NichG
2021-08-06, 04:53 PM
Before we had the GPT-3 rules-text and location description thread. There's a new toy out there which generates images based on a text prompt, and the method has been refined by a number of artists and is starting to look very, very good. I've just started playing with this notebook (https://colab.research.google.com/drive/1gFn9u3oPOgsNzJWEFmdK-N9h_y65b8fj), which lets you run the thing in browser on a cloud GPU.
A couple of example outputs...
"a photo-realistic and beautiful painting of a pirate ship sailing the cloud-ways into a storm." (https://drive.google.com/file/d/1dTWXZejKm_YM7kdesC0OhE7usBg5sXjn/view?usp=sharing)
"a photo-realistic and beautiful painting of the soul-forge churning out yet another weapon for the damned to fight their eternal war" (https://drive.google.com/file/d/1qMlc6mMErVZ6CP5cBaJcmsVi1a_jEjci/view?usp=sharing)
"a photo-realistic and beautiful painting of a jungle filled with purple, glowing moths" (https://drive.google.com/file/d/1W1Y8L6B8jbFwKjGKnT8d1w4jLkZtb3CF/view?usp=sharing)
So, obviously not perfect. It does a 'jungle filled with purple' + 'moths' rather than purple, glowing moths for example. But still, if you tweaked the prompts to get what you wanted, I think you could probably illustrate a campaign in real-time with this.
A couple of example outputs...
"a photo-realistic and beautiful painting of a pirate ship sailing the cloud-ways into a storm." (https://drive.google.com/file/d/1dTWXZejKm_YM7kdesC0OhE7usBg5sXjn/view?usp=sharing)
"a photo-realistic and beautiful painting of the soul-forge churning out yet another weapon for the damned to fight their eternal war" (https://drive.google.com/file/d/1qMlc6mMErVZ6CP5cBaJcmsVi1a_jEjci/view?usp=sharing)
"a photo-realistic and beautiful painting of a jungle filled with purple, glowing moths" (https://drive.google.com/file/d/1W1Y8L6B8jbFwKjGKnT8d1w4jLkZtb3CF/view?usp=sharing)
So, obviously not perfect. It does a 'jungle filled with purple' + 'moths' rather than purple, glowing moths for example. But still, if you tweaked the prompts to get what you wanted, I think you could probably illustrate a campaign in real-time with this.