@lemoncat7/dsh-partner 2.5.0 → 2.7.1

package/resources/skills/director-skills/upstream/skills/cinematic-video-prompting/SKILL.md

@@ -0,0 +1,84 @@
+---
+name: cinematic-video-prompting
+description: "Generate text-to-video and image-to-video prompts with shot timing, subject motion, camera movement, lensing, lighting, audio cues, continuity anchors, and model-aware constraints."
+---
+
+# cinematic-video-prompting
+
+## When to use
+- The user wants text-to-video, image-to-video, first/last-frame, video extension, multi-shot, audio-aware, or motion-control prompts.
+- The user has a shot list or still image and wants to animate it.
+- The task requires duration math, camera movement, subject motion, physical consequence, lip sync, or video-specific failure prevention.
+
+## When not to use
+- The user only needs a still image prompt.
+- The user is still developing story with no visual beat.
+- The user wants editing of a real video file rather than prompt design.
+
+## Required inputs
+- Scene or shot goal
+- Target duration
+- Subject and action
+- Camera/framing
+- Continuity anchors
+
+## Optional inputs
+- Reference image/video/audio roles
+- Target model
+- Aspect ratio/resolution
+- Dialogue line
+- Audio design
+- Negative constraints
+
+## Workflow
+1. Identify the shot purpose and emotional turn.
+2. Respect the clock: choose one dominant action and one dominant camera behavior for short clips unless the model explicitly supports multi-shot.
+3. Write time blocks or beat blocks when the model benefits from sequential structure.
+4. Pair every action with physical consequence: weight, friction, follow-through, environmental reaction, or sound.
+5. Specify camera movement, lens feel, framing, lighting, palette, texture, and realism guardrails.
+6. For I2V, preserve source composition and describe only motion, camera, lighting shift, and audio.
+7. Add audio layers where supported: dialogue, ambience, foley/SFX, and music mood.
+8. Add edit/post notes when this prompt belongs to a larger AI short-film package: take count, cut point, color-grain unification, and audio bed.
+9. Route to model-adaptation when a named model is involved.
+
+## Decision logic
+- If a prompt contains two or more big actions, split into shots.
+- If identity drift is likely, generate/reference a character sheet first.
+- If hands, mirrors, text, or logos are central, redesign framing or use a more suitable model.
+- If the model supports negative prompts, cap constraints to the few failures that would ruin the clip.
+
+## Output formats
+- Universal video prompt brief
+- Text-to-video prompt
+- Image-to-video prompt
+- Time-blocked prompt
+- Audio cue block
+- Continuity guardrails
+- Model-adaptation handoff
+
+## Quality checks
+- Duration, shot count, camera movement, and action complexity are compatible.
+- Motion is concrete and continuous, not a vague mood.
+- Subject and camera motion do not conflict.
+- Continuity anchors protect identity, wardrobe, props, and location state.
+- Known failure modes are addressed without over-constraining.
+
+## Anti-patterns
+- Trying to direct a full film in one prompt
+- Multiple camera moves in one short clip
+- Long dialogue in an 8-second clip
+- Describing an uploaded image again instead of animating it
+- Using negative prompts as a dumping ground for every fear.
+
+## Exit criteria
+- The video prompt is ready for the target workflow or for model-specific export.
+
+## Supporting files
+Read only the supporting file needed for the active task:
+- `references/video_prompt_components.md`
+- `references/camera_lens_movement.md`
+- `references/audio_motion_timing.md`
+- `references/ai_film_production_pipeline.md`
+- `references/prompt_anatomy.md`
+- `templates/text_to_video_prompt.md`
+- `templates/image_to_video_prompt.md`

package/resources/skills/director-skills/upstream/skills/cinematic-video-prompting/checklists/video_prompt_quality_checklist.md

@@ -0,0 +1,9 @@
+# Video Prompt Quality Checklist
+
+- Duration and shot count are realistic.
+- One action and one camera move dominate each shot.
+- Motion has physical consequence.
+- Lighting/style/texture are filmable.
+- Audio layers are included when supported.
+- I2V prompts preserve source image instead of redescribing it.
+- Constraints target the top failure risks only.

