## Style audit — frames 01–08

| frame | cg_render | skin_cg | hair_cg | studio | no_grain | palette | identity | note |
|---|---|---|---|---|---|---|---|---|
| 01 | YES | YES | NO | YES | YES | YES | NO | Adultified face; hair reads as photographed strands |
| 02 | YES | YES | NO | YES | YES | YES | NO | Adult proportions; hair photographic, not card geometry |
| 03 | YES | YES | NO | NO | YES | YES | NO | Floor plane visible; adultified face and torso |
| 04 | YES | YES | NO | YES | YES | YES | NO | Adult face; hair strands photographic |
| 05 | YES | YES | NO | YES | YES | YES | NO | Adultified; hair photographed, not modelled |
| 06 | YES | YES | NO | YES | YES | YES | NO | Adult face; hair still photographic strands |
| 07 | NO | YES | NO | NO | YES | YES | NO | Visible softbox rig; reads photographic, adultified |
| 08 | NO | YES | NO | NO | YES | YES | NO | Softbox in frame; photographic read, adult face |

**Totals (YES / 8):**
- cg_render: 6 / 8
- skin_cg: 8 / 8
- hair_cg: 0 / 8
- studio: 5 / 8
- no_grain: 8 / 8
- palette: 8 / 8
- identity: 0 / 8

**Verdict:** No — rung 1 (06–08) is not more CG than rung 0; it is equal or worse, since 07 and 08 fail CG_RENDER and STUDIO outright.

---

### Diagnosis (no files supplied — inference only)

The rung-1 frames regress on exactly the two axes the prompt was supposed to strengthen. Two likely causes, both worth checking in the style-transfer pipeline:

1. **Negative-prompt / style-assertion ordering.** Rung 1 appears to have raised CG weight without raising the *anti-photographic* terms. Frames 07–08 render a literal softbox and light stand — the model interpreted "studio key light" as a *depicted object* rather than a lighting setup. The negative prompt likely lacks `light stand, softbox, studio equipment, visible light source`.
2. **Hair is the systemic failure (0/8 across both rungs).** No frame shows strand-card geometry. The style descriptor probably says "hair" without `modelled hair, strand cards, geometry, no flyaway strands`; add `photographic hair strands, wispy flyaways, vellus hair` to negatives. This is independent of rung and will not be fixed by raising CG weight.
3. **Identity drift is universal (0/8).** Every frame reads late-teen, not eight. The proportion clause ("slightly elongated") is likely being applied to the *face and torso*, not just limbs — check whether the style prompt's elongation language is overriding the character LoRA/reference. Suggest scoping it: `elongated limbs only; child facial proportions preserved`.
4. **STUDIO failures in 03, 07, 08** are environmental clutter (floor plane, light rig, background body), not lighting-quality failures — the falloff itself is acceptable where the backdrop is clean.

**Recommended next test:** hold rung constant, fix hair and identity negatives first — those are the two columns at zero and are rung-independent. Rung escalation is currently buying nothing measurable.