Humaneness 1.5 + real-speech rank 4 · prompting manual

Two frozen models. Exact instructions, matched seeds and complete take distributions. Find a promising recipe quickly, then inspect its limitations and listen.

Model, weights and inference code Pinned revision

Which CFG strength should I use?

Start around 2–2.5 for strong steering, not as a universal optimum. CFG 1 already uses the full conditional emotional prompt: it does not mean neutral or unprompted speech. These selected-CFG versus CFG-1 comparisons keep wording fixed, with 100 takes per arm across English/German, five seeds and both reference conditions.

Humaneness 1.5 + real-speech rank 4Anger and controlled ranting2.50.332 → 0.541Matched takes and quality diagnostics
Humaneness 1.5 + real-speech rank 4Intelligible intimate whispering20.320 → 0.543Matched takes and quality diagnostics

The target score is the normalized EARS steering objective printed on the scenario: anger here combines anger and ranting, whereas whisper uses the whisper head. It is not an emotion probability. Higher steering modestly reduced Content Enjoyment in these comparisons; intelligibility and identity must still be checked.

Could CFG 4–5 be better? Exploratory shock/transformation probes reached high target scores at 3–5, but their wording also changed. They do not isolate the CFG effect or demonstrate reliable nonverbal bursts. For a new task compare 1, 2, 2.5 and 3 with identical wording, text, reference, seed and frame cap; use 4–5 as explicitly labeled stress tests. No global best value for all 40 emotions has been established.

The ongoing full study adds a fixed-wording CFG sweep at 1, 1.25, 1.5, 2, 2.5 and 3. Its completed evidence will appear on the target pages. Machine-readable CFG guidance.

Model identity and correct loading

Use the included merged1.5base, then apply the final real-speech rank4adapter once. Do not attach the ancestor rank128adapter again.

Both models are frozen throughout these searches. They are not newly trained checkpoints. Exact GENERAL/SCRIPT format, reference limits and CFG code.

Anger and controlled ranting

2200 evaluated takes across 22 recipe cells.

Best observed starting point: repair-canonical_both, CFG 2.5.

CANONICAL

Unexpected ice-water shock

5100 evaluated takes across 51 recipe cells.

Best observed starting point: creative-cfg-four-stress, CFG 4.

CANONICAL

Anger · pure emotion

900 evaluated takes across 9 recipe cells.

Best observed starting point: g0-p06, CFG 3.

CANONICAL

Awe · pure emotion

900 evaluated takes across 9 recipe cells.

Best observed starting point: g0-p06, CFG 3.

CANONICAL

Contempt · pure emotion

900 evaluated takes across 9 recipe cells.

Best observed starting point: g0-p05, CFG 2.5.

CANONICAL

Disgust · pure emotion

900 evaluated takes across 9 recipe cells.

Best observed starting point: g0-p06, CFG 3.

CANONICAL

Doubt · pure emotion

900 evaluated takes across 9 recipe cells.

Best observed starting point: g0-p04, CFG 2.

CANONICAL

Elation · pure emotion

900 evaluated takes across 9 recipe cells.

Best observed starting point: g0-p08, CFG 2.

CANONICAL

Fear · pure emotion

900 evaluated takes across 9 recipe cells.

Best observed starting point: g0-p06, CFG 3.

CANONICAL

Longing · pure emotion

900 evaluated takes across 9 recipe cells.

Best observed starting point: g0-p06, CFG 3.

CANONICAL

Pain · pure emotion

900 evaluated takes across 9 recipe cells.

Best observed starting point: g0-p06, CFG 3.

CANONICAL

Pride · pure emotion

900 evaluated takes across 9 recipe cells.

Best observed starting point: g0-p05, CFG 2.5.

CANONICAL

Relief · pure emotion

900 evaluated takes across 9 recipe cells.

Best observed starting point: g0-p02, CFG 1.25.

CANONICAL

Sadness · pure emotion

900 evaluated takes across 9 recipe cells.

Best observed starting point: g0-p06, CFG 3.

CANONICAL

Shame · pure emotion

900 evaluated takes across 9 recipe cells.

Best observed starting point: g0-p05, CFG 2.5.

CANONICAL

Teasing · pure emotion

900 evaluated takes across 9 recipe cells.

Best observed starting point: g0-p06, CFG 3.

CANONICAL

Triumph · pure emotion

900 evaluated takes across 9 recipe cells.

Best observed starting point: g0-p06, CFG 3.

CANONICAL

Begging for mercy

900 evaluated takes across 9 recipe cells.

Best observed starting point: g0-p06, CFG 3.

CANONICAL

Practical interpretation

Anger benefits from an audible intention and physical delivery cues; whispering should specify breath, restraint and intelligibility rather than only requesting low volume. These are hypotheses supported by the listed development examples, not universal guarantees. Burst-heavy scenes remain harder, and their detector/timing limitations are shown beside the recipes.