August 19, 2022

High-quality image generation is now on AI Dungeon!

We have launched a new high-quality image model, Stable Diffusion, for inline generation of images in AI Dungeon. Stable Diffusion is an incredible new text-to-image AI model that will generate a unique image from a simple text prompt. We’re excited about Stable Diffusion because not only are the images high quality, but the generation times and costs are lower than other models. We can now enable custom image generation in AI Dungeon adventures, unlike the current version of AI Dungeon 2D which injects images into stories from a pre-generated cache.

Changes we’re making to support image generation

Since Stable Diffusion makes inline image generation possible, we are making a number of changes to AI Dungeon to support image generation.

First, we’re changing how images are displayed in adventures. Instead of the two-column layout on larger screens, images will now be displayed at a larger size and full width. For wide screens, they’ll be placed over a faded background to fill the horizontal space. This design makes the images feel like a more integral part of the adventure—and they are!

Second, we’re adding a new AI setting to change the image model you are using. You can select between the Pixel Art image cache and Stable Diffusion image generation. Another image cache is coming soon, with images generated using Stable Diffusion, DALL-E 2, and other image models.

Third, we’re adding a new action, “See”, that will join the Do, Say, and Story actions. When selecting the “See” action, what you type will be submitted as an image prompt. This action will respect the image model selection you’ve chosen in your settings. If you have Stable Diffusion Image Generation selected as your image model, the prompt will generate a unique image using Stable Diffusion. If you’ve selected one of the image caches, the prompt will search the cache for a relevant image to insert into your story.

Fourth, we’re improving the editing interactions for text. The dropdown menu for images and text will now include the ability to add an image to an existing action, generate/fetch another image, or remove an image. We’re also auto-saving any text edits when clicking away from the text input. The new interaction is a first step towards document-style editing in AI Dungeon. Save and cancel buttons have been removed from the dropdown menu.

Players will now control when images are displayed in their stories, and what those images depict. This control means the images will be far more meaningful for the stories players create.

Changes to the AI Dungeon Economy

We’ve known for a while that our economy system (Scales, Energy, and Actions) needed changes to accommodate the rapidly changing landscape of AI models. We’ve made several significant changes to accommodate Stable Diffusion and be more resilient to future changes and innovations.

Here are a few highlights about this new system:

“Image Credits” are being renamed “Credits,” and Credits will be the currency of AI Dungeon and Voyage. When we transition from “Image Credits” to “Credits,” we’re going to 20x the totals of Credits to allow us to charge less for Stable Diffusion. We will multiply players’ current Credit amounts, and our other image models will require 20+ Credits for each image generation instead of 1 Image Credit.

Members will now receive Credits each month instead of Scales. Scales can still be used to tip other players or creators. Credits can be purchased on all platforms (except for Steam). Scales will no longer be available for purchase, but players will be able to trade Credits for Scales.

NSFW Images

NSFW image creation will be allowed in AI Dungeon, and we expect the content policy for images to mirror our current policies with text. We are working with, the makers of Stable Diffusion, to ensure measures are in place to prevent generating certain types of content—mainly content depicting the sexual exploitation of children. These measures would apply to both published and unpublished stories.

Unpublished images, like unpublished stories, will not be moderated. Players will not be punished or banned for anything they create in unpublished stories using AI models. Published stories are still subject to our Community Guidelines, and these guidelines will still be upheld when image publishing is ready. Images won’t be included in published stories until we figure out the right experience and safeguards for sharing images publicly.

We know we won’t be able to anticipate every situation regarding AI-generated images. Because this is cutting-edge technology, there is no clear precedent for how (or if) AI-generated images should be governed. There are several unanswered questions about the use of AI images that all of us will be working through as AI image models become more accessible. We’ve learned from the past and will be actively communicating any changes or adjustments that need to be made to our content policies as we learn more about how players will use this powerful technology.

More to come

We’re enjoying an exciting time in AI, with new models and capabilities being introduced regularly. Our plan is to bring as many of those innovations into AI Dungeon as quickly we can.

Here's a preview of some of the incredible images generated in Alpha recently.