Add Setting Portability discovery
BetterLeaks / scan (push) Successful in 54s

Courtesans can be rendered in any setting (text-only).
Identity + setting are independent modular components.
Discovered via accidental Cillian/Eiríkur identity swap.

Unlocks: cross-scene variants, setting libraries, guest appearances.
Build reusable setting descriptions, apply to any courtesan.
This commit is contained in:
Dolon (Agent)
2026-03-25 10:05:32 +00:00
parent 0de13597f1
commit eda0adeda4
+170
View File
@@ -0,0 +1,170 @@
# Setting Portability — Cross-Scene Rendering
**Discovered:** 2026-03-25 (accidental identity swap during Cillian TransitionScene generation)
**What it is:** Courtesan identities and setting descriptions are **completely independent** — you can render any courtesan in any described setting, regardless of their "original" scene.
---
## Discovery Story
While generating TransitionScene images for Cillian (sauna setting), accidentally used **Eiríkur's identity description** instead of Cillian's. Result:
- ✅ **Setting matched perfectly** (Cillian's stone sauna with tiered benches, firelight)
- ✅ **Person rendered perfectly** (Eiríkur with ash-blond hair, pale blue eyes)
- ✅ **No "scene ownership" lock** — the model had no problem putting Eiríkur in Cillian's room
**Implication:** Courtesans are **not caged to their original settings**. Settings are portable descriptions.
---
## The Formula
```
courtesan_A_identity + courtesan_B_setting_description = courtesan_A in courtesan_B's room
```
**Components:**
- **Identity block:** Physical description anchors WHO is in the scene
- **Setting block:** Architectural/decor description anchors WHERE they are
- **Model treats them independently:** No validation that they "belong" together
---
## Use Cases
### 1. Cross-Scene Variants
**Example:** "Show Kara in Eira's sauna instead of her bathhouse"
- Extract Eira's Undressed sauna setting description
- Apply Kara's identity
- Generate → Kara in sauna (even though her Undressed is bathhouse)
**Why:** Creative variety, testing different atmospheres for same courtesan.
### 2. New Location Testing
**Example:** Want to try a "firelit library" setting before committing?
- Write setting description for library
- Test with one courtesan's identity
- Review results before applying to full tier
**Why:** Validate new settings without generating full cohort.
### 3. Guest Appearances
**Example:** "Show the trio (Eira/Milla/Freya) in Kara's lanternlounge"
- Extract Kara's lanternlounge setting
- Apply triple identity block
- Generate → Trio in Kara's private room
**Why:** Narrative flexibility (shared spaces, visiting arrangements).
### 4. Setting Reuse
**Example:** Cillian's sauna description works great → reuse for Eiríkur, Milla, etc.
- One good setting description = portable asset
- No need to extract setting from every courtesan individually
- Mix-and-match library of validated settings
**Why:** Efficiency. Once you have 5-10 good setting descriptions, you can combine them with any courtesans.
---
## Practical Workflow
### Standard (Original Scene)
```python
# Extract setting from courtesan's own Undressed image
vision_extract(cillian_undressed) → "stone sauna with benches..."
generate(cillian_identity + cillian_setting) → Cillian in his sauna
```
### Portable (Cross-Scene)
```python
# Use another courtesan's setting
vision_extract(cillian_undressed) → "stone sauna with benches..."
generate(eirikur_identity + cillian_setting) → Eiríkur in Cillian's sauna
```
### Library Approach (Reusable Settings)
```python
settings_library = {
"stone_sauna": "traditional stone-walled sauna with...",
"bathhouse_tiles": "polished bathhouse with tiled pool...",
"lantern_lounge": "lantern-lit salon with silk cushions...",
"fireplace_furs": "stone-walled lounge with fieldstone fireplace...",
}
# Generate any courtesan in any setting
generate(kara_identity + settings_library["stone_sauna"])
generate(cillian_identity + settings_library["lantern_lounge"])
```
---
## Constraints & Gotchas
### ✅ What Works
- **Any courtesan in any setting** (no scene ownership locks)
- **Multiple courtesans in same setting** (shared space)
- **Same courtesan in multiple settings** (alternate locations)
### ⚠️ Watch Out For
- **Tier mismatch:** Putting Poor courtesan in Elite setting may feel incongruous (though technically works)
- **Gender-coded spaces:** Some settings may read as male/female-coded (sauna vs boudoir); cross-gender may feel off
- **Setting continuity:** If using for TransitionScene workflow, must match the courtesan's actual Undressed room for narrative coherence
### ❌ Doesn't Work
- **Reference-based portability:** This discovery is **text-only generation**. Reference images anchor both identity AND setting, so portability doesn't apply when using composite references.
---
## TransitionScene Implications
**Original workflow:**
1. Extract setting from courtesan's Undressed image
2. Generate TransitionScene using same courtesan + same setting
**Flexible workflow:**
1. Build **library of validated settings** (one per tier/style)
2. For each courtesan, pick the **closest setting** from library
3. Generate TransitionScene without individual vision extraction
**Trade-off:**
- **Library approach:** Faster, reusable, but less personalized
- **Extraction approach:** Perfect match per courtesan, but more work
**Recommendation:** Use extraction for first batch to build library, then reuse settings for similar courtesans.
---
## Future Experiments
### Multi-Courtesan Scenes
- Can we render **pair/trio Undressed** by combining identities + shared setting?
- Formula: `identity_A + identity_B + setting_shared → both in same scene`
- Already validated for Dressed/Topless (reference-based); text-only pairs/trios untested
### Cross-Tier Settings
- Render Modest courtesan in Wealthy setting (aspiration/fantasy framing?)
- Render Wealthy courtesan in Poor setting (incognito/disguise narrative?)
### Setting-Only Generation
- Can we generate **empty room** by omitting identity entirely?
- Test: `prompt = setting_description_only + "no people"`
- Use case: Establish location before adding courtesans
---
## Technical Note: Why This Works
**Text-to-image models separate semantic concepts:**
- **Subject tokens:** Person, face, body, hair, eyes → WHO
- **Scene tokens:** Walls, furniture, lighting, atmosphere → WHERE
- **Pose tokens:** Standing, walking, sitting → WHAT DOING
The model composes these independently. As long as you provide clear tokens for each, it doesn't enforce "canonical" pairings. **Identity + setting are modular.**
---
## Summary
**Setting portability unlocks creative flexibility.** Courtesans can appear in any setting you can describe. No scene ownership locks exist in text-only generation. Build a library of validated settings and mix-and-match with any identities.
**Key insight:** The accidental Cillian/Eiríkur swap wasn't a failure — it was a **proof of portability**. Embrace it.