Repository navigation
Images
Orbweaver can draw pictures into your chat, let you rework them, and keep a gallery for each of your characters. Drawing needs an image model; attaching your own images and video doesn't. The part I wanted most: you get to read the prompt before you pay for the picture.
Open Settings, then Connections, then Model roles, and point Image generation at a connection whose model can output images. It's optional. With nothing set, generating a picture fails with a "Couldn't generate the image." message. → Connect a model
Pictures are made on the image model of whoever asks for them. In a shared room that's the person who pressed Generate, even when it isn't the host.
There are three ways in, all from a chat:
- Type
/imagine. A Generate an image window opens. - Open the Message tools menu, and under Media choose Imagine — modes & preview.... It's the same window, started with whatever you'd typed.
- Under Media, choose Generate image from text. Your typed text goes out as the prompt, exactly as written, and the picture starts right away with no preview.
The window has five modes:
| Mode | What it draws | Shortcut after /imagine
|
|---|---|---|
| Free prompt | Exactly the words you type | Just type your prompt |
| Character | A full-body image of the character as they are in the story right now | you |
| Face | A close-up portrait of the character | face |
| Scene | The current moment of the story | scene |
| Background | The current location, with nobody in it | background |
Type /imagine scene and the window opens already on Scene.
In the four story modes, press Read the chat first. Orbweaver reads the recent chat, writes an image prompt, and drops it into the prompt box where you can edit it. Nothing is drawn until you press Generate. Reading the chat runs on your Utility model, and the window shows what that read cost once it's done, if your provider reports a cost.
If you press Generate in a story mode with an empty prompt, Orbweaver does both steps for you, and the window warns that this is two charges. Once there is text in the box, it's sent as written.
A picture usually takes between a few seconds and a minute. You can close the window, and the picture still lands in the chat when it's ready.
The picture posts into the chat as a message from you, with your prompt as its text if there is one. In a shared room everyone sees it. The file is yours and stays in your uploads.
If you own the first character in the room, the window shows an Add to (character)'s gallery box, switched on by default.
Some chat models can draw as well as write, Gemini's image models among them. Open your preset, go to Params, and switch on Reply pictures, and a model like that can answer with pictures in the chat beside its text. The hint under the switch says whether the model you're previewing can draw.
- It runs on your Chat model, not on Image generation, so it costs what that model charges for a reply with pictures.
- A text-only model ignores the switch, and its reply says it couldn't add pictures.
- The picture becomes part of the message, so it stays through edits, forks and exports.
Click a picture in the chat to open it large. For pictures you generated, the viewer's Details show the prompt, the model, the mode, an Edited tag if it was reworked, and what it cost when your provider reports one. A picture you didn't generate, such as someone else's or one you attached, shows No generation details recorded for this image. Videos and pictures from outside links open in a plain zoom view with no details.
Restoring an Orbweaver export can bring these recorded details back with the picture. Keep Image generation history selected when you export. Missing counts and prices stay unknown; restore does not invent them. → Back up and restore
- Edit image. Say how it should change, like "make it night" or "turn the coat red", and press Generate edit. You get a new picture, opened in the viewer, and the original stays as it was. The edit isn't posted to the chat. It works on pictures you own, whether generated or attached, and needs an image model that can edit images. If yours can't, the edit fails with a "Couldn't edit the image." message. Back to the image takes you out of the edit without changing anything.
- Set as background. Makes this picture the room's background. Only the host can do it, and only with a picture the host owns, so a picture another member generated won't work. You can change it back from the This chat panel, under Host controls, then Background.
Open Settings, then Chat behavior, then Image prompts. There's one box each for Character portrait, Character face, Scene and Background, plus the two "from avatar" versions used when a portrait is described from the character's picture.
The negative prompt lives in the same Chat behavior page, under Model-facing prose, as Negative-prompt base.
Each character you own has a gallery. Open it from Chat options (it's listed as the character's name plus "gallery"), by typing /gallery in the chat, or from the character's actions menu in your library. In a chat with several of your characters, Character galleries lets you pick one.
In the gallery you can drop or pick files to upload, add images you already own with Add images, open any picture large, remove one (it asks first), sort by newest or oldest, and, when you opened it from a chat, switch between Everywhere and This chat to see only what was made in that chat.
It's your own gallery of your own characters. /gallery opens the gallery of the first character in the chat, and says so if you don't own that one. Chat options lists every character you own.
Rules can draw for you while you play. The host adds them from the This chat panel, under Host controls, then Rules, using Add a rule. A new rule starts switched off, and Test runs it without spending anything. → Automation
- Illustrate the scene. At most once every so many beats, draws either the scene or the place, and posts it into the room. It reads the chat on the host's Utility model, then draws on the host's image model.
- Illustrate on lore reveal. Draws the scene when enough world book entries fire at once. → Lorebooks
- Auto-set scene background. As the story moves, swaps the room's background for one from the host's background library that fits. It runs on the host's Utility model, and does nothing if the library is empty. → Themes
- Living library. Found under Settings, then Automation, then Library-wide rules. When a character's card changes, a fresh portrait is made and saved with your images. Add it to the character's gallery with Add images. It needs a card with a picture: the Utility model describes it (so it must accept images), and the image model draws.
Put your own pictures into a message by choosing Attach images & video in the Message tools menu, dragging files onto the message box, or pasting from the clipboard. PNG, JPG, WEBP, GIF, MP4 and WEBM work. A file over the size limit is refused with a message that names the limit.
For rooms you host, Settings, then Chat behavior, then Attachments has Attachment quality, which controls the fidelity of what is sent to the model.
- Orbweaver shows the price of reading the chat after you've done it, and the price of each picture in its viewer. It doesn't quote the picture's price before you generate.
- Details show only for pictures you generated yourself. Edit works on pictures you own. Set as background is host-only.
- Galleries belong to characters you own.