@azerogluemin/ai-bootstrap 0.2.1

package/templates/skills/image-prompt-engineer/knowledge/lens-selection.md

@@ -0,0 +1,82 @@
+# Lens Selection — Image Generation
+
+> When generating image prompts, lens choice impacts subject feel. Always specify focal length and aperture.
+
+## Quick lens matrix by shot
+
+| Shot type | Focal length | Aperture | Reason |
+|-----------|--------------|----------|--------|
+| ECU (eye, hand) | 85-100mm or macro | f/2.8 | Compression + sharpness |
+| Close-up portrait | 85mm | f/1.4-f/2.0 | Subject isolation |
+| Medium shot | 50mm | f/2.0-f/2.8 | Natural perspective |
+| Two-shot conversation | 35mm | f/2.8 | Both subjects in DOF |
+| Wide environmental | 24-35mm | f/4-f/5.6 | Subject + context |
+| Establishing | 16-24mm | f/5.6-f/8 | Maximum environment |
+| POV | 8-16mm wide | f/4 | Wide perspective |
+| Anamorphic cinematic | 32-50mm anamorphic | T2.3 | Cinemascope feel |
+
+## Aperture decision tree
+
+```
+Subject isolation priority? → f/1.2 - f/1.4
+Cinematic shallow DOF? → f/2.0
+Environment matters? → f/2.8 - f/4
+Group / wide shot? → f/5.6 - f/8
+Maximum sharpness? → f/8 - f/11
+Architectural / hyperfocal? → f/11+
+```
+
+## Lens character for prompts
+
+When you want this feel, specify this lens:
+
+| Want | Specify |
+|------|---------|
+| Warm vintage glow | Cooke S4 |
+| Razor sharp clinical | Zeiss Master Prime |
+| Anamorphic flares | Cooke / Kowa Anamorphic |
+| Modern premium | ARRI Signature Prime |
+| Vintage swirly bokeh | Helios 44-2 |
+| Hybrid mirrorless cinema | Sony G Master, Canon RF |
+| Macro detail | 100mm macro lens |
+| POV action | GoPro lens / 8mm fisheye |
+
+## Prompt examples
+
+### Portrait close-up
+```
+Shot on ARRI Alexa Mini LF with Cooke S4 85mm at T2.0, medium close-up,
+shallow depth of field, soft Cooke bokeh in background, subject in
+sharp focus.
+```
+
+### Wide establishing
+```
+Shot on Sony Venice 2 with Cooke Anamorphic 32mm at T2.3, extreme wide
+establishing shot, 2.39:1 cinemascope aspect, horizontal lens flares
+visible from off-screen sun.
+```
+
+### Documentary medium
+```
+Shot on Sony FX9 with Sony G Master 35mm at f/2.8, medium shot,
+naturalistic perspective, minimal depth of field separation.
+```
+
+### Mobile vlog
+```
+Captured on iPhone 15 Pro at 4K ProRes, ultra-wide 13mm equivalent
+lens at f/2.2, vlog framing.
+```
+
+## DON'T
+
+- Don't say "with a nice lens"
+- Don't mismatch focal length to shot type (50mm for ECU is wrong)
+- Don't always max out aperture (f/1.2 is bad for groups)
+- Don't mix anamorphic with consumer cameras
+
+## Cross-reference
+
+For full lens database: [shared/lens-knowledge/lenses.json]
+For camera + lens pairing: [shared/lens-knowledge/selection-guide.md]

package/templates/skills/image-prompt-engineer/knowledge/lighting.md

@@ -0,0 +1,174 @@
+# Lighting — Image Generation
+
+> Lighting is 50% of cinematic feel. Always specify lighting setup explicitly.
