=== HERMES AI IMAGE GENERATION TEMPLATE START ===
════════════════════════════════════════════════════
PART 1 — CHARACTER REFERENCE IMAGES
════════════════════════════════════════════════════
TASK: Generate locked character reference images for a short film.
STYLE:
Photorealistic cinematic realism.
Grounded [COUNTRY/REGION] [GENRE] drama.
Natural skin texture, realistic proportions.
No beauty filter, no fantasy, no anime, no cartoon.
No exaggerated emotion.
Plain white background for reference sheets.
Full body.
Front view + side profile + back view.
Consistent face identity across all views.
Cinematic soft lighting, high detail, natural motion, no distortion.
ASPECT RATIO:
Vertical 9:16.
Note: gpt-image-2 defaults to square — embed “vertical portrait composition” in prompt.
GLOBAL NEGATIVE PROMPT:
Avoid anime, cartoon, fantasy, glossy beauty skin, perfect model face, luxury house, modern rich apartment, exaggerated crying, open mouth screaming, horror face, distorted hands, extra fingers, extra limbs, deformed face, wrong ethnicity, heavy makeup, colorful fashion outfit, unrealistic lighting, over-smiling, action pose, crowd, children, text, watermark, logo, blur, low resolution, duplicate characters, inconsistent face, distorted eyes.
API CALL FORMAT:
POST https://api.openai.com/v1/images/generations
Model: gpt-image-2
Params: {“model”: “gpt-image-2”, “prompt”: “”, “n”: 1}
Returns: base64 JSON (b64_json) — decode and save as PNG
Note: gpt-image-2 does NOT support negative_prompt or aspect_ratio params.
Embed negative terms and aspect ratio description directly in the prompt text.
FOR EACH CHARACTER, OUTPUT:

Character name and role
Character lock description (identity traits to preserve)
Reference sheet generation prompt (3-view turnaround)
Expression variation prompts (5 expressions)

════════════════════════════════════════════════════
PART 2 — OBJECT / PROP REFERENCE IMAGES
════════════════════════════════════════════════════
TASK: Generate locked symbolic object reference images.
STYLE:
Close-up product-style reference.
Plain white background.
Photorealistic, high detail.
Cinematic soft lighting.
Same object identity across all uses.
OUTPUT:

Object name and symbolic meaning
Object lock description
Object reference generation prompt
Variation prompts (different states: screen on/off, held, on table, etc.)

════════════════════════════════════════════════════
PART 3 — LOCATION / BACKGROUND REFERENCE IMAGES
════════════════════════════════════════════════════
TASK: Generate locked location reference images.
STYLE:
No people in reference image.
Modest [COUNTRY/REGION] setting.
Grounded realism, clean but poor living space.
Dim warm lighting, soft evening shadows.
Photorealistic cinematic realism.
OUTPUT:

Location name and description
Location lock description (props, materials, mood)
Location reference generation prompt
Variation prompts (wide shot, close-up details, doorway angle, wall angle, light variations)

════════════════════════════════════════════════════
PART 4 — SCENE IMAGES
════════════════════════════════════════════════════
TASK: Generate scene stills for each clip using locked character/object/location references.
RULE: Do NOT redesign characters. Use only the existing reference images as identity guides.
REFERENCE FORMAT IN PROMPTS:
@daughter_reference = young [COUNTRY] woman character image
@father_reference = [COUNTRY] father character image
@phone_reference = old smartphone object reference image
@home_reference = modest [COUNTRY] village home interior reference image
STYLE:
Photorealistic cinematic realism.
Grounded [COUNTRY/REGION] [GENRE] drama.
Quiet emotional atmosphere.
Natural acting.
No exaggerated crying, no overacting.
No advertisement feeling, no beauty filter.
No fantasy, no anime, no cartoon.
No luxury home, no modern rich apartment.
COLOR ARC:
Scene 1: Dim and slightly cool.
Scenes 2-4: Warm but heavy and shadowy.
Scene 5: Slightly warmer and more emotional.
Scene 6: Warmest and most hopeful.
FOR EACH SCENE IMAGE, OUTPUT:

Scene number and filename
Reference images to use (@daughter_reference, @father_reference, etc.)
Detailed image generation prompt (self-contained, 80-130 words)

Full character description (locked, consistent)
Environment / location details
Lighting style and color palette
Mood and atmosphere
Composition (framing, angle, lens feel)
Ends with: “vertical portrait composition, cinematic lighting, natural motion, no distortion”


Negative prompt (use global negative prompt)

OUTPUT FORMAT PER SCENE:
────────────────────────────────────
SCENE IMAGE XX — filename.png
References: @daughter_reference, @phone_reference, @home_reference
Prompt:
[Full detailed prompt]
Negative Prompt:
[Global negative prompt]
────────────────────────────────────
════════════════════════════════════════════════════
PART 5 — COPY-PASTE REQUEST TEMPLATE
════════════════════════════════════════════════════
HERMES — IMAGE GENERATION REQUEST
TYPE: [ ] CHARACTER REFERENCES ONLY
[ ] SCENE IMAGES ONLY
[ ] BOTH
FILM TITLE: