Picturing the world with Parti

When it comes to text-to-image models, they use a combination of language models and image generation to create illustrations and pictures from a simple text prompt. Parti is an autoregressive model that synthesizes images based on natural language prompts. It solves the inverse problem of image captioning, where given an image the task is to come up with a natural language description to describe the image.The Parti model takes text, the human language, as inputs, and produces images that are both visually realistic, and aligned well with the text inputs. In this episode of #MeetAGoogleResearcher, Drew Calcagno chats with Jiahui Yu and Yonghui Wu - two of the ground breaking researchers who have helped develop Parti. They dive into this autoregressive model, the personalities behind the people who made it, what makes Parti different from other text-to-image models, its challenges, and where the future lies for Parti and other models like it. Resources: Check out Parti → https://goo.gle/3Vk5wi2 Check out the “How AI creates photorealistic images from text” blog post → https://goo.gle/3CJQ5YW Follow Jiahui on Twitter → https://goo.gle/3VUGbvb Follow Drew on Twitter → https://goo.gle/3BTvKR8 Follow Drew on YouTube → https://goo.gle/3PaLrGo Chapters: 0:00 - Intro 0:50 - Researcher Intros 2:12 - Where are we with text-to-image models? 5:09 - What does Parti do? 7:14 - What fields have been applied to make Parti? 8:18 - How does Parti work? 9:13 - What did the first image of Parti look like? 11:53 - Do you think of yourselves as creatives? 15:01 - What are Parti’s strength and weaknesses? 17:07 - What are the different uses of Parti? 18:18 - Do we see these models helping other people? 20:16 - When will people be able to use these models? 21:39 - Recap Watch more: Watch more episodes of Meet A Google Researcher → https://goo.gle/MeetAGoogleResearcher Subscribe to the Google Research Channel → https://goo.gle/GoogleResearch #MeetAGoogleResearcher