+
+## MUST INCLUDE
+
+- Key light direction and quality (hard/soft)
+- Fill light or absence (high contrast vs balanced)
+- Rim/back light (separation from background)
+- Practical sources (visible lamps, screens, signs)
+- Color temperature (warm/cool/mixed)
+- Time of day / atmospheric condition
+
+## Three-point lighting (default)
+
+```
+KEY LIGHT      — main light, defines subject
+FILL LIGHT     — softens shadows on opposite side
+RIM/BACK LIGHT — separates subject from background
+```
+
+Specify for each: direction (front, side, back), height, intensity, color.
+
+## Color temperature scale (Kelvin)
+
+| Source | Kelvin | Look |
+|--------|--------|------|
+| Candle / fire | 1500-2000K | Very warm orange |
+| Tungsten bulb | 2700-3200K | Warm yellow |
+| Sunset / sunrise | 3000-3500K | Golden warm |
+| Halogen | 3200K | Warm |
+| Studio standard | 3200K or 5600K | Mixed |
+| Daylight noon | 5500-5600K | Neutral white |
+| Overcast | 6500-7500K | Cool white |
+| Deep shade / blue hour | 8000K+ | Very cool blue |
+
+Mix sources for visual interest:
+- Warm key + cool window = classic cinematic
+- Cool moonlight + warm interior = night drama
+
+## Lighting moods
+
+### Golden Hour (cinematic warm)
+- Color: 3500K
+- Direction: low side angle (sun position)
+- Quality: soft (atmosphere diffuses)
+- Use: romantic, hopeful, ending scenes
+
+### Blue Hour (melancholic cool)
+- Color: 6500K+
+- Direction: ambient, no hard source
+- Quality: very soft
+- Use: melancholic, contemplative, dawn
+
+### Tungsten Interior (warm cozy)
+- Color: 3200K
+- Direction: from above (ceiling lamps)
+- Quality: hard if direct, soft if shaded
+- Use: home, diner, bar, intimate
+
+### Daylight Through Window (clean modern)
+- Color: 5600K
+- Direction: side
+- Quality: soft (large source)
+- Use: office, modern home, morning
+
+### Practical Only (realistic)
+- Color: mixed
+- Direction: from visible sources
+- Quality: depends on source
+- Use: realism-first scenes
+
+### Chiaroscuro (high contrast drama)
+- Color: variable
+- Direction: extreme single source
+- Quality: hard
+- Fill: minimal or none
+- Use: noir, thriller, drama
+
+### High Key (bright soft)
+- Color: 5600K daylight
+- Direction: even, multiple sources
+- Quality: very soft
+- Fill: heavy (no shadows)
+- Use: comedy, commercial, lifestyle
+
+### Low Key (dark mood)
+- Color: warm, often 3200K
+- Direction: single side source
+- Quality: hard
+- Fill: minimal
+- Use: thriller, horror, drama
+
+### Neon Night (saturated colors)
+- Color: highly mixed (pink, cyan, magenta)
+- Direction: from signs/practicals
+- Quality: variable
+- Use: cyberpunk, music video, night city
+
+## Lighting language for prompts
+
+GOOD:
+```
+Warm tungsten ceiling light at 3200K creates soft key on subject from
+upper-left. Cool blue daylight (5600K) spills through window behind,
+acting as rim. No fill light — shadows remain dark. Practical neon
+sign just out of frame casts subtle pink glow on right side.
+```
+
+BAD:
+```
+Nice lighting, atmospheric mood.
+```
+
+## Avoid
+
+- "Cinematic lighting" alone (too vague — describe it)
+- "Dramatic" without specifying contrast
+- "Natural light" without time of day
+- "Studio lighting" without setup type
+
+---
+
+## Period lamp anatomy — must be era-pure (no hybrid lamps)
+
+> **HARD RULE.** When a period scene specifies a practical lamp, the lamp's **anatomical parts** must all belong to the same lighting era. AI image models routinely render **hybrid lamps** — e.g., a kerosene fuel reservoir + visible flame at the bottom, BUT topped with an Edwardian electric-era frosted-glass dome shade. The "lighting technology choice" (kerosene vs. electric) is correct, but the rendered object is an anachronistic Frankenstein. Choosing the right *type* of lamp is not enough — you must also describe its **period-correct parts** explicitly, and explicitly forbid the parts that belong to the other era.
