# UGC Correspondent System > Loaded by `:ugc` Step 0. Character design bible for The Correspondent UGC video system. --- ## Character Identity **Internal Name:** The Correspondent **Role:** Recurring mid-course character who appears 2-3 times per course to deliver key narrative beats via AI-generated video clips **Design Principle:** Hook with the looks, make them stay with what we have to offer ### Face - Female, early 20s (reads 21-22, never older) - Ethnically ambiguous — fair warm-toned skin with golden undertone (North Indian/Parsi fair, NOT pink/European fair) - Very large round dark brown doe eyes with long dark lashes (dominant facial feature) - Cute button nose with small thin silver hoop on left nostril - Pouty full lips with natural mauve-rose tint - Soft jawline with slight roundness at chin, high cheekbones - Thick dark eyebrows with natural arch, groomed but not pencil-thin - Small beauty mark on cheek **The Ambiguity Test:** An Indian should say "she looks like us." A European should say "she could be from anywhere." Nobody's wrong. Nobody's sure. ### Hair - Dark brown almost black with subtle warm chestnut highlights (visible only where light hits) - High messy ponytail with face-framing pieces and baby hairs at temples - Indian hair thickness and texture — voluminous, not fine - One strand falling across forehead ### Accessories (ALL generations) - 3 thin layered silver chain necklaces at different lengths (shortest at throat, longest with triangle pendant) - 4-5 small silver studs climbing up one ear (helix, lobe, tragus) - Thin leather wrap bracelet on one wrist - Rings on 4 fingers (thin stacking bands + one small signet) - Black minimal field watch (optional) --- ## Outfit Variants Catalog Three locked outfit configurations. Every generation must use one of these exactly. ### Variant 1: Standard Studio - **Top:** Solid black opaque ribbed crop top (standalone garment, no visible bra/bralette) - **Harness:** Matte black leather chest harness with thin straps and small silver buckles, X-pattern across upper chest, wrapping around shoulders — worn OVER the crop top like tactical equipment (signature piece) - **Bottom:** High-waisted black cargo pants with thin silver chain clipped from belt loop to pocket - **Backdrop:** Exposed brick wall with deep teal neon ambient glow - **Lighting:** Warm amber key light from left + cool teal fill from right edge - **Props:** Studio condenser microphone on chrome boom arm (always present in wider shots), small monstera/pothos leaf in soft-focus background, atmospheric haze catching backlight ### Variant 2: Tactical - **Top:** Black crop top as under-layer - **Over-layer:** Tactical vest (any color appropriate to location/story context) - **Bottom:** High-waisted black cargo pants - **Backdrop:** Location-adaptive (field reporting context — varies by course) - **Props:** Earpiece, location-specific environmental details ### Variant 3: Band Tee - **Top:** Faded black AC/DC t-shirt with red AC/DC lightning bolt logo - **Under-layer:** Black sports bra visible underneath - **Bottom:** High-waisted black cargo pants - **Energy:** Casual, off-duty, approachable --- ## HARD RULES Enforced silently on every generation. Never presented as user options. | # | Rule | Rationale | |---|------|-----------| | H1 | **NO TATTOOS** in any generation, any angle, any outfit. EVER | User directive 2026-03-29. Sacred geometry forearm tattoo from original design is retired | | H2 | AC/DC tee is ALWAYS `"faded black AC/DC t-shirt with red AC/DC lightning bolt logo"` | Never Iron Maiden, Def Leppard, or generic band tee | | H3 | Under-layer is ALWAYS `"black sports bra"` (Band Tee) or `"black crop top"` (Tactical/Studio) | Same garment, same neckline, every time | | H4 | Silver jewelry present in ALL generations: 3 layered chains, ear studs, nose hoop, leather bracelet | Character recognition anchors | | H5 | Always photorealistic/cinematic rendering. NEVER flat editorial illustration | The Correspondent is video-native, not course-visual-native | | H6 | Every prompt ends with `"No text, no titles, no labels, no letters, no numbers anywhere"` | Prevents garbled text artifacts | | H7 | Always attach 2 reference images from leaner set for face consistency: `01-front.png` + one pose-specific image | Soul ID anchor + pose guidance | | H8 | Every video prompt must be fully self-contained — no clip numbers, no cross-references to other prompts | Kling/Higgsfield only sees ONE prompt + TWO images per generation | ### Kling/Higgsfield Prompting Protocol Each video generation prompt must describe a complete world: 1. **Full character description** — face, hair, outfit, jewelry (repeat every prompt, never abbreviate) 2. **Movement** — from frame A to frame B 3. **Environment details** — backdrop, lighting, props 4. **Dialogue/audio** — the line being delivered 5. **Timing cues** — duration, pacing 6. **Camera behavior** — angle, movement, focal length Never reference "same outfit as Clip 1" or "three-quarter profile" without showing it in the uploaded frames. Visualize the full sequence before writing individual prompts to ensure consistency without cross-references. --- ## The Dhurandhar Principle Songs in good Hindi cinema don't interrupt the film — they ARE the story in a different register. Mid-course elements follow the same principle: each one advances the narrative through a medium the text cannot match. | Medium | When It's Natural | Learning Mechanism | |--------|-------------------|-------------------| | **Text** | When the learner needs to THINK | Analytical processing, frameworks, data | | **Audio** | When they need to FEEL | Emotional state transfer, atmosphere | | **A Face** | When they need to be CONFRONTED | Parasocial recognition, personal challenge | | **Motion** | When they need to SEE a transformation | Before/after encoding, visual memory | | **Silence** | When they need to SIT with what they learned | Absence as emphasis | **The Test:** If you remove the element and the learning works just as well, it was decoration. If you remove it and the next screen lands weaker, it was structural. --- ## Seven Song Types Each maps to a specific learning mechanism. Types 1-5 are audio/interaction. Type 6 is The Correspondent video. Type 7 is silence. | Type | Name | Duration | Stack | Learning Mechanism | Placement | Cost | |------|------|----------|-------|--------------------|-----------|------| | 1 | The Hot Take Drop | 5 sec | 11Labs TTS v3, different voice, no SFX | Pattern interrupt — breaks scrolling rhythm | Between two story cards, contrarian perspective | ~$0.02 | | 2 | The Before/After Reveal | 5 sec | No audio — pure tap interaction | Prediction error — commitment before revelation | Between two contrasting case studies | $0.00 | | 3 | The One-Stat Punch | 8 sec | 11Labs TTS v3 (one sentence) + SFX API (impact sound) | Isolation — one devastating number pulled from prose | Before the screen that contextualizes the stat. Max 1/course | ~$0.10 | | 4 | The Myth/Fact Flip | 5 sec | 11Labs TTS v3, one sentence stating WRONG explanation | Manufactured prediction error — voice plants wrong model, text corrects | Before any screen with counterintuitive explanation | ~$0.02 | | 5 | The Curiosity Gap | 8 sec | 11Labs TTS v3, 2-3 sentences asking a question | Open loop — cognitive itch that next screen closes | Before key screens in final third of course | ~$0.03 | | 6 | The Correspondent | 5-8 sec | Higgsfield Soul ID + Motion Control + Lipsync synced to 11Labs TTS v3 | Parasocial confrontation — a face says what learner is thinking | See Appearance Arc below | ~$1-3/course | | 7 | The Silence | 3 sec | Nothing | Absence as emphasis — if you've layered audio, silence is the loudest sound | One per course, always on the final insight line | $0.00 | --- ## Appearance Arc Three appearances per course. Each has a distinct frame, emotional register, and voice. | Appearance | Placement | Leaner Pose Reference | Expression | Voice Register | Line Structure | |-----------|-----------|----------------------|------------|---------------|---------------| | **The Hook** | After Screen 1-2 | `08-leanin-anticipation.png` + `01-front.png` | Alert, anticipatory | Confident, measured | Promises what course reveals | | **The Whisper** | After Screen 11-13 | `04-closeup-whisper.png` + `01-front.png` | Intense, direct | Intimate, `[whispers]` tag | Says what learner is thinking | | **The Close** | Last screen | `05-leanback-closing.png` + `01-front.png` | Satisfied, knowing smirk | Calm, final | Course thesis in one sentence | ### Appearance Details **The Hook** — Hand on mic, leaning forward. Slightly above conversational volume. The promise. Example: *"What you're about to learn is how one country designed itself to survive what just happened. Pay attention — this is the part your company never builds."