Prompts & Presets¶

Lumiverse builds the final image prompt from three things: the prompt mode you've chosen, any preset you've loaded, and the character / persona snippets bound to the active chat. This page covers all three and the prompt preview flow you can use to inspect the result before you generate.

Prompt Modes¶

Set the mode under Image Generation → Prompt Mode.

Scene tool¶

The default. A sidecar LLM reads the visible chat history and extracts a structured scene:

When Include Characters and Persona is enabled, the parser additionally returns visible character names, per-character appearance tags, composition subjects, shot framing, camera angle, and a list of composition rating tags. NovelAI uses these directly as character tags; the other providers fold them into the prose prompt.

Lumiverse caches the previous scene per chat and only generates a new image when at least Scene Change Sensitivity fields have changed. Turn on Ignore Scene Change Detection to bypass the cache and regenerate on every trigger.

Custom prompt¶

Your prompt is sent to the image provider verbatim. Standard macros ({{user}}, {{char}}, character lookups, etc.) and the character / persona snippet macros described below are still resolved.

Chat-aware custom¶

Your text becomes parser instructions, not the final prompt. A parser LLM reads the current chat context and rewrites it into an image prompt that follows your guidance.

For example, a parser instruction like:

Focus on the current pose, expressions, clothing, lighting, and room details. Use concise image-generation tags.

…will produce a different final prompt every reply, tracking what just happened in the chat. This is the best mode when you want both authorial control and scene awareness.

Picking the Parser LLM¶

Scene and Chat-aware Custom both need a parser LLM. By default Lumiverse uses the Council sidecar, but you can override it under Prompt Parser:

The parser is short-lived per request and only runs for Scene and Chat-aware Custom modes. Pure Custom prompts skip it entirely.

Prompt Presets¶

Custom and Chat-aware prompts can be saved as presets and reloaded later. Presets come in three kinds:

Use the Editing picker in the panel to switch which kind you're editing. Save, rename, and delete buttons sit underneath the editor.

Bindings¶

Character and persona presets aren't applied just because they exist — they have to be bound:

• Bind a character preset by editing it while a chat is open. The preset is bound to that chat's character. From then on, any chat with the same character will splice this snippet into {{character_prompt}}.
• Bind a persona preset by editing it with that persona active. The preset is bound to your user account against that persona.

A small banner under the editor confirms what the current preset is bound to. Bindings persist across sessions and are stored per user.

Resolution Order¶

When generation runs, Lumiverse resolves the prompt in this order:

1. The main preset (or the inline panel text if no preset is loaded) is taken as the template.
2. Any {{character_prompt}} / {{character_negative_prompt}} macro is replaced by the snippet from the character bound to the active chat — if one is bound.
3. Any {{persona_prompt}} / {{persona_negative_prompt}} macro is replaced by the snippet from the persona bound to your user — if one is bound.
4. Standard macros ({{user}}, {{char}}, etc.) are evaluated.
5. The result is sent to the image provider (Custom mode) or to the parser LLM (Chat-aware Custom mode).

If a macro is present but no binding exists, the macro expands to an empty string — your prompt still works, just without that snippet.

Example¶

Main preset (Custom mode):

masterpiece, best quality, {{character_prompt}}, {{persona_prompt}}, indoor lighting, cinematic composition

Negative:

worst quality, bad anatomy, {{character_negative_prompt}}

Character snippet (bound to "Aria"):

1girl, long red hair, leather jacket, gold earrings

Persona snippet (bound to your "Detective" persona):

middle-aged man, glasses, beige coat

When you chat as Detective with Aria, the prompt sent to the provider becomes:

masterpiece, best quality, 1girl, long red hair, leather jacket, gold earrings, middle-aged man, glasses, beige coat, indoor lighting, cinematic composition

Swap to a different persona or character mid-chat and the splices update automatically on the next generation.

Prompt Preview¶

Turn on Preview prompt before generating (under Scene Settings) to inspect the resolved prompt before the image provider is called.

When enabled, hitting Generate opens the Prompt Preview modal showing the final prompt and negative prompt — fully macro-resolved and (for Scene / Chat-aware modes) already passed through the parser LLM. From there you can:

• Edit the prompt and negative prompt directly.
• Generate — sends the (possibly edited) prompt straight to the image provider, skipping the parser this time.
• Cancel — discard and start over.

This is the easiest way to debug a misbehaving preset or to fine-tune the parser output for one specific scene without changing your saved instructions.

You can also preview without ever generating an image by setting the Output target to Preview only.

Generating¶

Generation runs cooperatively — a new request for the same chat aborts the in-flight one, so you can change your mind without waiting for the previous image to finish.

Tips¶