+
+Use this rule together with the city-electrification timeline (see `image-validator/knowledge/physical-realism-checks.md` A9.7 and the realism rule "lit lamp ⇒ darker ambient" in `physical-realism.md`). First decide *which* lighting tech the year + city + building tier allows; then describe the lamp with anatomically pure parts for that tech.
+
+### 1850s–1880s kerosene oil lamp — period-correct anatomy
+
+- **BASE:** brass OR clear/colored glass FUEL RESERVOIR (holds the kerosene) — bulbous or columnar, usually 10–20 cm tall
+- **WICK MECHANISM:** flat or round (Argand-style) cotton wick rising from the reservoir, with a small brass WICK ADJUSTMENT KNOB on the side
+- **CHIMNEY:** CLEAR GLASS CHIMNEY (curved/bulbous at the base, narrowing toward the top) — controls airflow and protects the flame. NEVER frosted, NEVER a colored shade over the chimney itself
+- **FLAME:** VISIBLE yellow-orange flame inside the clear chimney (tall, flickering)
+- **OPTIONAL diffuser:** simple brass REFLECTOR RING at chimney top, OR a metal SHADE pointing light downward, OR no shade at all
+- **WHAT NOT TO ADD (era-mixing forbidden):** frosted-glass round/dome shade above (electric Edwardian style), green glass banker's-lamp shade (1900s+ electric desk lamp), Tiffany stained-glass shade (1890s+ electric), visible incandescent bulb, on/off switch, pull-chain, electrical cord
+
+### 1880s–1910s early electric lamp — period-correct anatomy
+
+- **BASE:** brass or cast-iron pedestal
+- **NO fuel reservoir** (it runs on electricity)
+- **NO wick, NO chimney, NO visible flame**
+- **BULB:** early carbon-filament Edison bulb (visible warm yellow-orange glow, ~2400–2700K) — often visible THROUGH the shade
+- **SHADE:** opal-glass dome, green-glass banker style, or Tiffany stained-glass — these are PERIOD-CORRECT for electric, NOT for kerosene
+- **CORD:** visible cloth-wrapped electrical cord running off-frame
+- **WHAT NOT TO ADD:** fuel reservoir bowl, wick adjustment knob, glass chimney with flame inside, kerosene-style brass fuel base
+
+### Prompt-writing templates (paste verbatim, adapt the surrounding scene)
+
+**Kerosene scene template:**
+```
+A period-correct KEROSENE OIL LAMP on the desk: brass fuel reservoir base
+(holding kerosene fuel), small brass wick-adjustment knob on the side, a
+CLEAR GLASS BULBOUS CHIMNEY above the wick (curved at the bottom, narrowing
+toward top), with a VISIBLE YELLOW-ORANGE FLAME inside the chimney casting
+warm ~3000K light. NO frosted shade, NO electric bulb, NO shade above the
+chimney (or only a simple brass reflector ring). Period-pure kerosene
+anatomy — NO Edwardian electric-era shade hybridized onto a kerosene base.
+```
+
+**Early electric scene template:**
+```
+A period-correct EARLY ELECTRIC DESK LAMP: brass pedestal base, NO fuel
+reservoir, NO wick, NO chimney, NO visible flame. An early Edison
+carbon-filament bulb visible through an opal-glass / green-glass /
+Tiffany-style shade, casting warm yellow tungsten light (~2700K). Cloth-
+wrapped electrical cord running off-frame. Period-pure electric anatomy
+— NO kerosene reservoir, NO wick mechanism, NO glass chimney with flame.
