@azerogluemin/ai-bootstrap 0.2.1

package/templates/skills/video-prompt-engineer/knowledge/models/sora.md

@@ -0,0 +1,116 @@
+# Sora — Video Model Knowledge
+
+## Provider
+OpenAI
+
+## Type
+Transformer-based diffusion
+
+## Strengths
+
+- **Long-form** — 1+ minute clips
+- Strong physics simulation
+- Narrative understanding
+- Storyworld generation
+- Complex scenes possible
+- Multi-shot stitching attempts
+- Audio support (improving)
+
+## Weaknesses
+
+- **Limited access** — invite/waitlist
+- Very expensive
+- Slow generation
+- API access restricted
+
+## Best Practices
+
+1. **Narrative description** — describe like writing prose
+2. Less technical, more storytelling
+3. Establish setting first, then action
+4. Specify duration explicitly
+5. Camera direction in narrative form
+6. Multi-shot scenes possible
+
+## Prompt Anatomy
+
+```
+[Narrative scene description — setting, mood, atmosphere first]
+
+[Subject established in scene — who, what they wear, what they're doing]
+
+[Action — what unfolds during the clip]
+
+[Camera and time direction]
+
+[Audio if any]
+
+Duration: Ns
+```
+
+## Sample Prompt
+
+```
+A medium-shot of a 35-year-old man in a charcoal coat standing alone
+by a window in a 1960s diner. The exterior shows heavy rain falling
+under streetlamp light. The interior is warmly lit by tungsten ceiling
+pendants. He has been waiting. After holding still for a moment of
+quiet contemplation, he slowly turns his head to the left, his eyes
+finding the door. The camera holds steady throughout, observing.
+
+[Audio: heavy rain on the window glass, distant traffic, soft jazz
+playing from a jukebox somewhere off-screen, the man's quiet,
+shallow breathing.]
+
+The shot lasts 8 seconds.
+```
+
+## Long-Form Capability
+
+Sora can generate longer clips than competitors. Use this for:
+- Continuous action sequences
+- Long takes (one-shot scenes)
+- Establishing sequences
+- Documentary-style observation
+
+```
+A 30-second continuous shot following a young woman as she walks
+through a Tokyo neon-lit alleyway at night. She passes vendors,
+glances at signs, occasionally smiles at strangers. Camera tracks
+alongside her at medium distance, smooth steadicam motion. Practical
+neon and shop lights illuminate her face in shifting colors as she
+moves. Ambient city sounds throughout.
+
+Duration: 30 seconds.
+```
+
+## Multi-Shot Mode
+
+Sora attempts to handle scene with multiple shots:
+```
+Scene: Two people meet at a cafe.
+Shot 1 (4s): Wide establishing shot of cafe exterior, woman approaches.
+Shot 2 (3s): Medium shot, woman opens door, enters.
+Shot 3 (4s): Two-shot interior, she sees man at table, he stands up.
+Total duration: 11 seconds.
+```
+
+Results vary; for production-quality multi-shot, generate separate clips.
+
+## Aspect Ratios
+
+- 16:9, 9:16, 1:1
+- Wide aspect (2.39:1) supported
+
+## Common Mistakes
+
+- Treating like image generator (Sora wants narrative)
+- Too short prompts (Sora rewards detail)
+- Mixing styles within prompt
+- Forgetting duration
+
+## Pricing tier
+$$$$$ — highest
+
+## Speed tier
+Slow — several minutes per clip

