@lemoncat7/dsh-partner 2.4.2 → 2.7.0

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

@@ -0,0 +1,226 @@
+# Image Model Profiles
+
+## Grok Imagine / Aurora
+- Model name: Grok Imagine / Aurora
+- Modality: image generation, image editing, image reference source for video
+- Best use cases: photoreal portraits, cinematic stills, editorial/product imagery, social variants, fast iteration
+- Weak use cases: dense text, exact logos, automatic cross-session identity without references, close-up hands
+- Prompt structure recommendations: natural prose; front-load subject/action/environment; follow with lighting, camera/lens, style, aspect/output
+- Ideal prompt length: 40-120 words for many practical prompts; longer allowed by API guides
+- Style control behavior: strong with photographic and cinematic vocabulary
+- Realism behavior: strong when physical light, contact shadows, skin/fabric texture, and optical imperfections are named
+- Camera/lens understanding: strong for lens, shot scale, lighting, and composition terms
+- Motion understanding: not applicable for stills; use video profile for motion
+- Character consistency support: requires reference image workflow for reliable series; text-only character descriptions drift
+- Reference image support: supported via documented workflows/API in local guides
+- Negative prompt support: no separate consumer negative field in guides; use positive restatement unless current API says otherwise
+- Parameter support: aspect ratio and output fields documented in Grok API guide; verify current API
+- Aspect ratio support: many aspect ratios documented in guides
+- Duration support: n/a for image
+- Resolution support: resolution options documented in API guide; verify current values
+- Seed/reference controls: seed support unknown in distilled local guides; reference controls documented
+- Common artifacts: hands, text artifacts, identity drift, prompt auto-rewrite surprises
+- Recommended prompt template: [subject + action], [specific environment], [lighting], [camera/lens/framing], [style/texture], [aspect/output].
+- Example prompt: A tired night-shift nurse pauses under a buzzing hospital corridor light, one hand on a paper coffee cup, medium close-up, 50mm lens, cool fluorescent practical light from overhead, natural skin texture, soft film grain, quiet dramatic realism, 16:9.
+- Troubleshooting notes: Front-load identity, simplify hands/text, use references for continuity, inspect revised prompt when available.
+- Source confidence: high
+
+## Gemini image / Nano Banana
+- Model name: Gemini image / Nano Banana
+- Modality: image generation, editing, multi-image composition
+- Best use cases: natural-language scene construction, complex edits, text rendering, reference-heavy composition
+- Weak use cases: unverified surface-specific settings, overlong instructions, conflicting references
+- Prompt structure recommendations: declarative natural language; subject, action/kinetics, environment/context, lighting/atmosphere, camera/hardware, style/constraints
+- Ideal prompt length: can accept long structured briefs per guides; still prefer concise modular structure
+- Style control behavior: strong when style is a coherent brief rather than tag soup
+- Realism behavior: strong when lighting and physical context are explicit
+- Camera/lens understanding: good in source guides; use camera/composition language normally
+- Motion understanding: n/a for stills
+- Character consistency support: reference-heavy workflows described; exact limits should be verified
+- Reference image support: source guides favor semantic positives; exact negative support varies by surface
+- Negative prompt support: aspect/settings may be prompt, UI, or API dependent; verify current surface
+- Parameter support: supported behavior varies by surface; unknown without current docs
+- Aspect ratio support: n/a
+- Duration support: unknown without current docs
+- Resolution support: seed support unknown; reference image support described but limits may change
+- Seed/reference controls: reference conflicts, over-instruction, unsupported parameter assumptions
+- Common artifacts: Create [asset type]. Subject: ... Action/context: ... Environment: ... Lighting: ... Camera/composition: ... Style/constraints: ...
+- Recommended prompt template: Create a character reference sheet for a courier in a yellow raincoat, wet black hair, tired brown eyes, compact athletic build, plain studio background, front/three-quarter/profile/full-body views, even soft light, consistent facial features, natural fabric texture.
+- Example prompt: Reduce reference count when conflicts appear; convert tag lists to prose; verify exact current limits.
+- Troubleshooting notes: medium
+- Source confidence: medium
+
+## GPT-image-2
+- Model name: GPT-image-2
+- Modality: image generation, editing
+- Best use cases: high-fidelity image craft, typography/text rendering, editing, consistency workflows
+- Weak use cases: current API/settings may differ; exact prompt controls need official verification
+- Prompt structure recommendations: clear creative brief with purpose, subject, composition, lighting, style, text/edit instructions, and constraints
+- Ideal prompt length: unknown; use concise complete briefs unless current docs indicate otherwise
+- Style control behavior: strong per source guide, especially when purpose and layout are explicit
+- Realism behavior: strong with craft details, texture, and lighting
+- Camera/lens understanding: good for image composition; exact camera/lens weighting unknown
+- Motion understanding: n/a for stills
+- Character consistency support: editing/reference workflows described; verify current docs
+- Reference image support: unknown in distilled suite; do not guess separate negative support
+- Negative prompt support: parameter surface discussed in source guide; verify official docs
+- Parameter support: unknown without current docs
+- Aspect ratio support: n/a
+- Duration support: unknown without current docs
+- Resolution support: seed/reference controls unknown unless current docs provided
+- Seed/reference controls: over-complex text layout, policy/content edge cases, inconsistency without references
+- Common artifacts: [purpose]. [subject]. Composition: ... Lighting: ... Style: ... Text if any: ... Constraints: ...
+- Recommended prompt template: Poster-style still of a small-town observatory at night, single readable title at top, clean typographic hierarchy, blue-black sky, warm window light, cinematic but realistic, no extra copy.
+- Example prompt: Simplify text layout; use edit/reference workflow for consistency; verify model parameters before export.
+- Troubleshooting notes: medium
+- Source confidence: medium
+
+## Le Chat / FLUX Ultra
+- Model name: Le Chat / FLUX Ultra
+- Modality: image generation and editing through Le Chat
+- Best use cases: natural-language image prompting, typography, product/design visuals, prompt-enhanced imagery
+- Weak use cases: direct low-level FLUX controls through Le Chat, seed/CFG/steps, guaranteed aspect ratio on every surface
+- Prompt structure recommendations: natural language; no Stable Diffusion weighting syntax or comma-tag stacks
+- Ideal prompt length: moderate natural-language brief; avoid massive overloaded prompts
+- Style control behavior: strong with coherent natural-language style requests
+- Realism behavior: strong when material, light, and composition are explicit
+- Camera/lens understanding: good for image composition/lens terms; exact weighting unknown
+- Motion understanding: n/a for stills
+- Character consistency support: upload/edit workflow; guide notes uploaded aspect ratio preservation in edits
+- Reference image support: no separate negative field in guide; describe desired positive state
+- Negative prompt support: no seed/CFG/step controls through Le Chat per guide
+- Parameter support: prompted aspect ratio may be inconsistent; upload/edit workaround noted
+- Aspect ratio support: n/a
+- Duration support: unknown or surface-dependent
+- Resolution support: no seed; references via upload/edit workflow
+- Seed/reference controls: aspect inconsistency, literal interpretation of SD syntax, policy refusals
+- Common artifacts: A [subject] in [context], [composition], [lighting], [style/material detail], [desired aspect or edit preservation].
+- Recommended prompt template: Product photograph of a matte black pocket recorder on a clean grey surface, softbox reflection along one edge, crisp readable buttons, minimal studio composition, realistic plastic and metal texture, square crop.
+- Example prompt: Use natural language; rewrite negatives as positive descriptions; use upload/edit for aspect preservation when needed.
+- Troubleshooting notes: high
+- Source confidence: high
+
+## Midjourney
+- Model name: Midjourney
+- Modality: image generation
+- Best use cases: stylized cinematic stills and concept images when user targets Midjourney
+- Weak use cases: local sources mention mainly for comparison; exact current parameters unknown
+- Prompt structure recommendations: use universal image brief, then adapt with current Midjourney docs if supplied
+- Ideal prompt length: unknown
+- Style control behavior: unknown from local guides
+- Realism behavior: unknown from local guides
+- Camera/lens understanding: unknown from local guides
+- Motion understanding: n/a
+- Character consistency support: unknown
+- Reference image support: unknown
+- Negative prompt support: unknown
+- Parameter support: unknown
+- Aspect ratio support: n/a
+- Duration support: unknown
+- Resolution support: unknown
+- Seed/reference controls: unknown
+- Common artifacts: model-specific syntax mismatch if guessed
+- Recommended prompt template: Use universal brief and mark parameters unknown.
+- Example prompt: Unknown-profile export: preserve subject, light, camera, style, and aspect request in plain language.
+- Troubleshooting notes: Ask for current docs/settings or browse official/current sources before asserting flags.
+- Source confidence: low
+
+## Stable Diffusion
+- Model name: Stable Diffusion
+- Modality: image generation, editing, local workflows
+- Best use cases: parameterized local image generation and controllable pipelines
+- Weak use cases: not deeply profiled; capabilities depend on checkpoint, UI, ControlNet/IP-Adapter/etc.
+- Prompt structure recommendations: requires workflow-specific adaptation; do not use Le Chat/Grok natural-language assumptions blindly
+- Ideal prompt length: unknown
+- Style control behavior: workflow-dependent
+- Realism behavior: workflow-dependent
+- Camera/lens understanding: workflow-dependent
+- Motion understanding: n/a
+- Character consistency support: workflow-dependent
+- Reference image support: often supported in ecosystem, but exact workflow unknown here
+- Negative prompt support: workflow-dependent
+- Parameter support: workflow-dependent
+- Aspect ratio support: n/a
+- Duration support: workflow-dependent
+- Resolution support: seed support likely in many workflows but mark unknown unless target UI is known
+- Seed/reference controls: tag/prose mismatch, missing control modules, checkpoint-specific failures
+- Common artifacts: Create universal brief, then translate only after target SD workflow is known.
+- Recommended prompt template: A universal cinematic still brief with separate fields for subject, style, camera, lighting, negative risks, and references.
+- Example prompt: Ask which SD UI/model/checkpoint/control stack is used.
+- Troubleshooting notes: low
+- Source confidence: low
+
+## Ideogram
+- Model name: Ideogram
+- Modality: image generation, text/design images
+- Best use cases: text-heavy posters and design assets, based on comparison mentions
+- Weak use cases: not deeply profiled in local guides
+- Prompt structure recommendations: unknown; use universal image brief with text layout separated
+- Ideal prompt length: unknown
+- Style control behavior: unknown
+- Realism behavior: unknown
+- Camera/lens understanding: unknown
+- Motion understanding: n/a
+- Character consistency support: unknown
+- Reference image support: unknown
+- Negative prompt support: unknown
+- Parameter support: unknown
+- Aspect ratio support: n/a
+- Duration support: unknown
+- Resolution support: unknown
+- Seed/reference controls: unknown
+- Common artifacts: guessing text capabilities beyond source
+- Recommended prompt template: Use universal brief and mark exact parameters unknown.
+- Example prompt: Poster concept with exact desired text separated from visual description.
+- Troubleshooting notes: Verify current docs for text, style, and aspect controls.
+- Source confidence: low
+
+## DALL-E
+- Model name: DALL-E
+- Modality: image generation
+- Best use cases: general image generation; mentioned for comparison
+- Weak use cases: not profiled in local guides
+- Prompt structure recommendations: unknown; use universal image brief
+- Ideal prompt length: unknown
+- Style control behavior: unknown
+- Realism behavior: unknown
+- Camera/lens understanding: unknown
+- Motion understanding: n/a
+- Character consistency support: unknown
+- Reference image support: unknown
+- Negative prompt support: unknown
+- Parameter support: unknown
+- Aspect ratio support: n/a
+- Duration support: unknown
+- Resolution support: unknown
+- Seed/reference controls: unknown
+- Common artifacts: assuming current OpenAI image capabilities from memory
+- Recommended prompt template: Use universal prompt and route to current official docs if exact settings matter.
+- Example prompt: A concise natural-language cinematic still brief.
+- Troubleshooting notes: Verify current official documentation before model-specific claims.
+- Source confidence: low
+
+## FLUX direct
+- Model name: FLUX direct
+- Modality: image generation and editing depending implementation
+- Best use cases: photoreal/design imagery in FLUX-based systems
+- Weak use cases: direct FLUX outside Le Chat not profiled; implementation dependent
+- Prompt structure recommendations: workflow-specific; do not assume Le Chat limitations apply to all FLUX stacks
+- Ideal prompt length: unknown
+- Style control behavior: unknown
+- Realism behavior: unknown
+- Camera/lens understanding: unknown
+- Motion understanding: n/a
+- Character consistency support: unknown
+- Reference image support: unknown
+- Negative prompt support: implementation dependent
+- Parameter support: implementation dependent
+- Aspect ratio support: n/a
+- Duration support: implementation dependent
+- Resolution support: unknown
+- Seed/reference controls: workflow-specific artifacts
+- Common artifacts: Use universal brief; ask for host/UI/workflow.
+- Recommended prompt template: Natural-language image brief plus optional fields separated for the target UI.
+- Example prompt: Confirm whether this is Le Chat, local FLUX, API, or another host.
+- Troubleshooting notes: low
+- Source confidence: low