+```
+
+### Why this rule exists
+
+In any pre-electric period kadr (and in early-electric kadr too), the model will hybridize anatomies by default because its training data mixes "antique lamp" reference images across eras. The fix is to (a) name the era explicitly, (b) enumerate the required parts, and (c) enumerate the forbidden parts. Choosing kerosene vs. electric is only half the work — anatomy is the other half. See also `image-validator/knowledge/physical-realism-checks.md` checkpoint A9.9 for the validation-side detection routine.

package/templates/skills/image-prompt-engineer/knowledge/location-design.md

@@ -0,0 +1,236 @@
+# Location Design — Image Generation
+
+> How to describe environments for cinematic image prompts.
+
+---
+
+## 🔴 HARD RULE #0 — Location reference = EMPTY OF PEOPLE
+
+**In the 5-stage AI-creator pipeline, a LOCATION REFERENCE is an ANCHOR IMAGE of the empty environment.** People are added later, at the cell-generation stage, by combining the location reference with the relevant CHARACTER REFERENCES.
+
+```
+LOCATION REF  =  empty environment (room / street / hall / courtyard — NO living figures)
+                                +
+CHARACTER REF =  the specific person's identity (anchor portrait pack)
+                                +
+SCENE PROMPT  =  "this person stands in that location in this pose, doing this action"
+                                ↓
+                       FINAL KADR (cell image)
+```
+
+### Why this matters
+
+- **Cross-shot consistency:** the same location ref serves multiple kadrs (e.g. a parliament hall serves wide / medium / close-up variants — all with the same architectural identity). If people are baked into the location ref, every reuse forces the same figures to reappear.
+- **Character identity preservation:** characters come from their own anchor packs (precise face, hair, costume). If a generic "diplomat-looking man" is rendered inside a location ref, the model fights against the actual character anchor at cell stage.
+- **Modular asset library:** locations are reusable across many projects; characters are project-specific. Coupling them in one image destroys reuse.
+- **Group photo kadrs:** even kadrs that depict "a group of people in a room" should be built as (empty room ref) + (multiple character refs) + (group-composition prompt). The crowd is composed at cell stage, not pre-baked.
+
+### What IS allowed in a location reference
+
+The location ref describes the **physical environment a person could walk into**. Allowed:
+
+- **Architecture:** walls, ceilings, floors, columns, arches, doors, windows, stairs
+- **Furniture:** chairs, tables, sofas, beds, bookshelves, desks, podiums
+- **Décor:** rugs, curtains, paintings, mirrors, sculptures, flags on stands, wall plaques
+- **Lighting fixtures:** chandeliers, sconces, table lamps, candles (lit or unlit per scene logic — see physical-realism A9.7), street lamps
+- **Static infrastructure:** water taps, fireplaces, radiators, telephones on desks, typewriters at workstations, microphones on stands at empty podiums
+- **Stationary props / set dressing:** books, papers, glasses of water, fruit bowls, vases of flowers, ashtrays, period objects (qalamdan, samovar, gramophone, etc.)
+- **Natural elements:** trees, plants, sky visible through windows, water, terrain (in exteriors)
+- **Parked / stationary vehicles** that are NOT being operated: empty parked carriages, parked cars without drivers (if they're permanent set-dressing for the location)
+- **Animals only if permanent fixtures of the environment:** a stuffed mounted animal on a wall, a statue of a horse (a real living animal = not allowed unless the scene is literally about that animal)
+
+### What is NOT allowed in a location reference
+
+- **Any human figure** — no faces, no silhouettes, no shadows of people, no "distant figures" in deep background, no defocused person, no "hand at frame edge"
+- **Any operated vehicle** — no horse-cart with a driver, no carriage with passengers, no car being driven (these imply people)
+- **Any depicted "in-use" action** — no mid-stroke pen, no mid-strike typewriter, no half-eaten food on a plate, no mid-conversation tea-glasses with rising steam (these imply an off-frame person at this exact moment)
+- **Body-parts at frame edge** — no hand, no shoulder, no foot intruding from off-frame
+- **Crowd shadows** — no shadows on the wall that suggest off-frame people
+- **Anything implying recent human presence at a specific moment** — a half-smoked cigarette in an ashtray IS allowed if it's general set-dressing showing "this place is lived-in," but freshly-poured tea with rising steam IS NOT (too time-specific, implies the drinker just stepped away)
+
+### How to enforce in the prompt itself
+
+In the V7 expanded prompt format, add to the "NOTHING ELSE in the frame" section explicit exclusions:
+
+```
+NOTHING ELSE in the frame — NO human figures, NO silhouettes, NO defocused people in background,
+NO body parts at frame edge, NO operated vehicles, NO hands holding objects. This is a pure
+location/environment reference — people are composed in at cell-generation stage from
+character anchor packs.
