Eimi Wiki#
Eimi is an unofficial Discord bot for running AI image diffusion models. It is a hobby project, maintained by one person, running on one graphics card in their home.
Its job is to be a convenient way to run diffusion models from Discord. Right now that means 53 image models, 91 LoRAs, 8 upscalers and 14 editing models, plus some video, music and editing features that grew out of the same setup.
Read this before relying on anything here
Nothing about uptime, speed, image quality, or which models exist is promised. The bot goes down, models get swapped out, and old images stop loading. This wiki is written by hand against a moving target and may be out of date. See About This Wiki.
Start here#
| If you want to... | Go to |
|---|---|
| Make your first image | Getting Started |
| Know what you are signing up for | About This Wiki |
| See everything the bot can do | Command Reference |
| Pick the right model | Models |
| Write prompts that work | Prompting Tips |
| Understand the buttons under an image | The Result Card |
| Know why your job is still waiting | The Queue |
The commands#
Generate: /imagine · /turbo · /render · /compose
Change an image: /edit · /upscale · restyle · tweak
Everything else: /settings · /history · /manage · describe · waldo
How it works#
You submit a command. The job joins a queue. Eimi builds a ComfyUI workflow from your parameters, runs it on the GPU, and posts the result back with a row of buttons for what to do next -- re-roll, remix, upscale, edit, or inspect exactly how it was made.
Four things are worth knowing before you start:
- One job runs at a time, strictly first-come-first-served. There is one GPU, so there is one job. Admins and VIPs do not skip the line. See The Queue.
- A progress bar that stops in the low 90s is normal. It is built to do that. See Progress.
- Your settings follow you across servers. They are per-user, not per-server. See
/settings. - Save anything you want to keep. Old images stop loading eventually, and nothing here is a backup.
Reference#
| Section | Contents |
|---|---|
| Models | All 53 models, by family, with defaults, strengths and weaknesses |
| LoRAs | All 91 LoRAs and how to apply them |
| Guides | Prompting, enhancement, characters, parameters, upscaling, styles |
| Interface | Result cards, remix, detailer, details panel, delivery |
| Behavior | Access control, queue rules, failure handling |