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

@@ -0,0 +1,24 @@
+# Model Capability Matrix
+
+Source confidence reflects the local guides only. Unknown means the suite should not guess.
+
+| Model/tool | Modality | Best use | Negative support | Reference support | Text/logo reliability | Duration | Aspect/seed/resolution | Source confidence |
+|---|---|---|---|---|---|---|---|---|
+| Grok Imagine / Aurora | image, edit, T2V, I2V, extension | photoreal image/video, fast variants, permissive creative exploration | no separate consumer negative field in guides; use positive restatement | image references via supported workflows/API per guides | moderate, not best for dense text | 1-15s video; edits capped in guides | aspect ratios/API fields documented in Grok guides | high |
+| Gemini image / Nano Banana | image, edit, multi-image composition | natural-language image generation, editing, text rendering, reference-heavy composition | semantic/positive negatives favored; model-specific support varies | up to many references in source guides | strong in source guides | n/a for image | prompt/in-platform controls; some claims need verification | medium |
+| GPT-image-2 | image, edit | high-fidelity image craft, text rendering, editing, consistency workflows | model-specific; verify current tool | references/editing per guide | strong in source guide | n/a | parameter surface in source guide, verify current docs | medium |
+| Le Chat / FLUX Ultra | image, edit | natural-language image prompting, typography, product/design visuals | no negative field per guide; use positive restatement | upload/edit workflow; aspect ratio preserved on edit per guide | strong but not perfect | n/a | no seed/CFG/steps through Le Chat per guide | high |
+| Gemini Veo 3.1 | T2V, I2V, references, extension, audio | cinematic video with audio and reference ingredients | supports concise constraints per guide | up to 4 reference ingredients in guide | limited/risky in video | 4/6/8s clips, extendable per guide | model settings in guide; verify current docs | medium |
+| Gemini Omni video | video, conversational editing | cinematic prompt-to-video, audio, iterative edits | supported per guide | reference-driven workflows | unknown to medium | guide-dependent | parameter table in source guide | medium |
+| Kling AI | T2V, I2V, elements, start/end, extend, lip sync | controllable video, references, keyframe interpolation | negative prompts supported in guide | Elements/multi-reference workflows | text/logos risky | model/tier-specific; see profile | tier-dependent | high |
+| Seedance 2.0 | T2V, I2V, quad-modal refs | multimodal video, asset-role prompting, shot-by-shot prompts | constraints supported as prompt text | images/video/audio roles per guide | unknown | 4-15s in guide | input-driven/aspect support per guide | high |
+| Midjourney | image | stylized and cinematic stills | model-specific; unknown in local suite | reference features unknown in local suite | weak/unknown | n/a | unknown | low |
+| Stable Diffusion | image, edit, local workflows | parameterized local image generation | supported in common ecosystem but not profiled here | model/workflow-dependent | weak/unknown | n/a | extensive but unknown here | low |
+| Ideogram | image/text design | text-heavy images, posters | unknown | unknown | strong relative mention only | n/a | unknown | low |
+| Runway | video | video generation/editing | unknown | unknown | unknown | unknown | unknown | low |
+| Pika | video | video generation | unknown | unknown | unknown | unknown | unknown | low |
+| Luma | video | video generation | unknown | unknown | unknown | unknown | unknown | low |
+| Sora | video | cinematic video | unknown | unknown | unknown | unknown | unknown | low |
+| Hailuo | video | video generation | unknown | unknown | unknown | unknown | unknown | unknown |
+| DALL-E | image | general image generation | unknown in local suite | unknown | model-dependent | n/a | unknown | low |
+| FLUX | image | photoreal/design image generation | ecosystem-dependent | ecosystem-dependent | strong relative mention | n/a | unknown outside Le Chat | low |

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

@@ -0,0 +1,10 @@
+# Model Update Protocol
+
+Use this when adding or revising model profiles.
+
+1. Record date, source, and surface: web app, API, partner platform, or local model.
+2. Separate official docs from community reports.
+3. Update modality, parameters, reference support, negative support, duration, aspect ratio, resolution, seed/control, text/logo behavior, audio behavior, and failure modes.
+4. Mark confidence: high for official/current docs or repeated project evidence, medium for strong but indirect guides, low for comparison mentions, unknown for missing evidence.
+5. Add or update one example export and one test case.
+6. Do not delete older caveats unless the new source directly supersedes them.

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

@@ -0,0 +1,20 @@
+# Parameter Compatibility
+
+Track these fields separately from prompt text whenever possible:
+- aspect_ratio
+- duration
+- resolution/quality tier
+- seed
+- reference image/video/audio IDs
+- negative prompt
+- style preset
+- camera preset/control
+- start frame/end frame
+- extension mode
+- lip-sync/audio source
+
+## Compatibility policy
+- Supported: include exact field and value.
+- Unsupported: rewrite as prompt text if safe, or remove.
+- Unknown: mark unknown; do not invent.
+- Inherited: for editing/I2V, note when aspect/duration follows source media.

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

@@ -0,0 +1,18 @@
+# Prompt Translation Rules
+
+## Translation workflow
+1. Parse universal brief into intent, execution, and controls.
+2. Remove source-model syntax that the target model does not support.
+3. Preserve subject, action, emotion, continuity anchors, camera, light, and style.
+4. Convert settings into parameters only if the target supports them.
+5. Convert negatives according to target behavior.
+6. Add a translation report with unsupported and unknown fields.
+
+## Syntax conversion
+- Tag soup to natural-language model: rewrite as coherent sentences.
+- Natural prose to slot/template model: split into subject, action, camera, style, constraints.
+- Image prompt to video prompt: keep identity and setting, add motion, camera behavior, duration, audio, and continuity guardrails.
+- T2V to I2V: remove excessive appearance description and focus on animation instructions.
+
+## Unknown rule
+If a capability is not in the profile or current user-supplied docs, write `unknown` and provide a conservative fallback.

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

@@ -0,0 +1,251 @@
+# Video Model Profiles
+
+## Grok Imagine / Aurora video
+- Model name: Grok Imagine / Aurora video
+- Modality: text-to-video, image-to-video, reference images, editing, extension
+- Best use cases: short cinematic clips, I2V from locked stills, fast variants, audio-aware prompts in source guides
+- Weak use cases: full scenes in one prompt, too many camera moves, long dialogue, long-form continuity without chaining
+- Prompt structure recommendations: subject doing one thing, one camera move, style, lighting, audio cue, settings separated when supported
+- Ideal prompt length: concise, front-loaded, one-shot prompts; API max length exists but prompt economy is favored
+- Style control behavior: strong with cinematic and photographic vocabulary
+- Realism behavior: stronger with physical consequences, grounded contact, texture, light behavior
+- Camera/lens understanding: strong; source guides emphasize one camera move and lens/framing terms
+- Motion understanding: good when action is simple, continuous, and consequence-based
+- Character consistency support: reference images and image-first pipeline recommended
+- Reference image support: supported in documented workflows/API per source guides
+- Negative prompt support: use positive restatement unless current API/tool exposes a separate field
+- Parameter support: duration/aspect/reference fields documented in Grok API guide; verify current API
+- Aspect ratio support: multiple aspect ratios documented
+- Duration support: 1-15s T2V in guides; edits shorter/capped in guides
+- Resolution support: resolution options documented; verify current values
+- Seed/reference controls: seed unknown; reference image controls documented
+- Common artifacts: morphing, identity drift, hands, temporal blending, lip/audio mismatch
+- Recommended prompt template: [Subject doing one action], [one camera move], [style], [lighting], [audio cue], [duration/aspect settings].
+- Example prompt: A courier in a yellow raincoat runs through a wet alley for 8 seconds, low side tracking shot, cold neon reflections, rain splashing from each step, tense thriller realism. Audio: rain, breath, footsteps. Stable face and unchanged coat.
+- Troubleshooting notes: Split complex scenes, use references for identity, keep one camera movement, shorten dialogue.
+- Source confidence: high
+
+## Gemini Veo 3.1
+- Model name: Gemini Veo 3.1
+- Modality: text-to-video, image-to-video, reference ingredients, scene extension, audio
+- Best use cases: cinematic clips with audio, reference-based video, first/last-frame style workflows, extension chains
+- Weak use cases: long dialogue, many negative constraints, too many camera instructions
+- Prompt structure recommendations: five/six-part cinematic formula: subject, action, context, style, camera, audio/constraints
+- Ideal prompt length: medium concise prompt; enough detail for subject/action/camera/audio without overload
+- Style control behavior: strong with genre and cinematic vocabulary per guide
+- Realism behavior: strong with lighting, texture, and film stock cues
+- Camera/lens understanding: good; place one camera instruction early
+- Motion understanding: good with clear action and extension workflows
+- Character consistency support: references/ingredients and extension support in source guide
+- Reference image support: source guide states up to 4 reference ingredients; verify current docs
+- Negative prompt support: concise constraints supported; guide recommends limiting negatives
+- Parameter support: duration/reference/audio/settings depend on current surface
+- Aspect ratio support: supported; exact options need verification
+- Duration support: 4/6/8s clips in guide, extendable
+- Resolution support: 4K upscale mentioned in guide; verify availability
+- Seed/reference controls: seed unknown; reference ingredients supported per guide
+- Common artifacts: ignored camera moves, lip sync issues, unwanted music, prompt collapse from over-constraint
+- Recommended prompt template: [Subject] [action] in [context], [style], [camera], [lighting], [audio], [top constraints].
+- Example prompt: A woman in a yellow raincoat holds a steaming cup at a rainy cafe window, slow dolly-in, warm interior practicals against blue rain outside, natural lip movement as she whispers "not yet," soft cafe ambience and rain, no extra music.
+- Troubleshooting notes: Use one camera instruction, shorten dialogue, cap constraints, use references for identity.
