchmarks

Benchmark DALL-E
FID 17.89
IS 17.9
CLIP Score 32.5

Availability

Research Only

Official references