FAQ#
Common questions and the surprising answers.
Eimi is a hobby bot run by one person on one graphics card. A lot of the answers below come back to that. See About This Wiki.
The bot itself#
Is this an official bot? Who runs it?
No. It is an unofficial hobby project maintained by one person on their own hardware, in their own time, at their own cost. There is no company, no team and no support desk.
Is it free? Is there a paid tier?
It is free. VIP is not something you buy -- it is a flag the maintainer sets, and it raises some caps rather than making anything faster in the queue.
The bot is offline. When will it be back?
There is no status page and no schedule. It goes down for crashes, restarts, and whenever the machine is needed for something else. It usually comes back on its own. Waiting is normally the fix.
Can I rely on this for anything important?
No. Nothing about uptime, speed, quality, model availability or data retention is promised. If you need any of those, this is the wrong tool. See About This Wiki.
Are my prompts and images private?
They are stored in a database on a personal machine, and admins can view job history including prompts. Do not send anything you would mind the maintainer seeing.
Why did my favourite model disappear?
Models get added, swapped and removed whenever the maintainer feels like it, without notice. The lists in this wiki are a snapshot and drift out of date.
Is this wiki accurate?
It was accurate when it was written, against source code at one moment. There is no release process that keeps it in sync. Where the wiki and the bot disagree, believe the bot -- Discord's own command autocomplete is always current.
Jobs and the queue#
My job has said 🕒 Queued for a very long time. Is it broken?
Possibly not. One job runs at a time, so a busy queue means real waiting. But if it never starts at all, the bot may be in lockdown, and lockdown produces no message -- jobs simply sit queued forever. Ask an admin whether lockdown is on. See The Queue.
The progress bar stopped at 95%.
That is intentional. The bar caps at a random point between 93% and 97%, then a finishing animation ticks it to 99% before the result arrives. It also jitters on purpose. Neither means it is stuck. See Progress.
Where is my queue position?
It is not shown. The bot tracks position internally but never displays it in Discord.
Do VIPs and admins skip the queue?
No. The queue is strictly first-come-first-served with no priority tiers at all. What VIP and admin status grant is a higher cap on how many jobs you can have queued at once, and more GPU power -- never a better position.
How long will my job take?
No estimate appears on the job card. The bot's Discord status shows an aggregate estimate for everything currently queued.
The bot restarted while my job was running.
Wait rather than resubmitting. Restart recovery will either deliver the image anyway (marked as recovered), re-attach and resume tracking at the right percentage, or tell you the job was lost. See When Things Fail.
Prompts#
I turned on enhancement but my prompt is not being enhanced.
Enhancement failures are completely silent to you. A provider that fails three times in a row gets skipped entirely for a while. Also check that your ai_features setting is not on its default of Disabled. See AI Enhance.
Does typing a character's name load that character's LoRA?
No. Character matching replaces the name in your prompt with the correct tags. It never loads a LoRA automatically. If you want a character LoRA, add it yourself. See Character Matching.
Why does my prompt come out completely rewritten on some models?
Ideogram 4, ERNIE Image and Mage-Flow enable prompt enhancement by default, and Ideogram 4 force-enables it regardless of your setting. Use raw: true to bypass all prompt processing.
Why does the same prompt work so differently on two models?
Because the families want different things. Tag lists suit Illustrious, NoobAI and Pony; flowing prose suits Flux and Krea; Z-Image wants 150-250 words. See Prompting Tips.
Parameters#
I set steps to 100 but the details say something else.
Values outside a model's accepted range are clamped, not rejected. The range is derived per-model from its recommended values, so the ceiling differs between models. See Parameters.
I changed CFG and nothing happened.
Distilled and lightning models run at CFG 1 and ignore the setting.
My aspect ratio produced a size I did not ask for.
W:H is treated as a ratio and fitted to the model's pixel budget. WxH is a literal size and bypasses the budget, capped at 4096 on the long side. Some models also have their own aspect ratio tables that win over both.
Images and results#
Can other people press the buttons on my image?
Yes, apart from 🗑️ Delete and 📤 Post. Whoever clicks is the one charged for the job and the one whose saved settings apply. See The Result Card.
Why does my image look more compressed than I expected?
Eimi binary-searches WebP quality to fit your server's upload limit. Servers with a higher boost tier get higher quality automatically, because the target scales with the server's own limit. See Delivery.
My big upscale arrived small, with a 📥 button.
Discord will not preview images beyond 4096px per side or 89 megapixels, so you get a downscaled preview. The 📥 button serves the full-resolution original.
The ⬆️ upscale button disappeared.
You have reached the 8x ceiling. Detailer passes and nano 2K/4K jobs also count toward that ceiling.
The ✨ Detailer options are missing.
Detail Fixes and Hi-Res Fix are hidden at 4x and above to avoid running out of VRAM. The detailer also refuses to run at all on Z-Image, Lumina2, Neta-Lumina and Ideogram 4.
Are NSFW results blurred or spoilered?
No. The nsfw setting changes how prompts are handled -- it does not blur, spoiler, or gate anything. Nothing in the bot spoilers generation output.
Where are my old images?
/history. It is the only history surface -- there is no /stats, /favorites or /profile command.
Editing#
Can I outpaint / extend an image beyond its borders?
No. Outpainting is not supported. Inpainting, object removal, instruction edits, style transfer and multi-image composition all are. See /edit.
Why can I not select the Nano Banana or GPT models?
Those 5 edit models are VIP-only. Every image generation model, by contrast, is available to everyone.
Settings and servers#
Can I save presets or profiles?
No. There is one settings profile per user, and it applies across every server.
I just got my server delegated and now nobody can use the bot.
New delegations start in whitelist mode with an empty allow list, so only you have access until you add people. Use /manage to add users or switch to blacklist mode.
Can I set different defaults for my server?
No. Per-server configuration covers access control only. Generation settings are per-user.
I added the bot but a new command is not appearing.
Slash commands are not re-synced on every restart, so newly added commands can take a while to show up.
Bot behavior#
Does the bot respond to normal messages or mentions?
No. It has no message handlers and no auto-responses. Slash commands, right-click menus and result buttons are the only ways to interact with it.
Why are my results private when my friend's are public?
That depends on how each of you installed the bot. See Access and Visibility.