Image Generation Pipeline

The Image Generation Pipeline produces AI-generated visuals (backgrounds, portraits, illustrations) using ComfyUI, guided by configurable style profiles and automatic contextual triggers.

Style Profiles

Six built-in profiles define the visual aesthetic: Auto, Anime, Realistic, Cinematic, Fantasy, and Pixel Art. Each profile includes a prompt grammar (natural language, Danbooru tags, or comma-separated tags), positive/negative tag sets, and per-image-type overrides.

Built-in profiles are immutable but can be cloned to create editable custom variants. Select your active profile from the Providers settings page.

Prompt Compilation

The compileImagePrompt() function merges your base prompt with the active style profile, removes near-duplicate tags (case-insensitive), extracts negative phrases (text, watermark, blurry) to the negative prompt, and injects per-image-type tags (e.g., “scenic, wide shot” for backgrounds).

Contextual Triggers

Auto-generate images from game events:

• Location changed → scenic background
• Combat started → battle scene
• NPC introduced → portrait
• Dramatic moment → scene illustration
• Quest completed → commemorative illustration

Triggers include debouncing (max one generation per event type per 30 seconds) and NPC deduplication (one portrait per NPC per session). Enable or disable in settings.

Gallery Panel

Generated images are stored per-chat in a masonry grid with hover expand and full-res modal view. Each image retains its prompt metadata and image type badge.

Source

• Style profile service: apps/frontend/client/src/lib/services/image/style_profile_service.svelte.ts
• Prompt compiler: apps/frontend/client/src/lib/services/image/prompt_compiler.ts
• Contextual triggers: apps/frontend/client/src/lib/services/image/contextual_trigger_service.svelte.ts
• Gallery service: apps/frontend/client/src/lib/services/image/gallery_service.svelte.ts
• Dev sandbox: /dev/image-gen