package/resources/skills/director-skills/upstream/skills/cinematic-video-prompting/examples/image_prompt_to_video_prompt.md

@@ -0,0 +1,3 @@
+# Example: Image Prompt to Video Prompt
+
+Use the uploaded keyframe as the exact visual anchor. Preserve the brass key, the guard's hand position, chipped green paint, and cool fluorescent palette. Animate the key trembling once, then turning slowly in the lock. Camera remains macro and locked-off. The metal clicks, cloth sleeve shifts softly, fluorescent hum continues. End with the key fully turned. No extra fingers, no changing key shape, no background morphing.

package/resources/skills/director-skills/upstream/skills/cinematic-video-prompting/references/ai_film_production_pipeline.md

@@ -0,0 +1,21 @@
+# AI Film Production Pipeline
+
+Use this when video prompts are part of a finished short-film workflow rather than isolated clips.
+
+## Practical sequence
+
+1. Concept and script breakdown: logline, treatment, beat sheet, screenplay, shot list.
+2. Visual lock: golden images for characters, locations, props, costumes, and palette.
+3. Storyboard: verify composition and sequence before spending generations.
+4. Batch generation: create 2-3 variations per shot, favor I2V when continuity matters, keep shots short.
+5. Audio generation: dialogue, ambience, foley, and music or temp score.
+6. Assembly: cut clips together; trim weak AI reaction time and avoid lingering on artificial micro-expression gaps.
+7. Post polish: uniform color grade, consistent grain/texture, subtitles/captions, upscale if needed.
+8. Release assets: posters, vertical trailer, thumbnails, social clips.
+
+## Prompt implications
+
+- Every video prompt should include an editing note: where it starts, where it ends, and how it cuts.
+- Long scenes should be broken into controllable shots.
+- Keep audio beds active; silence exposes synthetic footage.
+- Track rights/provenance for voices, music, reference images, and AI outputs when release is intended.

package/resources/skills/director-skills/upstream/skills/cinematic-video-prompting/references/audio_motion_timing.md

@@ -0,0 +1,16 @@
+# Audio, Motion, and Timing
+
+## Audio layers
+- Dialogue: exact quoted line, speaker, emotional delivery, language, lip-sync note.
+- Foley/SFX: footsteps, cloth, glass, rain, machinery, door, breath.
+- Ambience: room tone, traffic, crowd murmur, wind, insects, fluorescent hum.
+- Music: mood and evolution, not copyrighted song titles.
+
+## Motion realism
+Write physical verbs and consequences. A fall has weight; a cup spill changes the table; a turn shifts fabric and hair; a door slam creates vibration and sound.
+
+## Duration guidance
+- 4 seconds: one gesture or simple reveal.
+- 6 to 8 seconds: one action plus consequence.
+- 10 to 15 seconds: one compact beat, or simple multi-beat if supported.
+- Longer narratives: chain clips with final-frame/first-frame continuity.

package/resources/skills/director-skills/upstream/skills/cinematic-video-prompting/references/camera_lens_movement.md

@@ -0,0 +1,26 @@
+# Camera, Lens, and Movement
+
+## Lens language
+- 18-24mm: wide, immersive, spatial distortion, kinetic proximity.
+- 28-35mm: naturalistic documentary presence, environmental context.
+- 50mm: balanced human perspective, restrained drama.
+- 75-100mm: portrait compression, isolation, emotional scrutiny.
+- Macro: tactile detail, evidence, fragility, product texture.
+- Anamorphic: wide cinematic frame, oval bokeh, horizontal flares; use only when the wide format supports the scene.
+
+## Camera movement
+Use one dominant move per short video prompt.
+- Locked-off: tension, deadpan comedy, surveillance, realism.
+- Slow dolly-in: attention, realization, dread, intimacy.
+- Pull-back: reveal, loss, loneliness, scale.
+- Tracking: pursuit, routine, journey, kinetic momentum.
+- Handheld: immediacy, documentary energy, instability.
+- Crane/rise: revelation, scale, transcendence.
+- Orbit: subject examination, glamour, uncertainty; risky for identity in weaker models.
+- Pan/tilt: discovery, geography, causal reveal.
+
+## Movement guardrails
+- Pair camera motion with subject motion only when they do not compete.
+- For AI video, prefer simple verbs: walks, turns, reaches, kneels, lifts, exhales.
+- State physical consequence: coat swings, glass trembles, dust lifts, water ripples.
+- Avoid mirrors, complex hand interactions, crowds, and fast rotations unless essential.