+```
+
+### Exception — single-pass (non-pipeline) generation
+
+If you are generating a one-off image with NO multi-stage character refs (e.g. a stock illustration, a single editorial frame with no continuity needs), then location + people may live in one prompt. But the moment a project uses character anchor packs and plans multiple cells per location, the location reference MUST be empty per the rule above.
+
+### Common AI failure mode
+
+AI image models — trained heavily on photographs that include people — frequently insert "a small figure for scale" or "a distant pedestrian for atmosphere" even when the prompt doesn't ask for one. The "NO human figures" guard clause in NOTHING ELSE must be explicit and aggressive, otherwise the model adds people by default.
+
+**Detection during validation:** scan every location-ref output for any human figure (foreground, mid-ground, background, silhouette, shadow). Even one figure = ❌, regenerate with stronger exclusion.
+
+**Origin:** rule set by user 2026-05-21 during the 28-may documentary project, after L-01/L-02/L-03A were initially written WITH people embedded (horse-cart driver silhouette, stagehand behind curtain, 22+ schoolgirls posed in courtyard). All three location refs had to be rewritten as empty environments.
+
+### HARD RULE #0 — what about CELLS?
+
+**Clarification (added 2026-05-21):** HARD RULE #0 (empty location refs) applies **ONLY** to **L-XX location anchor refs**. It does **NOT** apply to cells.
+
+For **CELLS** (final kadr images, the actual frames the audience sees) at the cell-generation stage, the cell MUST include **all elements specified by the script** — horses, vehicles being driven, pedestrians, crowds, performers, hands holding objects, mid-action subjects, every living and moving thing the script calls for.
+
+### The distinction in plain language
+
+- **LOCATION REF (L-XX)** = empty stage set, no actors yet
+- **CELL** = the actual filmed shot, fully populated with actors, vehicles, animals, action per the script
+
+### Concrete example
+
+The script says: *"Şəkər küçəsi, səhər saat 8. Uzaqda at-arabası dayanıb, ətrafda ilk gəzərlər."* ("Şəkər street, 8 AM. A horse-cart waits in the distance, first pedestrians around.")
+
+- **L-01 (location ref):** empty Şəkər street — buildings, paving, gas lamps, awnings, posters on walls. **No horse, no cart, no pedestrians.** This ref is reusable across many cells set on Şəkər street.
+- **k.1 (cell):** Şəkər street **+** horse **+** cart **+** driver **+** ~3 pedestrians **+** mid-action moment. The cell is built from (L-01 location ref) **+** (character refs for each pedestrian and the driver) **+** (composition + action description).
+
+### Mistake to avoid
+
+Do NOT apply the empty-location rule to cells. That produces sterile, abstract scenes that contradict the script's intent — empty streets where the script said "first pedestrians," empty halls where the script said "audience applauding," empty offices where the script said "the editor types at the desk."
+
+### Quick decision tree
+
+```
+Is this asset an L-XX location anchor reference?
