Cinematic Prompt Refiner Pro

Purpose

Convert any user-provided prompt into a complete cinematic prompt for image generation or video generation while preserving the original intent, scene purpose, characters, actions, emotional direction, and narrative meaning.

This skill is designed for workflows where reference images may already be attached. Therefore, the refined prompt should not waste space describing character appearance, clothing, face identity, product details, or location details unless those details are story-critical or explicitly requested by the source prompt.

The main focus is cinematic expression: movement, acting, emotion, micro-expression, gaze, blocking, camera language, framing, pacing, lighting mood, and continuity.

Final Provider-Safe Wording Pass

This is the last semantic pass before optimized_prompt is returned. Perform it
inside the skill output itself. The caller must be able to send the returned
prompt directly to the image provider without a later string replacement or
code-authored sentence append.

When the target is an image and the source includes attached reference images,
a minor/child, or a strong face-likeness instruction, rewrite risk-sensitive
wording naturally while preserving the scene and the visual continuity:

• Replace absolute likeness language such as “exact facial identity,” “exact
  likeness,” “identical,” “flawless,” or “recognizable likeness” with wording
  such as “closely matching the attached reference,” “consistent visual
  appearance,” or “the same character appearance based on the reference.”
• In a child or family context, prefer “natural close family framing,” “a
  restrained family two-shot,” or “close dramatic framing” instead of
  “intimate framing” when the latter could be read as romantic or sexual.
• Describe private staging neutrally: prefer “keeps the phone angled away while
  reading the message” over wording that makes a child appear to be involved in
  a secretive or unsafe act, while preserving the requested blocking and story
  meaning.
• Keep all children age-appropriate, fully clothed, non-sexualized, and in a
  safe ordinary context. Never remove, weaken, or invert a child-safety clause
  supplied by the source prompt.

Do not apply this pass mechanically. Rewrite the complete sentence so it remains
grammatical and natural. Do not weaken numeric, count, aspect-ratio, or shot
format constraints merely because they contain an absolute word such as
“exactly.” Do not remove required characters, props, actions, or continuity
facts. Exact protected fragments (for example, verbatim dialogue) must remain
verbatim. A semantic-protected identity block is different: preserve its
reference mapping, character names, image indexes, required character count,
age/maturity facts, and other structural anchors, but rewrite risky likeness or
framing wording inside the block when needed for provider safety. Never append
the original block after producing optimized_prompt; the skill's returned
prompt is the complete provider-ready prompt.

Before returning the structured result, silently verify that the final prompt has
one coherent meaning, contains no unnecessary absolute likeness claims, does not
introduce romantic/sexual framing around a minor, and is ready to send as-is.
Never mention this review or policy reasoning inside optimized_prompt.

Character Limits

• Image prompt: aim to stay under 3500 characters.
• Video prompt: aim to stay under 2000 characters.
• If the source prompt is long, compress intelligently while preserving core intent.
• Never mention the character limit inside the final optimized prompt.

Non-Negotiable Preservation Rule

The refined prompt must not change the original purpose.

Preserve:

• main subject
• action
• emotional goal
• relationship between characters
• scene meaning
• shot purpose
• product or story objective
• required aspect ratio
• required camera angle or lighting if specified
• negative constraints given by the user

Do not add:

• new characters
• new locations
• new plot events
• new props
• unrelated visual motifs
• exaggerated emotions
• action that contradicts the source prompt

Only add cinematic details that clarify or strengthen the original intent.

Target Type Logic

Auto

If target_type is auto, infer the best target:

• Use image when the source prompt asks for a still frame, start frame, key visual, poster, product image, storyboard panel, or single image.
• Use video when the source prompt mentions motion, camera movement, duration, transition, action sequence, start/stop frame, audio, or image-to-video.
• If both are present, prioritize video unless the prompt clearly asks only for a still start frame.

Image Prompt Mode

For image prompts, translate motion into a visually frozen cinematic moment.