+- Source confidence: medium
+
+## Gemini Omni video
+- Model name: Gemini Omni video
+- Modality: video generation, audio-aware prompting, conversational editing per source guide
+- Best use cases: cinematic prompt-to-video, audio-aware scenes, iterative conversational edits, reference-driven prompting
+- Weak use cases: undocumented capabilities and surface-specific settings
+- Prompt structure recommendations: official-style elements plus cinematic vocabulary: subject, context, action, style, camera, audio
+- Ideal prompt length: unknown exact limit; use concise complete prompt
+- Style control behavior: strong per guide but verify surface
+- Realism behavior: strong with physical/cinematic detail
+- Camera/lens understanding: good per guide
+- Motion understanding: good when action is specific
+- Character consistency support: reference-driven workflows described
+- Reference image support: source guide describes reference syntax/workflows; verify current docs
+- Negative prompt support: source guide states support; use sparingly and verify surface
+- Parameter support: parameter table in guide; verify current docs
+- Aspect ratio support: documented in guide; verify current options
+- Duration support: guide-dependent; verify current docs
+- Resolution support: guide-dependent; verify current docs
+- Seed/reference controls: seed/reference controls guide-dependent
+- Common artifacts: anatomy errors, physics violations, identity drift, audio-video desync, prompt collapse
+- Recommended prompt template: [Subject], [action], [setting], [camera/lens], [lighting/color], [audio], [constraints/settings].
+- Example prompt: A locked-off medium shot of an old radio on a kitchen table as the dial turns by itself, tungsten practical light, dusty air, subtle room tone and faint static, slow dread, no visible hands.
+- Troubleshooting notes: Mark undocumented features unknown; use conversational edits only for localized changes.
+- Source confidence: medium
+
+## Kling AI
+- Model name: Kling AI
+- Modality: T2V, I2V, Elements/multi-reference, motion brush, camera controls, start/end frames, extend, lip sync, video-to-video on some surfaces
+- Best use cases: controlled short video, keyframe interpolation, multi-reference continuity, product/commercial shots
+- Weak use cases: complex close-up hand interaction, exact text/logos, unsupported native audio on older tiers, overloaded multi-shot prompts
+- Prompt structure recommendations: hybrid prompt: subject, action, camera, lighting/style, negative prompt where supported
+- Ideal prompt length: tier/model dependent; concise shot prompts favored
+- Style control behavior: strong with clear style/camera/lens instructions
+- Realism behavior: good with lighting/texture and negative artifact control
+- Camera/lens understanding: strong, with some preset camera controls by tier
+- Motion understanding: good for simple direct motion; complex choreography risky
+- Character consistency support: Elements/multi-reference workflows support continuity
+- Reference image support: Elements/multi-reference and I2V supported in guides
+- Negative prompt support: negative prompts supported in many guide workflows
+- Parameter support: tier-specific controls; verify current model/tier
+- Aspect ratio support: tier-specific; guide includes aspect options
+- Duration support: tier-specific clip lengths; verify current model/tier
+- Resolution support: tier-specific quality/resolution
+- Seed/reference controls: seed unknown; reference/Elements controls documented
+- Common artifacts: identity drift, hand errors, background morphing, camera preset mismatch, audio limitations
+- Recommended prompt template: T2V: [subject/action]. Camera: [move]. Lighting/style: [look]. Negative prompt: [top artifacts].
+- Example prompt: A red ceramic mug slides across a wet cafe table as a hand stops it, macro static shot, warm window light, realistic water streaks. Negative: extra fingers, changing mug color, background morphing, blur.
+- Troubleshooting notes: Check tier/features, use negative prompt for top artifacts, reframe risky hands wider.
+- Source confidence: high
+
+## Seedance 2.0
+- Model name: Seedance 2.0
+- Modality: text-to-video, image-to-video, image/video/audio reference inputs in quad-modal workflow per guides
+- Best use cases: asset-role prompting, shot-by-shot cinematic clips, audio/reference-driven scenes
+- Weak use cases: too many shots for duration, unassigned references, mirrors/reflections, content/IP-sensitive prompts
+- Prompt structure recommendations: format line, asset legend, shot blocks, audio layers, logic rules, constraints
+- Ideal prompt length: shot-count and prompt length must respect 4-15s duration; concise shot blocks
+- Style control behavior: strong when style anchor is singular and assets have jobs
+- Realism behavior: good with physical logic rules and texture cues
+- Camera/lens understanding: good with explicit framing/lens/movement per shot
+- Motion understanding: good when each shot has one motion verb
+- Character consistency support: strong through assigned reference assets
+- Reference image support: guides cite up to 9 images, 3 videos, 3 audio files; verify current platform
+- Negative prompt support: constraints/stability boilerplate in prompt; exact negative field unknown
+- Parameter support: platform-specific; verify current surface
+- Aspect ratio support: input/setting-dependent; crop references to target aspect per guide
+- Duration support: 4-15s in guides
+- Resolution support: unknown/current platform dependent
+- Seed/reference controls: seed unknown; reference tags/roles central
