Give your community one shared image command instead of one member generating pictures for everybody. It runs the same Picasso IA models and the same credits as your own account, while your admins keep deciding who can use it, where, and what stays up.
Most servers already have one member who keeps generating pictures for everybody else and pasting them into the chat. A Discord AI image bot turns that favor into a shared tool: one image command that your community can use, running on Picasso IA models and the credits of a single account. Your admins still decide who can use it and where, and what gets posted stays under the same server rules as every other message.
A server is a place where people react to each other in real time, and images are some of the fastest reactions there are. A joke lands harder with a picture. A tabletop session gets a portrait of the villain the moment someone names him. When the picture is one message away, people use it the way they use any other emoji or sticker, which is to say constantly and without ceremony.
Today that usually runs through one generous person. They open an image tool, type what a friend asked for, download the result and upload it back to the channel. It works until that person is asleep, busy or tired of being the bottleneck. A shared command removes the middleman: the member who has the idea types it, and the picture arrives in the channel where the conversation is happening.
The Picasso IA connector for Discord is built around that habit. It links your account to your server, so a generation starts from a message and lands back in a channel. It uses the same catalog as the toolkit, 502 models in total, 212 of them text-to-image, and the same credit system, so there is no second bill and no second set of models to learn. You can browse the whole model catalog to see what a server could draw from.
A casual meme server and a painting community do not want the same thing from an image model, and the nice part of a catalog is that you do not have to pick once and live with it. These pairings are starting points to test in the toolkit before you decide what your server uses by default.
| Server type | Model to try first | What goes in the prompt | What to watch |
|---|---|---|---|
| Casual chat and memes | flux-schnell | One subject, one joke, one style word | Fast results can be loose on small details |
| Art and illustration community | seedream-4.5 | Medium, palette, mood and framing | Pick one house look or every post feels different |
| Tabletop and roleplay groups | nano-banana | A member's own character art plus one change | Only art the member has the right to use |
| Photographic looks | imagen-4 | Lens, light and setting | Fictional subjects only, never a real person |
| Event banners and announcements | ideogram-v3-turbo | Short phrase in quotes plus layout words | Proofread every letter before it goes up |
The column that matters most is the last one. Run the same prompt through two models, keep the one that understood it, and write that choice into your server notes so members know what to expect.
This is the journey from an empty server to the first picture in a channel. Before step one, spend ten minutes in the toolkit with flux-schnell, seedream-4.5 and nano-banana and write down two or three prompts your community would really send. Those become your test cases.
A shared command changes how prompts get written. In your own toolkit session you refine slowly, one image at a time. In a channel, ten people send ten different half-sentences, and the bot has no idea that the last message was a joke about a cat. What works is a short house habit that members can learn in one reading: subject first, then the style, then one detail that matters.
Give members a pinned example and they will copy its shape without being told. A prompt such as a tired wizard reading a menu in a tiny kitchen, watercolor, warm light is easy to remix and easy to read. The same wizard described in three paragraphs is the opposite: it takes longer to write, costs the same credits, and members stop experimenting after the second paste.
Keep the style words consistent if you want the channel to feel like one place. For lettering, say the exact words in quotes and keep them short. For people, describe fictional characters, and if a member wants their own face, they should upload their own photo and nobody else's.
Tip for admins: write the three best prompts from your test session into a pinned message. Members copy the shape of what they see, so the examples you pin teach the whole channel more than any rule does.
If your community also trades avatars, emoji or sticker packs, two companion pages cover those in depth: profile pictures and avatars and stickers and emoji packs. Tabletop groups will want Dungeons and Dragons character art.
An image command does not move any responsibility away from the people who run the server. Your admins and moderators still decide which channels it works in, which roles can use it, what is allowed and what gets removed. A picture posted by a bot is a picture posted in your server, so it falls under the same rules as everything else, and Discord's own terms and community guidelines still apply, so check their current version when you set things up.
Some habits make the job easier. Put the command in a dedicated channel at first, so a new tool does not spill into your serious channels. Write down in your rules what members may and may not ask for, in plain words, and say that generated pictures follow the same standards as any other post.
Faces and likenesses need one firm rule. Members can generate fictional characters and pictures of themselves from their own photos. They should never generate a real, identifiable person who has not agreed to it, whether a celebrity, a classmate or another member. Put that sentence in the channel description.
The key is the other responsibility. It belongs to the account that pays, so it should be known only to the admin who sets up the connection. If you ever suspect it has been shared, stop using it and create a new one from your account. Questions about connections and keys go to support.
A shared command is a convenience, and like any convenience it has edges. Here are the five that matter most before you announce it to the whole server:
None of these is a reason to skip the idea. They are the reasons to start in one channel, with one model and a person who checks the balance now and then.
No. The connection uses the same catalog as the toolkit, which has 502 models in total, and 212 of those are text-to-image. The catalog is open to browse on the all models page, and you can try any prompt in the toolkit before you decide what the server should use.
The account behind the key. Generations use the same credit system as the rest of Picasso IA, and the cost depends on the model and the settings, so there is no single price to quote. New accounts get free credits to start. Current plans live on the pricing page.
Your server admins and moderators do, the same way they decide everything else. They choose the channels where the command works, the roles that can use it and the rules about what may be asked for. The connector is a generator, not a moderator, so write your standards into the server rules and say that generated pictures are held to the same standards as any message. Check Discord's current terms and community guidelines when you set it up.
The idea is that you do not. You create a key in your account, an admin gives it to the connection and picks a channel, and members then send prompts as ordinary messages. The one thing that does need care is the key, which works like a password and should only be handled by the person who sets the connection up.
It depends on the community. For quick, casual rounds, flux-schnell is the sensible first test. For illustration and art-focused servers, seedream-4.5 is worth trying, and imagen-4 is worth trying when members want photographic looks. Run the same prompt through two of them in the toolkit, keep whichever understood it better, and tell members which model the server uses so their expectations match what they get.
Editing models such as nano-banana accept a reference image plus a plain instruction, so a member can attach a picture they own and ask for one change. That is useful for roleplay character art or for tidying up a profile picture. The rule to set is simple: members use their own images or fictional characters, and never someone else's face.
Treat the key like a password. Give it only to the connection you are setting up, do not paste it into a channel, a screenshot or a shared document, and keep it known to as few admins as possible. If you suspect it has been seen by someone who should not have it, stop using it and create a new one from your account.
Yes, and it is the best first step. The toolkit runs the same catalog today, so you can type the prompts your members are likely to send, compare models and see roughly what credits a generation uses with your chosen settings. Spend a short session there before you set anything up, and bring the best three prompts back to your server as pinned examples.
They should not. Faces and likenesses are limited to people who have agreed, such as a member using their own photo, and to fictional characters. A real, identifiable person who has not agreed, whether a celebrity, a classmate or another member, should be off limits, and the sentence belongs in your channel description and your server rules.
Generations cost credits, so when the account behind the key has none left, new requests cannot run until credits are added. That is another reason to give one admin the job of watching the balance and to keep the command in a few channels instead of everywhere. Plans and credit options are on the pricing page, and support can answer questions about your own balance.
Before you announce anything to the server, run your first three prompts in the toolkit, and send any question about connections to support.
Choose your language