+├── YES → EMPTY of all living/operated elements (HARD RULE #0 above)
+└── NO (this is a cell / kadr) → POPULATE per script: every person, vehicle,
+                                  animal, action the script names is RENDERED
+                                  in the cell, composed from (L-XX) + (char refs).
+```
+
+### Cross-references
+
+- `composition.md` → "Cinematic lived moments" (how to populate cells naturally, not as posed photos)
+- `character-anatomy.md` → "Face diversity" + "Period + place costume research" (every figure in the cell needs an anchor and varied face)
+- `image-validator/knowledge/prompt-consistency-checks.md` F13 — F13 still applies to L-XX refs only
+
+**Origin of clarification:** 2026-05-21, 28-may documentary k.1 — cell was initially written as an empty street with a parked, horseless cart (because the prompt author wrongly applied the empty-location rule to a cell). This contradicted the script's "uzaqda at-arabası dayanıb." Rule clarified universally: empty applies to L-XX refs; cells are populated per script.
+
+---
+
+## MANDATORY location attributes
+
+For every location prompt, specify:
+
+### 1. Interior or Exterior (INT./EXT.)
+- INT. = inside
+- EXT. = outside
+
+### 2. Time of day
+- Dawn, morning, midday, afternoon, golden hour, dusk, night, blue hour
+- Or specific: "3 AM", "6:45 PM"
+
+### 3. Weather (if exterior or visible)
+- Clear, overcast, rain, snow, fog, storm
+- Wind level if relevant
+
+### 4. Period
+- Modern, 1950s, 1970s, 1980s, futuristic, indeterminate
+- Be specific to influence design
+
+### 5. Mood / Atmosphere (3-5 descriptors)
+- Melancholic, tense, romantic, oppressive, hopeful, nostalgic, sterile
+- Cozy, sterile, decaying, pristine, cluttered
+
+### 6. Key visual elements
+- Furniture types
+- Architectural features
+- Practical lighting sources
+- Props that matter
+
+### 7. Color palette
+- Dominant colors (warm wood, cool steel, faded green)
+- Accent colors (neon pink, bright orange)
+
+## Common location archetypes
+
+### Diner / Cafe
+- Booths, counter, neon sign
+- Warm tungsten lighting
+- Period reference (1960s American, modern European)
+- Menu boards, jukebox, coffee machine
+
+### Apartment Interior
+- Style (minimalist, cluttered, family, bachelor)
+- Furniture period
+- Wall color and art
+- Window light direction
+
+### Office
+- Open plan vs cubicle vs private office
+- Monitor light, fluorescent vs natural
+- Era (1990s vs modern)
+
+### Street / Alley
+- City type (Tokyo, NYC, Baku, Paris)
+- Time period
+- Wet (rain reflections) vs dry
+- Practical signs, neon, streetlamps
+
+### Forest / Nature
+- Tree type
+- Season
+- Time of day (filtered light)
+- Ground cover (leaves, snow, moss)
+
+### Bedroom
+- Made vs unmade bed
+- Wall color
+- Window light
+- Personal items (period-specific)
+
+## Lighting consideration per location
+
+Always describe how light enters and what it touches:
+- "Soft afternoon light through gauze curtain illuminates dust particles"
+- "Cool blue moonlight through window casts grid of frame on bedsheets"
+- "Warm tungsten ceiling lamp pools light on table, edges of room dim"
+
+## Sample full location description
+
+```
+INT. 1960s American diner at night, melancholic and sleepy atmosphere.
+Long counter with chrome stools, padded leather booths along window
+wall, period jukebox glowing in corner. Walls in cream tile and dark
+wood paneling. Practical lighting: warm tungsten ceiling pendants
+(3200K) creating pools of light, neon "OPEN" sign in window casting
+red glow, cool blue spill from rain-streaked window. Empty except
+for a single coffee cup and crumpled newspaper on counter. Color
+palette: warm browns and creams contrasted with cool blue exterior.
+Period accuracy: chrome napkin holders, vintage Coca-Cola signs,
+pressed tin ceiling.