package/resources/skills/director-skills/upstream/skills/cinematic-video-prompting/references/negative_prompt_patterns.md

@@ -0,0 +1,18 @@
+# Negative Prompt Patterns
+
+Do not assume negative prompts work in every model.
+
+## If the model supports a negative field
+Use only the few failures that would ruin the output. Prefer concrete artifacts: extra fingers, warped hands, text artifacts, logo distortion, identity drift, blurry face, morphing background.
+
+## If the model does not support negatives
+Restate the desired positive condition:
+- Instead of "no blur": sharp focus on the subject's eyes.
+- Instead of "no extra people": single person alone in the frame.
+- Instead of "no morphing": stable face shape and unchanged wardrobe throughout.
+
+## Video constraint cap
+For video, limit constraints to the top 2 to 4 risks. Over-constraining can create prompt noise.
+
+## Safety/IP hygiene
+Do not provide bypass tactics. Replace named protected characters/logos with visual descriptions unless the user has a legitimate, allowed use case and the model supports it.

package/resources/skills/director-skills/upstream/skills/cinematic-video-prompting/references/prompt_anatomy.md

@@ -0,0 +1,18 @@
+# Prompt Anatomy
+
+Use the three-layer model.
+
+## Layer 1: Creative intent
+Story purpose, character emotion, theme pressure, audience experience, genre, and scene function.
+
+## Layer 2: Cinematic execution
+Shot type, framing, camera angle, lens, lighting, blocking, production design, texture, realism level, audio if video, and continuity anchors.
+
+## Layer 3: Model-specific adaptation
+Prompt syntax, length, parameters, negative support, reference handling, aspect ratio, duration, resolution, seed/reference controls, known failure modes.
+
+## Front-loading rule
+Place the identity, action, setting, and desired medium in the opening sentence. Put secondary style and constraints later.
+
+## Prompt economy
+Remove filler such as "we see" or repeated adjectives. Keep every clause tied to a visible or audible result.

package/resources/skills/director-skills/upstream/skills/cinematic-video-prompting/references/video_prompt_components.md

@@ -0,0 +1,25 @@
+# Video Prompt Components
+
+## Universal video stack
+1. Format: T2V, I2V, first/last frame, extension, editing, multi-shot.
+2. Duration and aspect ratio.
+3. Subject and stable identity anchors.
+4. One dominant action per short shot.
+5. One dominant camera behavior per short shot.
+6. Framing, angle, and lens feel.
+7. Lighting, color, and texture.
+8. Motion physics and consequences.
+9. Audio: dialogue, ambience, foley/SFX, music mood when supported.
+10. Continuity guardrails and model-appropriate constraints.
+
+## Time blocking
+Use time blocks when the model/source guide supports sequential behavior:
+- [0:00-0:02] establish subject and movement.
+- [0:02-0:05] action and camera continue.
+- [0:05-0:08] consequence/reveal/end state.
+
+## I2V rule
+Preserve the uploaded image. Do not redescribe every visual detail unless the target model asks for it. Focus on motion, camera behavior, lighting change, audio, and what must not change.
+
+## Dialogue math
+Short clips cannot carry long lines. Prefer one sentence, natural lip sync, and visible performance beats. If dialogue is not essential, use sound or behavior.

package/resources/skills/director-skills/upstream/skills/cinematic-video-prompting/templates/image_to_video_prompt.md

@@ -0,0 +1,5 @@
+# Image-to-Video Prompt Template
+
+Use the uploaded image as the exact visual anchor. Preserve character identity, wardrobe, composition, location geometry, palette, and object placement.
+
+Animate: [subject performs one clear action]. Camera: [one subtle movement or locked-off]. Lighting/atmosphere: [maintain or shift]. Motion physics: [specific consequence]. Audio: [ambience/foley/dialogue if supported]. End state: [clear final visual]. Constraints: no identity drift, no wardrobe change, no background morphing, no extra subjects unless intended.

package/resources/skills/director-skills/upstream/skills/cinematic-video-prompting/templates/text_to_video_prompt.md

@@ -0,0 +1,5 @@
+# Text-to-Video Prompt Template
+
+FORMAT: [duration]s / [aspect ratio] / [single shot or shot count]
+
+Shot: [subject] [one continuous action] in [setting]. Camera: [one movement or locked-off], [framing/lens feel]. Lighting: [source, direction, quality, color]. Motion physics: [weight, friction, environmental consequence]. Style/texture: [realism, grain, surface details]. Audio: [dialogue if any], [ambience], [foley/SFX], [music mood if supported]. Continuity: [identity, wardrobe, prop, location anchors]. Constraints: [top 2-4 model-appropriate constraints].

package/resources/skills/director-skills/upstream/skills/cinematic-video-prompting/templates/video_beat_prompt.md

@@ -0,0 +1,9 @@
+# Time-Blocked Video Prompt Template
+
+FORMAT: [duration]s / [aspect ratio]
+
+[0:00-0:02] [Establish subject, framing, action start.]
+[0:02-0:05] [Action continues; camera behavior; physical consequence.]
+[0:05-0:08] [Revealed result or emotional end state.]
+
+Lighting: [source/direction/quality]. Audio: [layers]. Continuity: [anchors]. Constraints: [top risks only].

package/resources/skills/director-skills/upstream/skills/cinematic-video-prompting/tests/video_prompt_completeness_tests.md

@@ -0,0 +1,5 @@
+# Video Prompt Completeness Tests
+
+- Prompt includes duration, modality, subject, one dominant action, one dominant camera behavior, lighting, motion physics, audio if supported, continuity, and constraints.
+- I2V prompt preserves source image and focuses on motion.
+- Dialogue length is plausible for duration.

package/resources/skills/director-skills/upstream/skills/model-adaptation/SKILL.md

@@ -0,0 +1,82 @@
+---
+name: model-adaptation
+description: "Translate universal cinematic image/video briefs into model-specific prompt formats, settings, constraints, and export packages for image and video models including Grok/Aurora, Gemini/Nano Banana, GPT-image, Le Chat/FLUX, Kling, Seedance, Veo, Sora, Runway, Pika, Luma, Midjourney, Stable Diffusion, Ideogram, Hailuo, and unknown tools."
+---
+
+# model-adaptation
+
+## When to use
+- The user names a specific image or video model/tool.
+- The user asks to convert a prompt from one model to another.
+- The user asks why a prompt works in one model but fails in another.
+- The user wants best settings, a model-agnostic package, or exports for multiple tools.
+- The prompt uses model-dependent features: negative prompts, reference images, seeds, duration, aspect ratio, resolution, camera control, audio, or text rendering.
+
+## When not to use
+- The user only wants story ideation with no target tool.
+- The user asks for unsafe policy bypasses.
+- The user wants live, up-to-date official specs and source verification; in that case browse or check current docs before asserting changed capabilities.
+
+## Required inputs
+- Universal prompt brief or existing prompt
+- Target model(s)
+- Modality: image, T2V, I2V, editing, extension, upscaling, etc.
+- Desired aspect ratio/duration/resolution if relevant
+
+## Optional inputs
+- Reference assets and their roles
+- Negative constraints
+- Seed/settings
+- Known failure from previous output
+- Current docs supplied by user
+
+## Workflow
+1. Parse the universal brief into creative intent, cinematic execution, and model-dependent controls.
+2. Look up the relevant model profile and capability matrix. If a capability is not in the references, mark it unknown rather than guessing.
+3. Choose prompt structure: natural prose, slot template, time-blocked prompt, shot list, or parameterized export.
+4. Translate negatives correctly: separate field where supported, positive restatement where unsupported, capped constraints for video.
+5. Map reference assets to supported syntax and roles. If unsupported, describe a workaround or mark unknown.
+6. Map settings: aspect ratio, duration, resolution, seed/reference controls, audio, and text/logo constraints.
+7. Produce a translation report noting preserved intent, changed syntax, unsupported fields, unknowns, and likely failure modes.
+8. For multi-model exports, keep one universal brief plus per-model prompts/settings.
+
+## Decision logic
+- If the target model is absent from profiles, create a conservative generic export and mark all capabilities unknown.
+- If the prompt depends on text/logos, route to models with stronger documented text support or simplify text.
+- If video motion is overloaded, simplify before model export.
+- If a source guide conflicts with another, prefer model-specific guide for that model and record confidence.
+
+## Output formats
+- Model-specific prompt export
+- Model comparison table
+- Prompt translation report
+- Multi-model export package
+- Unknown capability log
+- Failure-mode warning
+
+## Quality checks
+- No unsupported capability is invented.
+- The adapted prompt preserves creative intent while changing syntax for the model.
+- Settings are separated from prompt text where the target supports parameters.
+- Reference and negative-prompt behavior is explicit.
+- The export is ready to paste or hand to an API/tool.
+
+## Anti-patterns
+- Using one universal prompt format for all models
+- Assuming negative prompts work everywhere
+- Guessing current model specs from memory
+- Mixing source-image preservation with redescribing the whole image for I2V
+- Providing safety or filter bypass instructions.
+
+## Exit criteria
+- The user has model-specific prompt(s), settings, unknowns, and a translation rationale.
+
+## Supporting files
+Read only the supporting file needed for the active task:
+- `references/model_capability_matrix.md`
+- `references/image_model_profiles.md`
+- `references/video_model_profiles.md`
+- `references/prompt_translation_rules.md`
+- `references/parameter_compatibility.md`
+- `references/failure_modes_by_model.md`
+- `references/model_update_protocol.md`

package/resources/skills/director-skills/upstream/skills/model-adaptation/checklists/model_export_checklist.md

@@ -0,0 +1,9 @@
+# Model Export Checklist
+
+- Target model and modality are named.
+- Prompt syntax matches the model profile.
+- Settings are separate where supported.
+- Negative/reference/seed/aspect/duration support is explicit.
+- Unsupported fields are removed or rewritten.
+- Unknowns are marked unknown.
+- Failure modes and fallback are included.

package/resources/skills/director-skills/upstream/skills/model-adaptation/examples/failed_model_transfer_diagnosis.md

@@ -0,0 +1,5 @@
+# Example: Failed Model Transfer Diagnosis
+
+A prompt written for Kling used a separate negative prompt and start/end frame workflow. Transferred directly to Le Chat image generation, the negatives appeared to confuse the prompt and the video controls were ignored.
+
+Fix: convert to a still-image prompt, remove video-only settings, rewrite negatives as positives, and mark start/end frame unsupported for that target.

package/resources/skills/director-skills/upstream/skills/model-adaptation/examples/image_prompt_translation.md

@@ -0,0 +1,10 @@
+# Example: Image Prompt Translation
+
+## Source: tag-heavy prompt
+cinematic portrait, nurse, 85mm, Rembrandt, bokeh, realistic, hospital, dramatic, no blur, no extra people
+
+## Natural-language model translation
+A single tired night-shift nurse stands alone in a quiet hospital corridor, medium close-up, soft Rembrandt-style light from camera-left creating a small triangle of light on her cheek, 85mm portrait compression, shallow background blur, realistic skin texture, cool fluorescent corridor ambience, dramatic but grounded realism. Sharp focus on her eyes.
+
+## Notes
+Negative tags became positive conditions: single subject, sharp focus.

package/resources/skills/director-skills/upstream/skills/model-adaptation/examples/same_scene_across_models.md

@@ -0,0 +1,9 @@
+# Example: Same Scene Across Models
+
+Scene: a woman in a yellow raincoat looks through a rainy cafe window, holding a steaming cup.
+
+Universal intent: quiet longing, warm interior/cold exterior contrast.
+
+Image model prompt: medium close-up still, 50mm lens, warm cafe practical light against blue rainy window, natural skin texture, condensation and steam.
+
+Video model prompt: 6s, she exhales and lowers her gaze as steam curls from the cup; camera performs a very slow dolly-in; rain streaks continue on glass; ambience of muffled rain and cafe room tone.

package/resources/skills/director-skills/upstream/skills/model-adaptation/examples/universal_prompt_to_model_specific_exports.md

@@ -0,0 +1,14 @@
+# Example: Universal Prompt to Model-Specific Exports
+
+## Universal brief
+Night guard turns a brass key in a basement storage door. Mood: dread and grief. Macro close-up, cool fluorescent light, realistic texture, 8-second I2V.
+
+## Grok/Aurora export
+Prompt: Preserve the uploaded macro keyframe. The worn brass key stamped 12B trembles once between the guard's fingers, then turns slowly in the lock. Locked-off macro shot, cool green fluorescent practical light, realistic metal scratches and skin texture, quiet suspense. Audio: soft sleeve movement, metallic click, fluorescent hum. Stable hand shape and unchanged key design throughout.
+Settings: I2V, 8s, 16:9 if supported.
+Unknowns: current negative field support.
+
+## Kling export
+Prompt: Use source image as start frame. Macro locked-off shot of the brass key turning slowly in the lock; preserve hand, key, door paint, and lighting. Cool fluorescent basement mood, realistic metal texture. Audio added in post if current tier lacks native audio.
+Negative prompt: extra fingers, changing key shape, background morphing, blurry lock.
+Settings: I2V, 5-10s depending tier, 16:9.

package/resources/skills/director-skills/upstream/skills/model-adaptation/examples/video_prompt_translation.md

@@ -0,0 +1,9 @@
+# Example: Video Prompt Translation
+
+## Universal brief
+A courier runs through rain and stops at a red door, 8 seconds, thriller tone.
+
+## Seedance-style export
+FORMAT: 8s / 2 SHOTS / courier reaches the red door in rain
+SHOT 1 (0:00-0:04): Courier runs down a wet alley, coat whipping from speed, camera tracks beside him at waist height, rain splashes under each step.
+SHOT 2 (0:04-0:08): He stops hard at a red metal door, hand landing flat against it, water running down the paint. Camera pushes in slightly. Audio: rain, breath, hand slap on metal. Logic: preserve same courier, same coat, same red door; no extra people.

package/resources/skills/director-skills/upstream/skills/model-adaptation/references/failure_modes_by_model.md

@@ -0,0 +1,13 @@
+# Failure Modes by Model
+
+| Model | Watch for | First response |
+|---|---|---|
+| Grok/Aurora | hands, text, identity drift, prompt rewrite surprises | front-load, use references, positive constraints |
+| Gemini/Nano Banana | overlong briefs, reference conflicts, unverified surface claims | structure prompt, simplify references, verify docs |
+| GPT-image-2 | text layout complexity, policy/content edge, consistency without references | simplify layout, use editing/reference workflow |
+| Le Chat/FLUX | aspect ratio inconsistency, negative prompts ignored, SD syntax misread | use natural language and positive restatement |
+| Veo 3.1 | too many camera moves, long dialogue, unwanted music | one camera instruction, shorter dialogue, concise audio constraints |
+| Gemini Omni | undocumented capability assumptions | use official-profile fields only; mark unknowns |
+| Kling | hand interaction, background morph, audio tier mismatch | reframe wider, add negative prompt, check tier |
+| Seedance | too many shots, unassigned assets, mirror/reflection failure | respect duration math, assign every asset a job, avoid mirrors |
+| Unknown video model | unsupported params and motion overload | universal brief, one action/camera, unknown log |