BotDesign - dcsan/text2imageComp GitHub Wiki
Bot Design
For community-driven and social multiplayer experiences, a bot is a great choice. Everyone can see the work in realtime from each other and instantly chat about it.
But these type of bots, and the workflow need careful consideration to be successful.
workflow
Let the users see options first, then drill down in detail on the one they want.
This will also allow act as a valuable signal on 'voting' which images are best.
placeholder image, then stream
To act as visual confirmation, drop a placeholder image to the user first, and then stream the real image.
This reduces the feeling that the bot is slow as something is always updating.
use slash commands
Currently the bot uses text input, it would be better to accept slash commands.
In future discord bots will be prevented from eavesdropping a catch all on conversations, especially in larger servers.
Prompts also are clearer what is required, and there's a menu so they're easier to find.
Prompt editing
Allow gradual editing and addition to the prompts, rather than retyping everything each time.
Prompt Breakdown
We could provide initial analysis of the prompt, for example the semantic chunking, and any related images that are found in the LAION or other datasets