|
Welcome to the second episode of Generative AI - Short & Sweet! The online course about Generative AI (GAI) is released and, judging by all the feedback that we get, itโs a great way to get a solid GAI understanding. (grab discount here). ๐ฎ The GAI topic of the week is ..AI Image Generation (including Art Generation). Complex machine learning models that generate images often based on input text, other images or both. Recently, theyโve got a lot of attention as the modelsโ results are getting better and better. Almost monthly we can witness new astounding breakthroughs, making us wonder where could this lead to. As we have seen many different AI image generation models, we think these ones deserve special attention: Midjourney [By self-funded team called Midjourney; you can apply for their beta version for great art generation] DALL-E 2 [By OpenAI; image editing e.g. replacing a dog with a cat; realistic image generation e.g. Koala that makes a dunk; watch video for more] Nightcafe [By NightCafe Studio; compelling art generation; print to hang in your home or office] Deep Dream Generator [By Google; famous for running images backwards through differently pre-trained neural networks] Artbreeder [By Joel Simon; Tech - a StyleGAN + BigGAN - see open source version here; one can breed images and users can collaborate in the sense of merging their results] Parti AI [By Google; most imaginative results + understands to include the right text in the image itself; Tech - autoregressive]
๐ฅ Parti AI outperforms all other AI Image models by being able to accept large text prompts (a challenge for other models) and with its autoregressive approach (splits image generation into simpler sub-problems) generates images imaginatively in high-quality and high resolution. Most impressive: it is even capable of generating text on signs in its results, but only if the model is big enough (~20 Billion parameters). Watch Parti AIโs crazy results (read description below):
|