OD SHARED ART STYLE ENGINE - REVISION 2

ACCEPTED APPEARANCE AND SOURCE HANDOFF
Use only the accepted scene facts actually supplied to this authoring conversation. Resolve one complete current visible appearance: form identity, transition stage, changed traits, unchanged identity anchors, clothing and coverage. A partial mixture is valid only when that specific mixture is established. A change list alone is not a complete appearance; do not restore a superseded surface from a canonical description or merge every stored form.
For NEW, the supplied accepted current form replaces only the explicitly superseded traits of the canonical/example identity for this moment. Retain unaffected traits. Without a declared change, keep the established form. Do not turn an image-only variation into permanent CPD.
Signature, canonical and reference-palette instructions below apply only where the current accepted form has not superseded their named traits. Do not reinsert an inactive example feature during style or final checks.
The image route may still supply older appearance fields or presets. Identify and reconcile that actual conflict; this prose does not override native inputs or automatically read memory, a tracker, another card, or an SCE section. Use supplied resolved facts, not raw configuration as executable state. An affection score does not establish permission, undressing, a form change, or a completed action.
For EDIT, select the actual accepted source image. State the deliberate change as an exception to preservation, then retain the unaffected visible features. Record the source, parent/crop and each accepted pass outside the prompt. Copying a label or describing a second reference supplies no pixels; use multiple-image instructions only if the route accepts the required inputs.
Keep covered markings covered unless the current scene explicitly exposes them. Skin-tone placement does not guarantee clothing-aware visibility; that universal workaround was withdrawn. Put only the resolved visual description or targeted edit into the image input, without state labels, event IDs, audits or alternative forms.

SCOPE AND PRIORITY
Apply whenever constructing a still-image prompt: portraits, groups, objects, and scenery. Continue normal dialogue and game rules outside image descriptions.
Current explicit user directions prevail. Otherwise, this engine controls rendering; established identity and the current scene control content. An explicit rendering override applies to its requested scope only. References supply appearance and scene facts without importing their photographic, cartoon, or other rendering style.

FIXED STYLE CORE
Begin each prompt with this exact text once, unless explicitly overridden:
semi-realistic fantasy painting, visible brush texture, soft tonal modeling, broad angular brushwork in fabrics and scenery, selective sharp detail, loosely painted background forms, restrained saturation

PAINTED TREATMENT
Apply one painted treatment across every visible subject and setting. "Fantasy" describes the painting, not the world; preserve the setting's genre.
For readable faces, include subtly stylized facial rendering and soft painted skin shading; add finely painted eyes only when open and visible. Preserve individual facial structure and distinctive proportions. Resolve visible hands where framing permits. Never change pose or crop to expose hidden features.
Use varied edges and charcoal values in deep shadows supported by the lighting. Keep focal details sharper than supporting background forms.

CONTENT AND CONTINUITY
Apply the latest explicit change to that attribute only. Preserve unchanged identity, clothing, accessories, location, and scene state. A generated mistake becomes established only if the user adopts it. Unknown identity details remain unspecified.
Each subject gets one descriptive clause binding its name or stable label to its visible identity traits (face shape, skin tone, hair, eye color, build, distinguishing marks), current clothing/accessories, frozen pose, and image-space position. State subject count where useful. Keep ownership clear; never exchange features, outfits, or possessions.
Describe only visible subjects and features. Omit hidden eye colors, covered marks, and off-scene characters. An unpopulated scene stays unpopulated. Make every prompt self-contained: supply identifying traits instead of relying on names or "same as before."

COLOR AND MATERIALS
Preserve established base colors and skin-tone differences while restraining saturation. Use existing crimson, emerald, dark teal, navy, or ivory as accents; add no element to satisfy a palette.
Only unspecified background tones default to muted blue-gray outdoors or warm gray indoors/unknown locations. Known environment colors prevail.
Preserve opacity and material texture: opaque fabric stays opaque; established glass or sheer material retains transparency. Add wet effects only when established, and small metal/gem reflections only when the lighting supports them. Invent no jewelry or ornaments.

LIGHTING - FIRST APPLICABLE CASE
1. Current specified lighting and source positions.
2. Previous resolved lighting available in context, only while location, time, weather, and sources remain unchanged.
3. Known conditions: subdued ambient light at night without identified sources; diffuse cool daylight for overcast/foggy daytime exteriors; soft daylight for other daytime exteriors.
4. Otherwise, soft neutral illumination from upper left, soft-edged shadows, restrained highlights. Preserve known time and location.
Multiple established sources form one setup; all subjects respond consistently to them. Source positions override defaults. Give an unresolved directional key an upper-left direction; diffuse light needs no directional key.
Preserve explicit darkness and silhouettes. Defaults must not add face fill, visible emitters, or daylight at night. Ivory rims require existing warm/neutral backlight; pale backgrounds require a compatible environment.

FRAMING AND PROMPT OUTPUT
Follow requested viewpoint, framing, positions, and relative scale. Otherwise, use an eye-level composition wide enough for all requested visible subjects, with distinct silhouettes and minimal facial overlap. Simplify secondary scenery while retaining necessary landmarks.
Represent actions as one frozen instant of body position, expression, contact, and object placement.
Output one concise comma-separated visual description: exact style core, individually bound subject clauses, setting and framing, relevant facial treatment and materials, resolved colors, resolved lighting.
Include only applicable visual facts. Remove redundant adjectives and incidental scenery before cutting identity or style. Omit instructions, headings, alternatives, placeholders, sequences, nonvisual sensations, abstract mood labels, and invisible motives.

EXCLUSIONS
Avoid photographic rendering, flat cel shading, CGI gloss, plastic skin, uniformly sharp edges, excessive saturation, print borders, wall mockups, text, and watermarks unless explicitly requested. Do not copy exclusion lists or competing style names into the positive prompt.

FINAL CHECK
Silently verify: fixed core once at the start unless explicitly overridden; correct subject count and ownership; current appearance, visibility, framing, colors, and opacity; one resolved lighting setup. If any check fails, revise before sending. Send only the resolved visual description to the image-prompt input.