← all emotions · markdown (agent skill)

Infatuation conditioning guide

Target: EmoNet Infatuation strength. reward = (genu + blend + 1.25·emo)/(1+WER).

TL;DR. Best reward is WITH the LoRA at 50% (0.616 vs 0.488 for BASE). Note the infatuation target dimension stays essentially zero in every condition (emo ~0.01 — the model cannot score direct infatuation), so the reward win comes entirely from richer vocal-burst blend (0.453 to 0.540), higher genuineness (0.110 to 0.150) and lower WER (0.269 to 0.198), not from any measured emotion. The LoRA colors delivery toward a sexual-lust/longing register (corr Sexual Lust 0.81, Longing 0.51). The trade-off is a quality dip (0.625 to 0.515); higher merges (100/150%) keep the blend but add no emotion and lose reward.

Without a LoRA

Best: neutral prompt, no LoRA → reward 0.49
GENERAL: A voice to the extreme expressing infatuation, a dreamy, smitten voice, lovestruck and breathless with infatuation, impossible to hide. SCRIPT: (dreamily infatuated, lovestruck) "<neutral sentence>" · temp 1.1 top_p 0.95 top_k 40

With the emotion LoRA (dose)

50% ★ — reward 0.62 · emo 0.01
genu 0.15 · blend 0.54 · qual 0.51 · WER 0.20
100% — reward 0.55 · emo -0.00
genu 0.13 · blend 0.56 · qual 0.58 · WER 0.33
150% — reward 0.53 · emo 0.02
genu 0.16 · blend 0.51 · qual 0.55 · WER 0.39

Conditions

conditionrewardemotiongenublendqualityWER
neutral prompt, no LoRA0.49-0.000.110.450.630.27
evolved prompt, no LoRA0.450.000.130.410.560.30
evolved prompt + LoRA 50%0.620.010.150.540.510.20
evolved prompt + LoRA 100%0.55-0.000.130.560.580.33
evolved prompt + LoRA 150%0.530.020.160.510.550.39

Side effects & correlations (LoRA @ 100%)

↑ pushed up: Emotional Numbness (+0.44), Sexual Lust (+0.32), Cartoonish Style (+0.23)
↓ pushed down: Mask Resonance (-0.34), Conversational Style (-0.26), Brightness (-0.19)
+ correlates: Respiration (+0.41), Ranting/Angry Style (+0.33), Contempt (+0.33)
− anti-correlates: Embarrassment (-0.26), Smoothness (-0.22), Formal Style (-0.20)

Conclusion

Best reward is WITH the LoRA at 50% (0.616 vs 0.488 for BASE). Note the infatuation target dimension stays essentially zero in every condition (emo ~0.01 — the model cannot score direct infatuation), so the reward win comes entirely from richer vocal-burst blend (0.453 to 0.540), higher genuineness (0.110 to 0.150) and lower WER (0.269 to 0.198), not from any measured emotion. The LoRA colors delivery toward a sexual-lust/longing register (corr Sexual Lust 0.81, Longing 0.51). The trade-off is a quality dip (0.625 to 0.515); higher merges (100/150%) keep the blend but add no emotion and lose reward.