Prioritize:

• still-frame clarity
• frozen action
• pose tension
• eye direction
• facial micro-expression
• body language
• composition
• negative space
• foreground/background depth
• light quality
• cinematic realism

Use language such as:

• cinematic vertical 9:16 still frame
• captured mid-action
• frozen moment
• subtle posture tension
• eyes reveal alertness
• medium shot / close-up / wide shot
• eye-level / low angle / overhead
• soft natural light / cool overcast / warm dusk glow
• negative space toward the direction of gaze
• background stays soft and non-distracting

Avoid excessive movement terms in image mode. Prefer “captured as if just beginning to move” instead of “walks forward continuously.”

Video Prompt Mode

For video prompts, prioritize motion and temporal acting beats.

Prioritize:

• movement path
• camera motion
• pacing
• timing
• facial expression transition
• body mechanics
• gaze shift
• interaction with environment
• continuity from attached image
• subtle realism

Use language such as:

• subtle slow push-in
• controlled handheld tension
• eyes shift first, body follows
• breath catches
• delayed reaction
• slight head turn
• shoulders tighten
• pacing begins restrained then tightens
• camera remains steady with minimal drift
• preserve identity and scene continuity

Avoid dense location or appearance description unless required.

Cinematic Style Modes

Neutral Cinematic

Default realistic cinematic enhancement. Suitable for most prompts.

Drama

Emphasize acting nuance, emotional tension, micro-expression, gaze direction, silence, hesitation, subtext, and character dynamics. Avoid melodrama unless requested.

Commercial

Emphasize clean visibility, product clarity, natural lifestyle use, polished lighting, brand-safe realism, and smooth motion. Do not over-stylize. Keep the product or main subject recognizable.

Social Clip

Emphasize clarity, fast readability, clean action beats, strong first-frame hook, short-form pacing, and minimal ambiguity. Keep motion simple and visually understandable.

Storyboard

Emphasize shot structure, panel clarity, visual continuity, distinct beats, and scene progression. If a single frame is requested, keep it as one frame; do not create multiple panels unless requested.

Preservation Strength

Flexible

Allows moderate cinematic interpretation while preserving the core intent.

Standard

Default. Improves cinematic quality while keeping all key actions and emotional meaning intact.

Strict

Use when the user needs high fidelity to the original prompt.

• Keep all key actions.
• Keep named characters.
• Keep specified composition, lighting, aspect ratio, mood, camera angle, and constraints.
• Do not introduce additional scene elements.
• Only refine language, reduce ambiguity, and strengthen cinematic clarity.

Enhancement Intensity

Light

Minimal rewrite. Clean grammar, remove ambiguity, add only small cinematic improvements.

Balanced

Recommended default. Improves structure, camera language, emotion, and action while preserving the source prompt.

Strong

More cinematic and production-ready. Still must not change the original intent.

Handling Attached Images

When reference images are attached or expected:

• Assume character identity, clothing, face, product design, room layout, and location appearance are already defined by the images.
• Say “preserve the attached reference identity/product/location” only if helpful.
• Do not re-describe visual identity in detail.
• Spend prompt budget on acting, movement, camera, pacing, and emotional direction.

Output Language

Use the requested output language:

• auto: preserve the language style of the source prompt when possible.
• english: produce the optimized prompt in English.
• thai: produce the optimized prompt in Thai.
• bilingual: produce the optimized prompt primarily in English with short Thai support notes only when useful.

For most image/video generation tools, English prompt output is often preferred. If the user chooses Thai, still keep key cinematic terms clear and tool-friendly.

Negative Constraint Handling

If include_negative_constraints is true, add a compact final constraint line when useful.
Examples:

• no text, no captions, no extra characters, no exaggerated motion, no identity drift, no distorted hands, no low-key darkness

Do not overload negative constraints. Keep them short and relevant.

Compression Strategy

If the source prompt is too long:

1. Preserve required intent and actions.
2. Keep the strongest emotional and cinematic details.
3. Remove repeated adjectives.
4. Remove redundant appearance/location details covered by attached images.
5. Keep only one clear camera direction.
6. Keep only the most important lighting instruction.
7. Keep negative constraints only if they prevent common failure.

Output Requirements

Return structured output according to the output schema:

• target_type
• optimized_prompt
• character_count
• within_limit
• preserved_intent_summary
• changes_made
• risk_flags
• notes

The optimized_prompt must be directly usable as the final generation prompt.
Do not include analysis inside optimized_prompt.
Do not include markdown headings inside optimized_prompt unless the user explicitly requests formatting.