+- Common artifacts: subject stops, identity drift, mirror/reflection breaks, filter rejection, overloaded dialogue
+- Recommended prompt template: FORMAT: [duration]s / [shot count] SHOTS / [concept]; ASSETS: @image_1 role; SHOT blocks; AUDIO; LOGIC.
+- Example prompt: FORMAT: 8s / 2 SHOTS / courier reaches red door. ASSETS: @image_1 locks courier wardrobe. SHOT 1: courier runs through rain, low side tracking. SHOT 2: he stops at red door, hand hits metal. Audio: rain, breath, hand slap. Logic: same courier and coat throughout.
+- Troubleshooting notes: Assign every asset a job, count actions and words, avoid mirrors, simplify shot math.
+- Source confidence: high
+
+## Runway
+- Model name: Runway
+- Modality: video generation/editing
+- Best use cases: not deeply profiled; use when user targets Runway and supplies docs/settings
+- Weak use cases: unknown local capability surface
+- Prompt structure recommendations: universal video brief; model-specific syntax unknown
+- Ideal prompt length: unknown
+- Style control behavior: unknown
+- Realism behavior: unknown
+- Camera/lens understanding: unknown
+- Motion understanding: unknown
+- Character consistency support: unknown
+- Reference image support: unknown
+- Negative prompt support: unknown
+- Parameter support: unknown
+- Aspect ratio support: unknown
+- Duration support: unknown
+- Resolution support: unknown
+- Seed/reference controls: unknown
+- Common artifacts: unsupported assumptions, motion overload
+- Recommended prompt template: Universal video brief with subject/action/camera/light/audio/constraints separated.
+- Example prompt: 8s thriller shot brief with one action and one camera movement.
+- Troubleshooting notes: Verify current Runway docs before exact settings.
+- Source confidence: low
+
+## Pika
+- Model name: Pika
+- Modality: video generation
+- Best use cases: not deeply profiled; use conservative video briefs
+- Weak use cases: unknown
+- Prompt structure recommendations: universal video brief
+- Ideal prompt length: unknown
+- Style control behavior: unknown
+- Realism behavior: unknown
+- Camera/lens understanding: unknown
+- Motion understanding: unknown
+- Character consistency support: unknown
+- Reference image support: unknown
+- Negative prompt support: unknown
+- Parameter support: unknown
+- Aspect ratio support: unknown
+- Duration support: unknown
+- Resolution support: unknown
+- Seed/reference controls: unknown
+- Common artifacts: unsupported assumptions
+- Recommended prompt template: Universal brief only.
+- Example prompt: One-shot video description with model fields marked unknown.
+- Troubleshooting notes: Verify current Pika docs.
+- Source confidence: low
+
+## Luma
+- Model name: Luma
+- Modality: video generation
+- Best use cases: not deeply profiled; use conservative video briefs
+- Weak use cases: unknown
+- Prompt structure recommendations: universal video brief
+- Ideal prompt length: unknown
+- Style control behavior: unknown
+- Realism behavior: unknown
+- Camera/lens understanding: unknown
+- Motion understanding: unknown
+- Character consistency support: unknown
+- Reference image support: unknown
+- Negative prompt support: unknown
+- Parameter support: unknown
+- Aspect ratio support: unknown
+- Duration support: unknown
+- Resolution support: unknown
+- Seed/reference controls: unknown
+- Common artifacts: unsupported assumptions
+- Recommended prompt template: Universal brief only.
+- Example prompt: One-shot video description with model fields marked unknown.
+- Troubleshooting notes: Verify current Luma docs.
+- Source confidence: low
+
+## Sora
+- Model name: Sora
+- Modality: video generation
+- Best use cases: cinematic video target mentioned in sources/user request
+- Weak use cases: not profiled deeply in local guides
+- Prompt structure recommendations: universal cinematic video brief until current docs are provided
+- Ideal prompt length: unknown
+- Style control behavior: unknown
+- Realism behavior: unknown
+- Camera/lens understanding: unknown
+- Motion understanding: unknown
+- Character consistency support: unknown
+- Reference image support: unknown
+- Negative prompt support: unknown
+- Parameter support: unknown
+- Aspect ratio support: unknown
+- Duration support: unknown
+- Resolution support: unknown
+- Seed/reference controls: unknown
+- Common artifacts: unsupported assumptions
+- Recommended prompt template: Universal video brief plus unknown log.
+- Example prompt: A concise scene prompt with one action, camera, light, and continuity anchors.
+- Troubleshooting notes: Verify current official docs before exact controls.
+- Source confidence: low
+
+## Hailuo
+- Model name: Hailuo
+- Modality: video generation
+- Best use cases: activation placeholder from user request
+- Weak use cases: no local guide evidence
+- Prompt structure recommendations: universal cinematic video brief until current docs are provided
+- Ideal prompt length: unknown
+- Style control behavior: unknown
+- Realism behavior: unknown
+- Camera/lens understanding: unknown
+- Motion understanding: unknown
+- Character consistency support: unknown
+- Reference image support: unknown
+- Negative prompt support: unknown
+- Parameter support: unknown
+- Aspect ratio support: unknown
+- Duration support: unknown
+- Resolution support: unknown
+- Seed/reference controls: unknown
+- Common artifacts: unsupported assumptions
+- Recommended prompt template: Universal video brief plus unknown log.
+- Example prompt: A concise one-shot video prompt with unknown settings.
+- Troubleshooting notes: Ask for docs or verify current specs.
+- Source confidence: unknown