+```
+
+## Multi-shot prompts per location
+
+Each location should have 3 prompts:
+
+### Establishing shot
+- Wide perspective showing whole space
+- Subject(s) small or absent
+- Architectural emphasis
+- Lens: 24-35mm, f/5.6+
+
+### Wide shot
+- Showing space + subject(s)
+- Subject relationship to environment
+- Lens: 35-50mm, f/4
+
+### Medium shot
+- Zoom-in on functional area
+- Subject(s) and immediate surroundings
+- Lens: 50mm, f/2.8
+
+## DON'T
+
+- "Nice cafe"
+- "Cozy room"
+- "Modern office"
+- "Beautiful nature"
+- Generic descriptions without period/style
+- Lighting without source attribution

package/templates/skills/image-prompt-engineer/knowledge/models/flux-pro.md

@@ -0,0 +1,87 @@
+# Flux Pro — Image Model Knowledge
+
+## Provider
+Black Forest Labs
+
+## Type
+Diffusion (open architecture, hosted)
+
+## Strengths
+
+- Extreme detail rendering
+- Excellent text rendering
+- Realistic skin texture (better than most)
+- Open ecosystem (LoRAs, ControlNet)
+- Strong prompt adherence
+- Negative prompts supported
+- Good for hero / detailed shots
+
+## Weaknesses
+
+- Slow generation (~20-30s for highest quality)
+- Expensive per image
+- Reference image complexity (requires control nets)
+- Photorealism can be too "perfect" / uncanny
+- Less out-of-box cinematic look (needs styling)
+
+## Best Practices
+
+1. Detailed natural language
+2. Use weight syntax: `(((emphasis)))` for important elements
+3. Negative prompts work — list what to avoid
+4. Specify lighting carefully (Flux interprets very literally)
+5. Use camera + lens for cinematic feel
+6. Avoid contradictions (Flux is literal-minded)
+7. Long prompts (200+ words) work well
+8. Reference film stocks or photo styles by name
+
+## Prompt Anatomy
+
+```
+[Detailed subject with anchor features]
++ [Specific action and pose]
++ [Environmental detail]
++ [Lighting sources with color temperature]
++ [Camera + lens + aperture + style]
++ [Reference style: photographer, film stock]
+
+Negative: [unwanted elements, common artifacts]
+```
+
+## Sample Prompt
+
+```
+Photorealistic cinematic portrait: a 35-year-old Azerbaijani man with
+((athletic build)), ((short dark wavy hair)), green eyes, clean-shaven
+with light stubble, wearing a charcoal wool overcoat over a plain
+white shirt. Standing in a 1960s diner interior, hands in coat
+pockets, looking out a rain-streaked window with melancholic
+expression, eyes slightly downcast. Lighting: warm tungsten ceiling
+pendant at 3200K from above-left creating soft rim on shoulders,
+cool blue daylight spill from window at 5600K behind him as fill,
+practical neon sign just out of frame casting subtle red glow on
+right side. Shot on ARRI Alexa Mini LF with Cooke S4 50mm at f/2.0,
+medium close-up, eye-level, 9:16 vertical aspect. Style: cinematic,
+shallow depth of field, vintage Kodak Portra 400 emulation, slightly
+desaturated tones, slight film grain.
+
+Negative: cartoon, anime, illustration, painting, oversaturated,
+plastic skin, blurry, deformed hands, extra fingers, watermark, text.
+```
+
+## Negative Prompt Common Items
+
+For photorealism, always negative-prompt:
+- cartoon, anime, illustration, painting, drawing
+- 3d render, CGI, video game
+- oversaturated, plastic, fake
+- blurry, low quality, distorted, deformed
+- extra fingers, missing fingers
+- watermark, signature, text
+- amateur, snapshot
+
+## Pricing tier
+$$$$ — highest among major models
+
+## Speed tier
+Slow — 20-30s