package/templates/skills/video-prompt-engineer/knowledge/models/veo-3.md

@@ -0,0 +1,95 @@
+# Veo 3.1 — Video Model Knowledge
+
+## Provider
+Google DeepMind
+
+## Type
+Diffusion + transformer hybrid
+
+## Strengths
+
+- **Audio sync supported** via `[Audio: ...]` syntax — UNIQUE among models
+- Natural language prompts work best
+- Up to 8 seconds with high stability
+- Excellent for long static takes
+- Strong physics simulation
+- Photorealistic skin, lighting, fabric
+- Cinematic look out-of-box
+
+## Weaknesses
+
+- Cannot render text, signs, captions
+- Fast motion creates artifacts
+- Camera move + subject move simultaneously is hard
+- Sometimes adds "magic" elements not requested
+- Limited to 8s clips
+
+## Best Practices
+
+1. **Story-driven natural language** — write like screenplay action
+2. Specify "static" for camera when possible
+3. Audio cues in `[Audio: ...]` brackets
+4. 6-8 seconds optimal duration
+5. "Slowly" prefer over instant verbs
+6. NO negative prompts — just don't mention unwanted
+7. Reference cinematic style, not photo style
+8. Specify single camera movement if any
+
+## Prompt Anatomy
+
+```
+[Camera setup — angle, shot type, movement]
+
+[Subject + action description in natural language]
+
+[Audio: ambient sounds, specific sounds, dialogue]
+
+Lighting: [setup description]
+Duration: Ns
+```
+
+## Sample Prompt
+
+```
+[Static medium shot, eye-level]
+
+A 35-year-old man in a charcoal coat stands by a rain-streaked window
+in a dimly lit 1960s diner. He slowly turns his head left to look
+toward the entrance. Subtle breathing motion. Camera holds steady — 
+no movement.
+
+[Audio: heavy rain on glass, distant traffic, soft jazz from jukebox,
+his quiet exhale]
+
+Lighting: warm tungsten from above, cool blue from window.
+Duration: 6 seconds.
+```
+
+## Audio Cue Examples
+
+```
+[Audio: city ambient — distant traffic, occasional horn, street vendor calling]
+[Audio: woman's voice, soft, urgent: "We need to leave now."]
+[Audio: piano playing slow melancholic melody, minor key, distant]
+[Audio: silence except wind and breathing]
+```
+
+## Common Mistakes
+
+- Asking for text rendering ("sign reading 'OPEN'") — Veo can't
+- Multiple simultaneous camera moves
+- Action verbs without "slowly" or timing
+- Too much happening in 4 seconds
+- Negative prompts (use positive descriptions instead)
+
+## Aspect Ratios
+
+- 16:9 — primary
+- 9:16 — supported, specify in prompt
+- 1:1 — supported
+
+## Pricing tier
+$$$$ — premium
+
+## Speed tier
+Medium — 30-60s per clip