package/resources/skills/director-skills/upstream/skills/model-adaptation/templates/model_comparison_table.md

@@ -0,0 +1,4 @@
+# Model Comparison Table Template
+
+| Model | Modality | Strength for this scene | Risk | Required prompt changes | Unknowns |
+|---|---|---|---|---|---|

package/resources/skills/director-skills/upstream/skills/model-adaptation/templates/model_specific_prompt_export.md

@@ -0,0 +1,21 @@
+# Model-Specific Prompt Export Template
+
+## Target model
+
+## Prompt
+
+## Settings
+- Modality:
+- Aspect ratio:
+- Duration:
+- Resolution/quality:
+- Seed/reference controls:
+- Negative prompt/constraints:
+
+## Reference asset mapping
+
+## Unsupported or unknown capabilities
+
+## Expected failure modes
+
+## Troubleshooting note

package/resources/skills/director-skills/upstream/skills/model-adaptation/templates/multi_model_export_package.md

@@ -0,0 +1,18 @@
+# Multi-Model Export Package Template
+
+## Universal brief
+
+## Exports
+### Model A
+- Prompt:
+- Settings:
+- Notes:
+
+### Model B
+- Prompt:
+- Settings:
+- Notes:
+
+## Comparison
+
+## Unknowns and verification tasks

package/resources/skills/director-skills/upstream/skills/model-adaptation/templates/prompt_translation_report.md

@@ -0,0 +1,13 @@
+# Prompt Translation Report Template
+
+- Source prompt/model:
+- Target model:
+- Preserved creative intent:
+- Syntax changes:
+- Parameter changes:
+- Negative/constraint handling:
+- Reference handling:
+- Unsupported fields:
+- Unknown fields:
+- Likely failure modes:
+- Revised prompt:

package/resources/skills/director-skills/upstream/skills/model-adaptation/templates/universal_prompt_brief.md

@@ -0,0 +1,25 @@
+# Universal Prompt Brief Template
+
+## Creative intent
+- Story/scene purpose:
+- Subject/character:
+- Emotion/audience experience:
+- Genre/tone:
+
+## Cinematic execution
+- Action:
+- Location/environment:
+- Framing/lens/camera:
+- Lighting/color:
+- Texture/realism:
+- Audio/motion if video:
+- Continuity anchors:
+
+## Desired controls
+- Modality:
+- Aspect ratio:
+- Duration:
+- Resolution/quality:
+- References:
+- Negative/constraints:
+- Seed/settings: