Skip to content

Image Generation

Lumiverse generates scene illustrations, character shots, and chat attachments from your conversations. It can run a hands-off Scene tool that watches your chat and refreshes the background as the setting shifts, or accept a custom prompt you write yourself — with full preset and macro support.

What's New

The image generation system was substantially rebuilt. If you used Lumiverse before this release, the parts you'll notice first:

  • Three prompt modes — Scene tool, Custom prompt, and Chat-aware custom (your instructions, but applied to the live chat context by a parser LLM).
  • Prompt presets with per-character and per-persona bindings — splice character or persona snippets into a main preset with {{character_prompt}} / {{persona_prompt}}.
  • Output targets — send the result to the chat background, insert it as a new chat image, or attach it to the most recent message.
  • Prompt preview — see (and optionally edit) the resolved prompt before the image provider is called.
  • Six providers — ComfyUI and SwarmUI for local generation, plus Google Gemini, NovelAI, NanoGPT, and Pollinations in the cloud.
  • Configurable timeouts — separate limits for the parser/scene LLM and for the image provider itself.
  • Gallery integration — generated images can be auto-added to the active character's gallery and (optionally) re-sent to the LLM as multimodal context.

How It Works

  1. You pick a prompt mode — Scene tool, Custom, or Chat-aware Custom.
  2. Lumiverse builds the final prompt (running the parser/scene LLM if needed and substituting any character / persona snippets).
  3. The active image-gen connection generates the image.
  4. The result is routed to the output target you chose — background, new chat image, or attached to the last message — and optionally linked into the character gallery.

In Scene mode the cycle is automatic: each new reply is checked against the previous scene, and a new image is generated only when enough fields have changed (you control the sensitivity). In Custom and Chat-aware modes you trigger generation yourself with the Generate button.

Prompt Modes

Mode What it does
Scene tool A sidecar LLM reads the chat and extracts a structured scene (environment, time of day, weather, mood, focal detail, and — optionally — visible characters and composition). The provider-specific prompt is built from those fields.
Custom prompt Your prompt is sent to the image provider verbatim. Macros ({{user}}, {{char}}, {{character_prompt}}, {{persona_prompt}}, …) are still resolved.
Chat-aware custom Your text becomes parser instructions, not the final prompt. A parser LLM rewrites the current chat context into an image prompt following your guidance.

See Prompts & Presets for the full picture, including how to save presets and bind them to a character or persona.

Output Targets

Target Result
Set as background Image becomes the chat background at your configured opacity.
Insert into chat A new chat message is created with the image as an attachment.
Attach to last message The image is appended to the most recent message's attachments.
Preview only The image is generated but not placed in the chat — useful for testing a preset.

Generated images are persisted with thumbnails and a public URL, are addressable from the image gallery, and (when enabled) are automatically linked into the active character's gallery.

Providers At a Glance

Provider Runs Strength
ComfyUI Local Bring-your-own workflow. Full control over samplers, schedulers, checkpoints, and any custom node graph you've already built.
SwarmUI Local Friendlier wrapper around Comfy with built-in model browsing and component overrides (VAE, text encoders).
Google Gemini Cloud Prose prompts, multiple aspect ratios, up to 4K.
NovelAI Cloud Anime/illustration with Danbooru-style tags and director reference images (character / persona avatars or your own uploads).
NanoGPT Cloud Aggregator — access to Flux, HiDream, DALL·E 3, Imagen 4, Midjourney, Recraft, SDXL, SD 3.5, Reve, and others under one key.
Pollinations Cloud Lightweight provider with optional enhance, transparency, and quality tiers.

See Setup & Providers for connection setup and per-provider quirks.

Guide What You'll Learn
Setup & Providers Create an image-gen connection and configure each provider (including importing a ComfyUI workflow).
Prompts & Presets Use Scene / Custom / Chat-aware modes, save presets, bind them to a character or persona, and preview the resolved prompt.
Scene, Output & Timeouts Configure scene detection, output targets, gallery and context recycling, timeouts, and background display.