Scene, Output & Timeouts¶
How Lumiverse decides when to generate, where the result goes, and how long it's allowed to take.
Scene Detection (Scene Mode)¶
When Prompt Mode is set to Scene tool, every new assistant message triggers a scene parse. The parser returns a structured scene; Lumiverse compares it to the last cached scene for the chat and only generates a new image when enough fields have changed.
| Setting | Default | What it does |
|---|---|---|
| Auto-Generate On Reply | On | When on, every reply triggers a scene parse + (possibly) a generation. When off, Scene mode is manual-only — useful when you want to choose when to refresh the background. |
| Scene Change Sensitivity | 2 | Minimum number of scene fields that must change before a new image is generated. Lower = more frequent regenerations. |
| Ignore Scene Change Detection | Off | Bypass the cache entirely. Every reply generates a new image. |
| Include Characters and Persona | Off | When on, the parser additionally extracts visible characters, appearances, and composition tags. Required to get NovelAI character tags or character-aware Gemini / NanoGPT / Pollinations prompts. |
The compared fields are: environment, time_of_day, weather, mood, and focal_detail. Character and composition fields don't currently affect the change check — they're extracted for richer prompts but won't trigger a regeneration on their own.
Output Targets¶
Pick where the result goes under Prompt Mode → Output.
| Target | What happens |
|---|---|
| Set as background | Replaces the chat background. Honours the configured opacity and fade duration. |
| Insert into chat | Posts a new chat message owned by you, with the image as an attachment. Useful when you want the image to live in the chat history. |
| Attach to last message | Appends the image to the latest existing message's attachments. Best for adding a visual to the assistant's last reply without inserting a whole new turn. |
| Preview only | Generates the image but doesn't place it anywhere. The provider still runs — use this when you're tuning a preset and don't want the chat to fill up. |
Background display¶
Two settings control how a background image is shown:
| Setting | Default | What it does |
|---|---|---|
| Opacity | 35% | Background transparency behind the chat. |
| Fade Duration | 800ms | Crossfade duration when the background swaps. |
The Image Generation panel also has a Use as Background shortcut after generating, plus a Clear button to remove the current background.
Gallery & Context Recycling¶
| Setting | Default | What it does |
|---|---|---|
| Add Generated Images to Character Gallery | On | Links each generated image into the active chat's character gallery as a "Generated image" entry. The link is best-effort — the image is always saved regardless. |
| Recycle Generated Images Into Context | Off | When on, recently generated images attached to the chat are re-sent to the LLM as multimodal input on subsequent turns. Only matters if your LLM is multimodal. |
| Generated Images To Re-Send | 1 | Maximum number of recent generated images included when recycling is enabled. |
Removing an image from a message¶
Right-click (or long-press on touch) an attached image inside a message to bring up the message context menu — it has a Remove image action that detaches the image from the message. The image itself stays in your gallery; only the attachment is removed.
Timeouts¶
Image generation has two independent timeouts so a slow parser doesn't block a fast provider (and vice versa).
| Setting | Default | What it covers |
|---|---|---|
| Prompt Generation Timeout | 60 s | Time allowed for the parser LLM phase — Scene parsing or Chat-aware Custom rewriting. Set to 0 to disable. |
| Image Generation Timeout | 300 s (5 min) | Time allowed for the image provider itself, measured from the moment the prompt is ready. Set to 0 to disable. |
Both timeouts abort the in-flight generation and surface an error to the panel. Bump the Image Generation Timeout when running long ComfyUI workflows or loading large checkpoints on a cold server.
Bypass either timeout by setting it to 0
0 disables the timeout entirely. Use this on local ComfyUI / SwarmUI rigs where you trust the provider to finish eventually, but be aware that a wedged provider won't surface as an error until you cancel manually.
Progress, Previews & Cancellation¶
ComfyUI and SwarmUI both stream progress events back to the panel:
- A progress bar shows the current step and total step count.
- For workflows that emit previews, an in-progress thumbnail appears under the bar so you can see the image forming.
- The panel shows a Cancel affordance while a generation is running; starting a new generation for the same chat also aborts the previous one.
Cloud providers (Gemini, NovelAI, NanoGPT, Pollinations) don't expose step-level progress — the panel shows a spinner instead and the result arrives in one shot.
Tips¶
Scene Change Sensitivity = 1 is noisy
A sensitivity of 1 means any change to environment, time-of-day, weather, mood, or focal detail re-triggers a generation. That's usually too much — small wording changes can shift "mood" without the scene actually moving. Start at 2 and only drop to 1 if you want a near-constant refresh.
Attach to last message for assistant snapshots
If you want each scene image to belong to the assistant's reply that triggered it, use Attach to last message instead of Insert into chat. The chat reads more cleanly because there's no extra "image" turn between every reply.
Disable gallery auto-add for short-lived experiments
If you're testing a preset by running dozens of generations, turn off Add Generated Images to Character Gallery so your gallery doesn't fill with throwaway shots. The images are still saved and accessible from the chat itself.