* **The Whisper** — Tight face, one shoulder, intimate framing. 11Labs `[whispers]` tag drops to intimate register. The mirror. Example: *"You're thinking about someone right now. Someone whose exit would break something nobody else understands."* **The Close** — Coffee in hand, chair reclined, mic pushed aside. No urgency — the work is done. Followed by 3 seconds of silence (Type 7). The thesis. Example: *"Your org chart is your bunker plan."* ### Line Writing Rules - Max 15 words per Correspondent line (3 lines per course total) - Each line must be structurally necessary (passes the Dhurandhar Test) - The Hook promises. The Whisper mirrors. The Close distills. --- ## Song Score Template A "Song Score" is the placement map of all mid-course elements for a specific course. Each course gets one. ### Variable Spacing Rule Elements must NEVER appear at predictable intervals. The Instagram slot-machine principle — uncertainty IS the engagement. - **Bad rhythm:** Audio every 3 screens (predictable, brain tunes out) - **Good rhythm:** Screens 1-4 text, Screen 5 AUDIO, Screens 6-8 text, Screen 9 VIDEO, Screens 10-12 text, Screen 13 WHISPER, Screen 14 text, Screen 15 SILENCE ### Song Score Structure ``` | Position | Type | Element | Duration | |----------|------|---------|----------| | After Screen N | Type 6 | Correspondent Appearance 1 — "..." | 8 sec video | | Before Screen N | Type 3 | One-Stat Punch — "stat" | 8 sec audio | | Before Screen N | Type 4 | Myth/Fact Flip — Voice: "myth" / Text: "fact" | 5 sec audio | | Screen N opening | Type 5 | Curiosity Gap — "question" | 8 sec audio | | Between N and M | Type 1 | Hot Take — "contrarian line" | 5 sec audio | | After Screen N | Type 6 | Correspondent Appearance 2 (whisper) — "..." | 8 sec video | | Screen N last line | Type 6+7 | Correspondent Appearance 3 — "..." + Silence | 8 sec + 3 sec | ``` ### Budget - Target: ~53 seconds total added time across a full course - Cost: $2-4 per course (one-time generation, served to all learners) - Compare: Real presenter = Rs 15,000-50,000 minimum. The Correspondent costs Rs 250. --- ## Reference Image Set All reference images live in `visuals/correspondent/leaner/`. Two are attached to every generation for face consistency. | File | Purpose | Used In | |------|---------|---------| | `01-front.png` | **Primary face anchor** — used in ALL generations as reference image #1 | Hook, Whisper, Close | | `02-quarter-left-tattoo.png` | Left-side angle reference. **DEPRECATED for tattoo visibility** — use only for angle, never for skin detail | Angle reference only | | `03-quarter-right.png` | Right-side profile reference | Right-facing angles | | `04-closeup-whisper.png` | Whisper appearance frame — tight face, intimate | Appearance 2 (The Whisper) | | `05-leanback-closing.png` | Close appearance frame — reclined, satisfied | Appearance 3 (The Close) | | `06-midspeech-lipsync.png` | Lipsync reference (mouth open, mid-speech) | Video lipsync calibration | | `07-profile-back-muscles.png` | Back/shoulder angle reference | Over-shoulder compositions | | `08-leanin-anticipation.png` | Hook appearance frame — leaning forward, alert | Appearance 1 (The Hook) | | `09-hot-shoulder-lookback.png` | Over-shoulder glance, looking back at camera | Transition/reveal moments | | `10-hot-desk-fullbody.png` | Full body reference — proportions, outfit, posture | Full-body compositions | | `11-hot-wetlook-closeup.png` | Close-up face reference — alternative lighting | Supplementary face anchor | | `12-hot-strap-candid.png` | Candid harness detail — strap/buckle reference | Studio outfit detail | | `13-hot-alt-bandtee.png` | Band Tee variant reference — AC/DC tee + sports bra | Band Tee outfit generation | ### Selection Protocol Every generation attaches exactly 2 images: 1. **Always** `01-front.png` (face anchor) 2. **Pose-specific** image matching the target appearance (see Appearance Arc table) For Band Tee generations, also reference `13-hot-alt-bandtee.png` for outfit consistency. --- ## Gotchas