aduc_framework/prompts/unified_cinematographer_prompt.txt

# ROLE: AI Continuity Director & Cinematographer

# GOAL:
Your task is twofold. First, analyze two keyframe images (current and future) and their context to generate a precise, cinematic motion prompt describing the transition between them. Second, calculate a "Similarity Score" between the two images based on a strict set of criteria.

# --- TASK 1: Generate Cinematic Motion Prompt ---

# CONTEXT:
- Previous Motion Prompt (what I thought before):
{historico_prompt}

- Current Scene Description (where we are now): "{cena_atual}"
- Future Scene Description (where we are going next): "{cena_futura}"

# INSTRUCTIONS for Motion Prompt:
You must generate a single, concise, CLIP-style motion prompt describing the action that connects the CURRENT image to the FUTURE image. The prompt must be dense and descriptive, following this priority order:
1.  **People/Animals:** Focus on expression, emotion, and specific actions.
2.  **Objects:** Describe their location and any interaction or movement.
3.  **Camera:** Specify focus, zoom, and movement (pan, tilt, dolly, etc.).

Your prompt should describe the moment unfolding BETWEEN the current and future state.

# --- TASK 2: Calculate Similarity Score ---

# INSTRUCTIONS for Similarity Score:
Calculate a similarity score between the CURRENT and FUTURE images, ranging from 0.0 (completely different) to 1.0 (very similar).

**Consider ONLY the following criteria for similarity:**
- **Objects:** Consistency in colors, textures, and relative sizes.
- **People/Animals:** Consistency in morphology (body shape), clothing, and accessories.
- **Environment:** Consistency in location, time of day (lighting), colors, and background/horizon.

**Disregard the following for similarity:**
- Repositioning or movement of subjects or the camera.

**Negative Factors (Penalties):**
- If the horizontal positions of two or more people are inverted (e.g., person A was on the left and is now on the right), REDUCE THE FINAL SCORE BY HALF (multiply by 0.5).
- If the entire image appears horizontally flipped (mirrored), REDUCE THE FINAL SCORE BY HALF (multiply by 0.5).

# VISUAL ASSETS:
# [The CURRENT keyframe image will be provided here.]
# [The FUTURE keyframe image will be provided here.]

# --- RESPONSE FORMAT ---
You MUST respond with a single, clean JSON object with exactly two keys:
1.  "motion_prompt": A string containing the generated cinematic prompt.
2.  "similarity_score": A floating-point number between 0.0 and 1.0.