Prompts are comprised of two components: the prompt and the negative-prompt. The prompt allows you to guide generation towards a set of concepts. In contrast, the negative-prompt allows you guide the bot away from a set of concepts. Both the prompt and negative-prompt work best when provided by a series of comma-separated phrases or words. The words at the start of each list have a greater weight, so make sure to put key concepts first.
Example: prompt: female elven monk, black clothing, strong, assassin negative-prompt: bald, asian, large breasts, weapon, staff
You may wish to make use of the /prompt
feature of the bot, which can take a body of text (story/description/etc) and constructs the prompt component for you.
Example (input): There was no hope, no glimmering star of dissension, no whispered insurrections. There was only Seraphina, her dark reign engraved in the millennia to come. Her victory was the empire's perpetual night, and her rule the unending eclipse. Beneath the cloak of the authoritarian empress, the cosmos was plunged into a darkness so profound it echoed the distance between stars: endless, infinite, and utterly oppressive.
Example (output): Female Emperor, sinister, draped in darkness, infinite expanse of the cosmos, endless oppression, a reign written in the stars, oppressive eclipse, celestial, dark empire
It's time to generate your images by using /img generate
. I recommend choosing a gen-style which helps to guide the AI towards a particular style or genre. It doesn't always improve the image, but it may help to produce stylistically consistent images.
Example: /img generate prompt: female elven monk, black clothing, strong, assassin negative-prompt: bald, asian, large breasts, weapon, staff
Disabled | DungeonsAndDragons | Warhammer 40k |
---|---|---|
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
Once you've found an image you like, you may wish to create variations of that image. To do this, you may use the /img variations
command. This command requires the original file generated by /img generate
or /img variations
(including the same file name). You can provide a closeness value which determines how closely the generation will adhere to the original image.
Very closely | Closely | Standard | Little | Very little |
---|---|---|---|---|
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
![]() |
You may upscale the image by using /img upscale
, which will increases the resolution of the image to 4k.
Upscaled |
---|
![]() |
The bot is capable of producing nudity. I have introduced some safeguards, but they are not perfect. Please only generate images that are suitable for your current Discord server. Presuming this is on the Arelith public Discord, that means PG-13 only.
- Accidentally nudity can sometimes happen: if it does, I recommend informing a moderator so they can remove it, and tweaking your prompt accordingly (hint: you can add naked, nude, breasts, etc to negative-prompt to make this less likely).
- If you intentionally bait the AI to produce booba then it will be treated by the moderators as if you were sharing booba from your degenerate booba stash. In other words, you'll get in trouble, so don't do it.
The bot is currently hosted on my GPU (when it is available) and on the cloud (when it is not). Depending on usage, I may change the amount of images that are generated for every command.
There are a bunch of advanced features! I won't explain them all, yet, but here are some examples:
/img generate
takes inspiration, which takes an initial image to generate a new one based off, and seed, which you can use to get deterministic results, and landscape, which allows you to generate 768x512 images./img variations
takes prompt-guidance and negative-prompt-guidance, which can help to make subtle tweaks to the image, for example changing hair or eye colour.