package/templates/skills/video-prompt-engineer/knowledge/motion-vs-static.md

@@ -0,0 +1,229 @@
+# Motion vs Static — Video Generation
+
+> When to use static cameras vs movement. The most important creative decision in AI video, given current model limitations.
+
+## Default: STATIC
+
+For AI video in 2026, static camera is almost always the safer choice:
+- Most stable output
+- Lowest artifact risk
+- All models excel
+- Subject motion can still be present
+
+## When STATIC is right
+
+### Dialogue scenes
+- Two people talking
+- Single character monologue
+- Voiceover with character listening
+
+### Reaction shots
+- Emotional close-ups
+- Subject responding to news
+- Face changing expression
+
+### Atmospheric establishing
+- Wide shot of location
+- Subject placed in environment
+- Subtle ambient motion (curtain, smoke)
+
+### Product shots
+- Hero product views
+- 360 rotations should be done in post
+- Static highlights texture and shape
+
+### Contemplative moments
+- Subject thinking, looking
+- Lone figure in landscape
+- Quiet moments
+
+## When MOVEMENT is justified
+
+### Following action
+- Subject walking through space
+- Tracking shot accompanying motion
+- Camera follows the eye
+
+### Revealing information
+- Pan to reveal character
+- Tilt up to reveal scale
+- Dolly out to show context
+
+### Building tension
+- Slow dolly in on subject
+- Push toward subject's eyes
+- Crane down into scene
+
+### Stylistic intent
+- Music video aesthetic
+- Documentary handheld
+- Whip pans for energy
+
+## Movement complexity ladder
+
+From safest to riskiest:
+
+### Level 1: Static (safest)
+```
+Camera: static, no movement.
+```
+
+### Level 2: Single slow movement
+```
+Camera: slow dolly in over 6 seconds, no other movement.
+```
+
+### Level 3: Single medium movement
+```
+Camera: pan right 45° following subject's gaze.
+```
+
+### Level 4: Combined movement (risky)
+```
+Camera: subtle handheld tracking shot following subject walking.
+```
+
+### Level 5: Complex (failure-prone)
+```
+Camera: crane down combined with slow pan and gradual zoom in.
+```
+
+Don't go above Level 3 unless you understand the model's limits.
+
+## Subject motion within static frame
+
+A static camera doesn't mean a static scene. Subject can:
+- Move within frame
+- Enter/exit frame
+- Change pose
+- Express emotion
+- Interact with props
+
+Static camera + subject motion = most successful AI video formula.
+
+## Movement direction guide
+
+### Camera moves IN (dolly/zoom in)
+- Increases intimacy
+- Builds tension
+- Focuses attention
+- "Coming closer to truth"
+
+### Camera moves OUT (dolly/zoom out)
+- Increases context
+- Releases tension
+- Reveals isolation
+- "Stepping back from intensity"
+
+### Camera moves UP (crane/tilt up)
+- Reveals scale
+- Hopeful/aspirational
+- Subject becomes smaller
+- Heavenward / future
+
+### Camera moves DOWN (crane/tilt down)
+- Lowers status
+- Defeated / grounded
+- Earthbound / reality
+- Subject becomes accessible
+
+### Camera moves LEFT or RIGHT (pan/track)
+- Following action
+- Comparing two things
+- Westward (left) tends to feel like ending
+- Eastward (right) tends to feel like beginning
+
+## Prompt phrasing
+
+### For static
+```
+Camera: completely static, no movement.
+```
+
+### For dolly in
+```
+Camera: medium shot at start. Slow dolly in over the duration of the
+clip, ending in close-up. No pan, tilt, or zoom — only forward
+movement.
+```
+
+### For tracking
+```
+Camera: medium shot. Tracks right alongside the subject as they walk,
+maintaining same distance and framing throughout.
+```
+
+### For crane up
+```
+Camera: starts at eye-level. Slowly cranes up over 6 seconds, ending
+in high-angle bird's-eye view of the scene below.
+```
+
+## Decision checklist
+
+Before adding movement, ask:
+1. Does the scene require it? (or is it tradition/habit?)
+2. Will the model handle this movement reliably?
+3. Is the duration enough for the movement to feel natural?
+4. Will subject motion suffice instead?
+5. Can I add this in post (zoom in editing)?
+
+If answer to any is doubtful → use static.
+
+## Default for AI video
+
+When unsure: **static**. Always.
+
+---
+
+## Ken Burns archive / photo-doc — interpreting "frozen photo" cues correctly
+
+When a script says **"frozen photo"**, **"still photograph"**, **"donmuş zaman"**, **"the image is not alive"**, or similar poetic cues, this is **NEVER a literal taxidermy command.** Do not write a prompt that forbids breathing, blinking, fabric in wind, animal life signs, or natural ambient motion. A scene with zero subject motion looks like a CGI freeze-frame, not an archival photograph, and breaks the documentary spell the script is trying to create.
+
+### What "frozen photo" actually means in documentary grammar
+
+The Ken Burns / Adam Curtis / Errol Morris tradition has a precise meaning for this:
+
+| Yes (preserve) | No (avoid) |
+|---|---|
+| Camera as primary motion (slow push, pan, tilt) | Subject driving the narrative through action |
+| Natural micro-life (breath, blink, hair, fabric, ambient sway, light flicker) | Dramatic action (running, gesturing, shouting, turning to camera) |
+| Subjects continuing what they were already doing (walking, waiting, looking) | Subjects changing what they're doing mid-clip (sudden gestures, reactions) |
+| Soft ambient motion (dust in light, breeze, drifting smoke) | Acted "events" that didn't exist in the still frame |
+| Animals breathing / shifting weight / swiveling ears | Animals bolting, rearing, walking out of frame |
+| Pedestrians continuing to walk at their original pace | Pedestrians stopping, turning back, addressing the camera |
+
+The script's intent is observational stillness — the world keeps existing, but no one performs for the camera. **Life continues; drama doesn't.**
+
+### Prompt phrasing for "frozen photo" cues
+
+GOOD — preserves the breath of life:
+```
+Action — natural observational life, no drama, no camera awareness:
+- The horse breathes gently, shifts weight, mane catches the breeze
+- The driver sits with the reins, glances down the street once
+- Distant pedestrians continue walking at the same calm pace
+- A curtain stirs in a window, dust drifts in the backlight
+- The lantern flame flickers softly
+
+Camera: very slow continuous push-in. No subject performs for camera.
+```
+
+BAD — kills the scene (literal interpretation):
+```
+NO subject motion of any kind. The horse does not breathe, blink, or shift.
+The driver does not move. Pedestrians remain locked in mid-stride. No fabric
+flutter, no dust, no wind, no flame flicker. This is a still photograph;
+only the camera moves.
+```
+
+The "BAD" version is what an inexperienced prompt writer produces when they translate "frozen photo" word-for-word. It looks dead on screen.
+
+### When literal full-freeze IS correct
+
+There are exactly two cases where every subject must be truly motionless:
+
+1. **Photo of a photo** — the camera is filming a printed photograph hanging on a wall, lying on a table, or being held in hand. The photograph is an object inside the scene; objects don't move.
+2. **Title-card / typography / inanimate-only frames** — manuscript pages, documents, signs, statues with no living subject in frame.
+
+For frames containing living subjects (people, animals) in an archival aesthetic, micro-life is mandatory, not optional.