@aswless_854771076/ai_short_studio_cli 0.1.98 → 0.1.100

package/skills/using-vvicat-ai-short-studio-cli/references/hypit/playbooks/craft/examples/image-direction.md

@@ -0,0 +1,123 @@
+# Image direction examples
+
+Read these with [Directing generated images](../image-direction.md). Each complete prompt uses
+Capture, Person, Shot and Setting. The Kit supplies Capture and assembles `person`, `shot` and
+`setting` in that order; the [Craft](../image-direction.md#use-a-concrete-capture-direction) explains
+assembly. For these full-screen portraits, set `aspect-ratio="9:16"` and usually `resolution="2K"`
+on `gpt:Image`.
+
+These production prompts belong to the accepted balcony portrait and five reviewed character
+portraits. They retain the same capture language while choosing the person, framing and setting
+for each picture: soft goth, clean girl, coquette, Harajuku decora and streetwear. The styling terms
+organize a particular appearance; the selected details make that direction specific.
+
+## Japanese sailor on an apartment balcony
+
+![Sailor-style presenter on a balcony, with navy clothing, terracotta tiles and green plants](https://storage.googleapis.com/hypit-public-assets/assets/skills/hypit/image-direction/2026-09-10/japanese-sailor-balcony.webp)
+
+The user selected the result generated from this exact prompt. The Person paragraph gives a
+specific casting direction, strong beauty, broad shoulders and head-to-shoulder proportions, with
+long straight hair and sailor styling. Shot establishes the frontal speaking state, free-hand
+gestures and face placement. The table gives the offset composition a physical relationship.
+
+In the selected result, navy clothing, terracotta architecture and plants have distinct color roles.
+The glass door, railing and neighboring buildings establish a place with different distances.
+The model supplied individual tiles, leaves, reflections and clothing details within the authored
+direction; those incidental arrangements are not additional requirements for another portrait.
+
+Production prompt:
+
+```text
+A photograph captured as a single frame from a video actually shot on an iPhone, with the texture of real iPhone footage. The image looks real, without an oily, overprocessed finish. The background is clearly visible, with no depth-of-field blur. Skin texture is natural and fine, and the lighting is natural. The image is coherent and free of visual artifacts.
+
+A Japanese woman in her early twenties. She is exceptionally, strikingly beautiful, with the looks of a top Korean girl-group idol and very fair skin. She has beautiful long, straight black hair with blunt bangs and pretty makeup, with bold eye makeup giving her an intensely alluring gaze. She has a beautiful figure, very broad shoulders and excellent head-to-shoulder proportions. She wears a sweet-but-edgy navy-blue sailor-style outfit with white trim and a matching sailor cap.
+
+Generate a vertical medium half-body shot. She sits on a chair, fairly close to the camera, holding a handheld microphone and speaking directly to the viewer, with her free hand gesturing. Her face points straight toward the lens, with no head tilt or rotation, and sits toward the upper-right part of the frame. A small wooden table beside her enters the lower-left corner.
+
+She is on an apartment balcony. Behind her are a sliding glass door, textured exterior tiles, metal railings and green plants, with neighboring buildings visible beyond the balcony. The background color scheme is terracotta and green.
+```
+
+The frontal face and speaking gesture give a reusable direct-to-viewer state. The upper-right face
+and lower-left table are choices for this composition. A centered standing view keeps the casting
+and proportion direction while changing posture, placement and the scene relationship. A podcast
+view instead directs attention toward the partner. See the [idle guidance](../image-direction.md#choose-an-idle-state-for-the-encounter).
+
+## Goth girl outside a Korean cafe
+
+![Goth presenter outside a navy-and-yellow Korean café, holding a microphone beside a dessert table](https://storage.googleapis.com/hypit-public-assets/assets/skills/hypit/image-direction/2026-09-10/goth-korean-cafe.webp)
+
+Soft goth styling gives the person a clear visual character within an ordinary café setting. The grape bingsu and nearby table connect her rightward placement to a visit already underway.
+
+```text
+A photograph captured as a single frame from a video actually shot on an iPhone, with the texture of real iPhone footage. The image looks real, without an oily, overprocessed finish. The background is clearly visible, with no depth-of-field blur. Skin texture is natural and fine, and the lighting is natural. The image is coherent and free of visual artifacts.
+
+A Japanese-American woman in her early twenties. She is exceptionally, strikingly beautiful, with the looks of a top Korean girl-group idol and very fair skin. She has beautiful long, straight black hair with blunt bangs and soft goth makeup, with bold eye makeup giving her an intensely alluring gaze. She has a beautiful figure, very broad shoulders and excellent head-to-shoulder proportions. She wears a sweet-but-edgy black beret, a black goth-style outfit and a few pieces of silver jewelry.
+
+Generate a vertical medium half-body shot. She sits on a chair, fairly close to the camera, holding a handheld microphone and speaking directly to the viewer, with her free hand gesturing. Her face points straight toward the lens, with no head tilt or rotation, and sits toward the upper-right part of the frame. A small wooden table beside her enters the lower-left corner, with a half-eaten grape bingsu on it.
+
+She is seated outside a street-side café in Korea. Behind her are entrance steps, a wooden shopfront, metal railings, green plants and a few other tables. The background color scheme is navy blue and yellow.
+```
+
+## Clean Girl in a Korean cafe
+
+![Clean-girl presenter in navy knit beside a café counter, with green tiles and wooden shelves](https://storage.googleapis.com/hypit-public-assets/assets/skills/hypit/image-direction/2026-09-10/clean-girl-cafe.webp)
+
+Clean-girl makeup, a sleek ponytail and navy knit give the person a clear style. The wooden counter and iced latte connect the seated composition to the café; pistachio green and walnut support the navy outfit.
+
+```text
+A photograph captured as a single frame from a video actually shot on an iPhone, with the texture of real iPhone footage. The image looks real, without an oily, overprocessed finish. The background is clearly visible, with no depth-of-field blur. Skin texture is natural and fine, and the lighting is natural. The image is coherent and free of visual artifacts.
+
+A Korean-Chinese woman in her early twenties. She is exceptionally, strikingly beautiful, with the looks of a top Korean girl-group idol and very fair skin. She has a sleek high ponytail and beautiful clean-girl makeup, with clear, expressive eyes and a relaxed, confident presence. She has a beautiful figure, very broad shoulders and excellent head-to-shoulder proportions. She wears a navy-blue knitted top in a minimalist clean-girl style, with small gold earrings and a fine gold necklace.
+
+Generate a vertical medium half-body shot. She sits on a stool beside a café counter, fairly close to the camera, holding a handheld microphone and speaking directly to the viewer, with her free hand gesturing. Her face points straight toward the lens, with no head tilt or rotation, and sits toward the upper-left part of the frame. The wooden counter beside her enters the lower-right corner, with a partly finished iced latte on it.
+
+She is in a neighborhood café in Seoul. Behind her are wooden shelves, a tiled counter, an open doorway and green plants beside a window. The background color scheme is pistachio green and walnut brown.
+```
+
+## Coquette girl in a bay-window bedroom
+
+![Coquette presenter with a pink bow and lace top, seated in a green bay-window bedroom](https://storage.googleapis.com/hypit-public-assets/assets/skills/hypit/image-direction/2026-09-10/coquette-bedroom.webp)
+
+Coquette styling is carried by the hair bow, doe-eyed makeup and pink lace. A bay-window alcove supports the centered seat, while green and wood give the pink clothing a distinct place in the palette.
+
+```text
+A photograph captured as a single frame from a video actually shot on an iPhone, with the texture of real iPhone footage. The image looks real, without an oily, overprocessed finish. The background is clearly visible, with no depth-of-field blur. Skin texture is natural and fine, and the lighting is natural. The image is coherent and free of visual artifacts.
+
+An African-American woman in her early twenties. She is exceptionally, strikingly beautiful, with the looks of a young Hollywood film star and beautiful brown skin. She has long black waves with a pink ribbon bow and pretty coquette makeup, with captivating doe eyes and rose-colored lips. She has a beautiful figure, very broad shoulders and excellent head-to-shoulder proportions. She wears a pink lace-trimmed top in a romantic coquette style, with a pearl necklace.
+
+Generate a vertical medium half-body shot. She sits in an upholstered chair in a bay-window alcove, fairly close to the camera, holding a handheld microphone and speaking directly to the viewer, with her free hand gesturing. Her face points straight toward the lens, with no head tilt or rotation, and sits toward the upper center of the frame. The curved window alcove frames her chair.
+
+She is in an apartment bedroom. Behind her are a bay window with floral curtains, a wooden wardrobe and a small vanity with a vase of roses. The background color scheme is sage green and walnut brown, with pink accents.
+```
+
+## Harajuku Decora girl outside a fashion shop
+
+![Decora presenter with pink double buns and a lilac hoodie outside colorful Harajuku shops](https://storage.googleapis.com/hypit-public-assets/assets/skills/hypit/image-direction/2026-09-10/harajuku-decora.webp)
+
+Decora supplies the styling idea; pink double buns, a graphic hoodie and a few accessories make it specific. The shop entrance gives the centered standing view a natural frame.
+
+```text
+A photograph captured as a single frame from a video actually shot on an iPhone, with the texture of real iPhone footage. The image looks real, without an oily, overprocessed finish. The background is clearly visible, with no depth-of-field blur. Skin texture is natural and fine, and the lighting is natural. The image is coherent and free of visual artifacts.
+
+A Japanese woman in her early twenties. She is exceptionally, strikingly beautiful and cute, with the looks of a top Japanese girl-group idol and very fair skin. She has pink hair in two high buns, a few star-shaped hair clips and pretty, playful makeup, with large, captivating eyes. She has a beautiful figure, very broad shoulders and excellent head-to-shoulder proportions. She wears a lilac graphic hoodie in a playful Harajuku decora style, with colorful beaded jewelry.
+
+Generate a vertical medium half-body shot. She stands outside a fashion shop, fairly close to the camera, holding a handheld microphone and speaking directly to the viewer, with her free hand gesturing. Her face points straight toward the lens, with no head tilt or rotation, and sits toward the upper center of the frame. The shop's recessed doorway frames where she stands.
+
+She is on a shopping street in Harajuku, Tokyo. Behind her are a shop window displaying bags and plush toys, capsule-toy machines and neighboring storefronts farther along the pavement. The background color scheme is turquoise and yellow.
+```
+
+## Streetwear guy beside a basketball court
+
+![Streetwear presenter in a navy graphic shirt beside a terracotta-and-teal basketball court](https://storage.googleapis.com/hypit-public-assets/assets/skills/hypit/image-direction/2026-09-10/streetwear-basketball.webp)
+
+A hip-hop-star beauty comparison, short twists and relaxed streetwear establish the person. The bench and open court explain the offset framing, with terracotta and teal giving the surroundings a deliberate palette.
+
+```text
+A photograph captured as a single frame from a video actually shot on an iPhone, with the texture of real iPhone footage. The image looks real, without an oily, overprocessed finish. The background is clearly visible, with no depth-of-field blur. Skin texture is natural and fine, and the lighting is natural. The image is coherent and free of visual artifacts.
+
+An African-American man in his early twenties. He is exceptionally, strikingly handsome, with the looks of a young hip-hop star, brown skin and a magnetic, confident presence. He has short twists beneath a backward baseball cap and expressive eyes. He has an athletic build, very broad shoulders and excellent head-to-shoulder proportions. He wears relaxed New York streetwear: an oversized navy-blue graphic T-shirt and a silver chain.
+
+Generate a vertical medium half-body shot. He sits on a courtside bench, fairly close to the camera, holding a handheld microphone and speaking directly to the viewer, with his free hand gesturing. His face points straight toward the lens, with no head tilt or rotation, and sits toward the upper-left part of the frame. The bench extends beside him toward the open court on the right.
+
+He is beside an outdoor neighborhood basketball court in New York. Behind him are the painted court, a chain-link fence, a mural and trees beyond the fence. The background color scheme is terracotta and teal.
+```

package/skills/using-vvicat-ai-short-studio-cli/references/hypit/playbooks/craft/generated-dependencies.md

@@ -0,0 +1,38 @@
+# Reference relationships
+
+Every image request asks the painter for a complete picture. Connected references bring useful visual memory into that request: a person's appearance, a product's form, a place's atmosphere and geometry, a camera relationship, or a physical state. Some details need close continuity; others give the painter a world to reinterpret from a new viewpoint. Direct what the current picture carries forward and what it now shows. The relationships among images take the shape of those creative needs.
+
+## Let one picture open another view
+
+A podcast host image can establish a person, an encounter and the restaurant around them in one camera view. The image for the other host can use it as an anchor while changing several parts of the picture at once:
+
+| Part of the new picture | How the new view uses the host image |
+| --- | --- |
+| Capture | The same phone-video photographic language. |
+| Person | Cast the other host with their own appearance within this shared encounter. |
+| Shot | The two views form a conversation: the new host sits on the other side, looks toward their partner, and receives the microphone from the corresponding side. |
+| Setting | Both cameras inhabit the same Korean street-side restaurant and palette; this camera sees a different sector of its architecture, furniture and activity. |
+
+The painter makes a full new view. The shared world and conversational geometry connect the pictures even as person, composition and visible background change. The symmetry belongs to the encounter; the second camera sees the part of the place available from its position.
+
+The relationship matters more than copying a backdrop. A two-person encounter can carry who faces whom, where the microphone enters and which part of the place each camera sees. Separate call or reaction feeds can instead establish different rooms; their shared fact is who addresses or responds to whom, not a common wall.
+
+The first host view can also meet a product image in another request. In that picture, the host and much of the camera view continue, while her hands, the phone and its displayed product change. One reference can support very different kinds of variation because each new picture has its own intention. The [podcast and interview image examples](examples/conversation-images.md) show the actual prompts and inputs behind these relationships.
+
+## Give each new picture the references it needs
+
+A useful first camera image can establish person and place together. It may show one person, several people or the whole encounter. When another view needs that world, carry its atmosphere, palette, materials and spatial relationships into the view the new camera should see. Several Takes can then use the resulting images.
+
+A new location can begin with a camera image that already places the people in that setting. Their established views can supply identity or styling; a product or continuing prop can supply its form and state. The new picture establishes the location and composition together, while the prior images carry the facts that persist across the move. A character reference can also carry identity directly into video direction when the new place does not need its own key image.
+
+If one character leaves a living-room argument holding a phone, a hallway view can inherit the person's identity and the phone's relevant state when that moment merits an image. The furniture and camera angle of the living room have finished their job. An intentional change of clothes, time or setting similarly changes which facts the next picture needs to carry. A view may grow from the previous story view when its changing state matters, or branch from a character reference when earlier settings have nothing to contribute.
+
+A portrait may contribute identity while clothes, action and place change. A product-holding variation may carry nearly the whole view while changing the hands and object. When there is no person, the picture can still draw on a product, place, graphic or composition reference. Choose the useful reference by the relationship the next picture needs. Keeping close to the image that actually knows that relationship also keeps incidental details from unrelated intermediate scenes from accumulating.
+
+An existing image can serve the intended view directly. When the work asks for another picture, connect the images that carry its needed facts and describe their roles in the prompt. With several references, “reference 2” means the second image connected to this request; the actual image input gives that phrase its meaning. A request whose references already establish much of the picture may need only a short direction for the change. The selected image package owns the exact reference-input syntax.
+
+Keep useful prompts and source images with the project when later views may need the same relationship. [Image direction](image-direction.md) owns the picture's casting, framing, palette and capture language; [Run Candidates](../../production/authoring.md#reuse-produced-work-explicitly) explain reuse of completed Outputs.
+
+## Carry the shared world into motion
+
+The image graph records visual facts to carry forward, not the video work's shot order or request boundaries. One camera image may guide a passage with several views or multiple Takes in the same world; several images may contribute different facts to one request. A video excerpt can carry performed action or camera behavior that a still cannot show; [video direction](video-direction.md#let-footage-carry-temporal-behavior-that-matters) owns those input choices and the requested action. [Voice and performance](voice-and-performance.md) and [B-roll](b-roll.md) own the resulting footage's roles in the work.

package/skills/using-vvicat-ai-short-studio-cli/references/hypit/playbooks/craft/graphic-compositions.md

@@ -0,0 +1,85 @@
+# Directing graphic composition
+
+Read this for boards, comparisons, charts, titles, cards, diagrams and designed visual fields.
+Start with what the viewer should understand or feel. A graphic can explain, organize, set a tone,
+punctuate a joke or create delight; it does not have to justify itself as literal information alone.
+[Motion graphics](motion-graphics.md) owns how that picture changes over time.
+
+## Compose an idea, not a stack of decorations
+
+Decide what the eye should find first, what supports it and what changes at the next meaningful beat.
+Use scale, spacing, grouping, contrast, color and motion to express that hierarchy. A ranking's rows,
+an interview's unanswered slots and a product comparison's two sides communicate relationships even
+before the viewer reads the words.
+
+Coordinate the whole frame. A presenter's position can leave space for a board; the same accent can
+connect Caption, an answer icon, a question mark and a reveal flash. Treat palette, occupied area and
+brightness together. A large pale panel behind pale clothing can flatten the shot even when each
+element looks attractive alone. Work with the actual material's framing and movement; the next
+section explains how source direction and authored presentation can answer each other.
+
+Make the visual direction concrete before extending it across scenes. Translate a reference's vibe
+into a few shared choices: palette, type, edges, depth and motion character. A small representative
+passage can show those choices together with the actual footage. Carry the accepted direction into
+shared Recipes or project styling so a later change reaches the related components consistently.
+
+Give graphic text its own job: identify, compare or emphasize what the picture needs beyond Caption.
+When removing a redundant heading or label, recompose the remaining content and its spacing. The
+former text slot need not remain an unexplained hole; the demonstration should still feel complete.
+
+## Give material and graphics room together
+
+Keep the generated or recorded scene, its camera view and the final Canvas in view as distinct
+relationships. People interact with the people, props and setting in their scene. The camera frames
+that scene. Authored media views, Caption and graphics organize the final picture. A recorded sign
+can belong to the scene while an explanatory label belongs to a component.
+
+Plan the useful room for recurring graphics when directing new footage. Translate that need into
+camera facts: subject placement, visible body extent, gesture room, distance and background. Give
+the image model those picture requirements; the future MG, editorial text and explanation remain
+with their own components. [Image direction](image-direction.md#frame-the-image-for-what-it-will-become)
+owns the wording. Accepted footage supplies concrete framing to work with: adjust its view, the
+graphic or their shared arrangement according to what should remain legible.
+
+Decide which contribution leads now. A small moving face can command attention while a large quiet
+field supports it. Protect the mouth, meaningful gesture, product fact or text that carries this
+beat. Placement, source crop, occupied extent and paint order all affect the result;
+[Spatial layout](../../production/spatial.md) owns their geometry. The material's A-roll role gives
+it no reserved region or permanent place behind everything else.
+
+For a screen that carries the claim, [screen demonstrations](screen-demonstrations.md) owns the
+choice between captured evidence and an authored explanation. Its view shares the Canvas geometry
+and visual hierarchy of the surrounding composition.
+
+## Put the right facts in editable structure
+
+Use authored text, exact fonts, geometry and state when wording, alignment, timing, data or repeated
+visual relationships need control. Use supplied, found or generated media for photographs, artwork,
+textures, product imagery and other material whose picture is itself the content. A real screenshot
+or finished graphic can remain media; recreating every pixel as a component is not automatically useful.
+
+The boundary follows the work. A sign photographed in a scene belongs to that scene. A score that
+must change on a spoken answer is editable graphic state. A textured paper field may be drawn,
+procedurally made, found or generated; choose according to its intended appearance, theming and
+editing needs.
+
+An icon, portrait, interface or other picture can identify a subject, carry a theme, show a factual
+state or deliver a visual joke. Choose its source from the information the viewer needs. When exact
+current wording, data, UI, private identity or a particular brand asset is itself evidence, use the
+authoritative supplied or found material and record its source and relevant use conditions.
+
+For a familiar public icon, person, object, meme or place whose job is visual recognition or creative
+adaptation, a broadly informed image model can paint the needed material directly or from references;
+a found asset is another ordinary choice. Judge whether the result carries the intended identity and
+graphic role. Reconstruction preserves the useful recognizability and relationship; adaptation may
+replace it with a new theme. Provenance alone does not make one image more visually truthful than
+another. Keep generated prompts and reference edges in Source so their creative intent stays editable.
+
+## Carry the picture into the component
+
+A ranking board can share one visual owner when its rows move and retain state together; a label
+and independent photo can remain peers. That boundary follows behavior, not the size or beauty of
+the graphic. [Component design](../../production/component-design.md) owns the chosen boundary,
+external inputs and author controls. [Motion graphics](motion-graphics.md) owns the change the
+viewer experiences; [review](../../production/review.md#make-the-composition-work) tests how that
+change reads with the actual material.

package/skills/using-vvicat-ai-short-studio-cli/references/hypit/playbooks/craft/h3-prompt-writing.md

@@ -0,0 +1,54 @@
+# H3 Prompt Writing
+
+## Workflow
+
+1. Identify the input mode: T2VA, I2VA, FL2VA, L2VA, or full-reference Ref2VA.
+2. For base text/keyframe modes, read `examples/h3-base-prompt-writing.md` and follow its final prompt structure.
+3. For full-reference mode, read `examples/h3-reference-prompt-writing.md` and follow its six-section rewrite format.
+4. Preserve the exact field names, section order, labels, and timing notation from the selected guide.
+
+## Base Modes
+
+- T2VA: build the full audiovisual timeline from text.
+- I2VA: start from the first frame and develop forward from it.
+- FL2VA: describe the continuous path between the first and last frames.
+- L2VA: infer a plausible opening and converge to the supplied last frame.
+
+Use `integrated_multimodal_description`, `overall_soundscape`, and `non_diegetic_music` in the order shown in `examples/h3-base-prompt-writing.md`.
+
+## Full-Reference Mode
+
+Ref2VA rewrites use `subject_definitions`, `summary`, `retention_analysis`, `detailed_description`, `overall_soundscape`, and `non_diegetic_music` in that order. Reference labels stay consistent across all sections.
+
+Read `examples/h3-reference-prompt-writing.md` for label rules, retention analysis, and complete examples.
+
+## Routing Guard
+
+When the target system separates the video model from the prompt type, verify that the effective prompt type is `h3` before using this skill's output. A valid node-level `h3` or `seedance` override may take precedence over a project setting; clear or correct a stale `seedance` override and read the effective value back. If the H3 model and prompt route do not match, stop instead of silently sending H3 output through a Seedance prompt path.
+
+## Output Rules
+
+- Write rewrite sections in English; preserve dialogue, lyrics, and visible scene text in their original language.
+- Describe each shot by composition, subjects, environment, actions, camera, sound, and the exact point where referenced content appears.
+- Avoid plot summaries, unresolved reference labels, and timing that does not match the requested duration.
+
+## Exact Generated State Sequences
+
+For a countdown, clock or other difficult changing photographed value, create a temporal storyboard
+before video generation. Default to six cells, while choosing another count when it represents the
+required states more accurately. Put each exact reading in its own reviewed cell, preserving formatting,
+typeface, color, scale, placement, perspective, lighting and background. Approve the cells visually,
+attach the board as a Ref2VA image reference, and explain in `retention_analysis` that it supplies state
+and order rather than a visible grid. `detailed_description` gives the exact transition timing and
+requires one normal full-frame video. Reject malformed output, improve the board or prompt, and regenerate
+the smallest failing shot; never patch or locally retouch generated video.
+
+Visible dialogue defaults to joint audiovisual generation. Put the exact line in `<d>`, fit the request
+duration to the measured delivery, describe synchronized mouth movement and delivery in the timeline,
+and specify matching ambience and physical sounds in `overall_soundscape`. Use `non_diegetic_music: N/A`
+when no audience-only score is intended.
+
+For multilingual source dialogue, detect and confirm language per utterance. Put the language code beside
+every `<d>` line and explicitly preserve that turn's original language: for example
+`WOMAN [ja]: <d>待って</d>` followed by `MAN [en]: <d>No.</d>`. Do not translate, transliterate, infer
+language from subtitles, or let H3 collapse several turns into the clip's dominant language.

package/skills/using-vvicat-ai-short-studio-cli/references/hypit/playbooks/craft/image-direction.md

@@ -0,0 +1,95 @@
+# Directing generated images
+
+An image model can paint the visual world the work needs: a person in a camera view, a lived-in place, a product, an icon, or a fantastical event seen through ordinary footage. Each request directs a complete picture. The author gives it a subject, aesthetic force and visible relationships; the model realizes the details that make those choices into an image. References bring selected visual facts into that same act of painting.
+
+For the common phone-video portrait, the [phone-UGC Kit](#use-a-concrete-capture-direction) supplies tested capture wording; the author directs the person, camera encounter and place. The [complete examples](examples/image-direction.md) show prompts alongside their results.
+
+## Two ideas behind every sentence
+
+**A capable painter has visual habits.** “A person at a desk” leaves the face, styling, encounter and room to the model’s usual associations. A casting noun, the kind and strength of beauty, a hairstyle or the social character of a place can shift the whole picture. Choose the few dimensions that matter to this work and direct them clearly; the model can complete the rest.
+
+**A few decisive strokes carry the picture.** For a presenter that may be an arresting face and a clear camera encounter; for a product, its defining form and use; for an icon, a readable visual idea; for a place, a few telling objects and spatial relationships. These choices establish more than a long inventory of incidental details. Spend words where they change the image and let the model complete the rest. “Few” describes the number of important decisions, not a word limit. The [balcony portrait](examples/image-direction.md#japanese-sailor-on-an-apartment-balcony) shows the difference between authored anchors and details the model supplied.
+
+When the work calls for a presenter people want to watch, direct attractiveness with conviction at the character’s actual age. Photographic credibility and exceptional beauty can coexist. Fatigue, comedy or a difficult expression can also be part of the character or moment when the story calls for them.
+
+## Four paragraphs, four responsibilities
+
+For a phone-UGC image establishing a person in a scene, the tested starting shape is the Kit’s fixed **Capture** paragraph followed by **Person**, **Shot** and **Setting**. These are directing responsibilities, not a form to fill attribute by attribute. A reference may already settle a person or place, while a different kind of image may need a different shape altogether.
+
+| Part | What the author decides |
+| --- | --- |
+| Capture | The photographic language; [the Kit](#use-a-concrete-capture-direction) supplies the tested phone-video wording. |
+| Person | Who this is, why they are compelling, and the few appearance choices that make them particular. |
+| Shot | What they are doing, whom they address, and how the camera meets their face, body and surroundings. |
+| Setting | Where this happens, with a palette and visible details that make the place feel inhabited. |
+
+The parts influence each other: an outfit contributes to the palette; a table can make an offset framing feel natural. Keep each paragraph focused without treating the boundaries as walls.
+
+## Seven questions for a presenter image
+
+For a complete person-and-place image, these questions help find the high-value decisions:
+
+1. What kind of captured image is this?
+2. Who is this person, and what kind and strength of appeal should they have?
+3. Which appearance choices make them recognizable rather than generic?
+4. What ongoing activity or disposition can the next video moment continue?
+5. How does the camera frame this encounter, including face, body, gaze and useful space?
+6. What visible relationships make the place and the person belong together?
+7. Which colors give the whole image its intended character?
+
+One phrase can answer several questions; some are already answered by the brief or a reference. They are a way to notice what matters in this picture, not seven required sentences.
+
+## Cast the person with force
+
+When casting an adult presenter in this GPT Image 2 phone-video direction, start with **girl** or **boy** followed by the actual adult age. This tested wording has produced appealing, lifelike adult faces; the noun steers the portrait while the explicit age and role still define the person. For example: *a Chinese girl in her mid-forties, exceptionally beautiful, with the presence of a leading actress of her generation*; *an Indian boy in his late thirties, strikingly handsome, with the easy confidence of a leading actor in his late thirties*. When a reference establishes the person’s appearance, let it carry that identity into the current picture.
+
+Give beauty or handsomeness both **strength and kind**. *Exceptionally, strikingly beautiful, with the looks of a top Japanese idol* carries more direction than *attractive*; a different age, culture or role calls for its own equally strong comparison. Cultural background, visual appeal and a social or aesthetic type—jirai-kei, soft goth, clean girl, a particular kind of founder—can work together, but only the choices that define this character need words. A role alone rarely casts the face.
+
+Choose a few visible anchors that realize that direction: hair around the face, eye makeup or gaze, a memorable garment or accessory. For a half-body human presenter, *very broad shoulders and excellent head-to-shoulder proportions* is a useful structural anchor against the large-head, narrow-shoulder look. Its value is the proportion, not an instruction to expose every part of the shoulders.
+
+A portrait can establish a person’s identity, a chosen appearance cue or a useful camera view. Direct what the current picture carries from it and what changes in the camera view, styling or setting. [Reference relationships](generated-dependencies.md) show how an image can anchor another complete view while its person, composition and visible environment change.
+
+## Choose an idle state for the encounter
+
+A reusable character image gives the video a state it can continue. Speaking toward the viewer with a free hand gesturing is an ongoing activity, not a frozen pose; an alluring, curious or teasing presence can persist through several beats. A wide laugh or a startled face commits the image to one instant, which is useful when that instant is the shot’s subject. Let the intended performance decide.
+
+For a direct-to-viewer presenter, *the face points straight toward the lens, with no head tilt or rotation* gives a dependable starting view while the body can turn and the hands can work. Podcast partners or interviewees instead have a gaze relationship with each other. Their reference images should establish that relationship, with faces readable from the camera angle the work needs.
+
+## Frame the image for what it will become
+
+Direct framing, camera distance, body arrangement, face angle and placement as one encounter. A close half-body presenter can make the face, shoulders and gesture legible; a wider view can carry an action or a second person. Place the face where the later composition needs it, while describing only the physical scene the image model should paint. A table, another person or a doorway may make an offset position natural when it also belongs to the story. An offset can also be a straightforward camera choice.
+
+For a camera view that will sit alongside later MG, direct the photographed scene the viewer will see. An icon painted on a sign or phone screen belongs to that image and travels with it. A ranking icon meant to appear or move independently can be made as its own image Resource; a component can then place and animate that Resource, or draw the icon itself. Describe the person’s actual interaction with people and objects in their scene. Choose the source view for the body, gestures and setting the performance needs; [spatial fitting and crop](../../production/spatial.md) can present it in other viewports. [Graphic composition](graphic-compositions.md#give-material-and-graphics-room-together) owns its relationship with surrounding content.
+
+## Establish the place with a few details
+
+A setting should carry a social world, not serve as a color field behind the face. Name the recognizable place and a few details that explain why this person is there: café seating and a half-finished dessert, a record shelf continuing into an aisle, a classroom door and student posters. A view through a window or an opening behind the person can give the scene distance. Choose objects for their story and visual character, then let the model complete the ordinary remainder.
+
+Give the place a deliberate palette. *Coral red and yellow* or *terracotta and green* can be carried by the actual architecture, clothing, plants and objects rather than painted as a uniform wall. A broad light or dark field can gain interest from the place’s real surfaces and contents—wood grain, tiles, shopfronts, foliage—when those belong to the scene. Describe color as color and material; the Kit already supplies natural-light capture language. Time of day, weather or a particular light event earns a place in the prompt when the story or reference depends on seeing it.
+
+## Use a concrete capture direction
+
+For a phone-video frame, `@hypit/gpt-image-kits/phone-ugc-v1` preserves the tested iPhone capture paragraph. Keep that paragraph intact while directing the picture’s content in ordinary prose. A fantastical person can still inhabit a photographed world; an illustration, product render or other visual goal calls for capture language that serves that goal.
+
+The Kit assembles `person`, `shot` and `setting` after the fixed paragraph:
+
+```svml
+<import as="text" from="@hypit/text@1"/>
+<import as="ugc" source="@hypit/gpt-image-kits/phone-ugc-v1"/>
+
+<text:Render id="portrait-prompt" template={ugc.phone-ugc-v1}>
+  <text:Set name="person" text={portrait-person}/>
+  <text:Set name="shot" text={portrait-shot}/>
+  <text:Set name="setting" text={portrait-setting}/>
+</text:Render>
+```
+
+These slots carry paragraphs, not individual appearance fields; the Kit allows `person` or `setting` to be omitted when a reference already supplies those facts. Pass `{portrait-prompt}` to the image Surface and connect reference images there. GPT Image 2 is the usual choice for this direction; `2K` often serves a full-screen picture or video reference, while `1K` can serve an inset. Aspect ratio and resolution are model parameters, not prompt words. The Kit package README owns its assembly interface.
+
+## Let references supply the facts they own
+
+A supplied portrait can establish identity; a camera view can establish a shared place; a product image can establish the product. Say what each reference contributes and what the current picture should show, and connect the actual images on the Source. When references already provide the difficult visual facts, the prompt can concentrate on the new encounter or viewpoint. The model paints that whole view, carrying forward the facts the work needs and realizing the new ones.
+
+Carry the phone-video capture direction into another view when the work keeps the same photographic world. [Reference relationships](generated-dependencies.md) owns multi-view branches; [transformations](../../creation/transformations.md) owns a face or product swap across the video.
+
+The finished prompt should read as one picture with a clear subject, visible relationships and visual character, with words spent on the decisions that make it useful to this work. The examples preserve complete prompts as evidence of those particular pictures, not as sentences to transfer unchanged into every new one.

package/skills/using-vvicat-ai-short-studio-cli/references/hypit/playbooks/craft/motion-graphics.md

@@ -0,0 +1,68 @@
+# Directing motion graphics
+
+Read this when authored graphics, effects or a coordinated scene change the picture over time.
+Motion graphics can explain a relationship, make a comparison felt, give a joke its timing or
+establish a visual attitude. The work may use a small accent, a moving presenter and diagram,
+or an entire animated passage. [Graphic composition](graphic-compositions.md) owns the hierarchy
+of the picture they share; this page owns what their changes do to the viewing experience.
+
+## Direct the change, not just the entrance
+
+Understand what the viewer sees before an event, what changes, and what remains afterward. A board
+can accumulate answers; a picture can be picked up, moved to a slot and rejected; a camera view can
+recede while another idea takes the foreground. In each case the action gives the next state its
+meaning. A satisfying entrance cannot compensate for a settled state that fails to show the claim.
+
+Follow the identity of an object through its changes. When a card becomes an attachment, the same
+card can keep its image, shape or trajectory; when a demonstrated interface changes style, its
+recognizable controls can remain while its treatment changes. That continuity lets the audience
+understand *what changed* without learning a new diagram at every beat. A clean cut is also a
+strong handoff when the next idea is better understood immediately than through a journey.
+
+An effect can serve feeling as well as literal explanation. A springy arrival may make an answer
+playful; a sudden hard replacement may give a claim force; a slow, precise construction may make a
+relationship easier to study. Choose the motion character that belongs to this film, then give the
+result enough visible life for its content to register. The appropriate hold and pace depend on
+the words, sound, picture size and amount the audience must read.
+
+## Make the motion belong to one visual language
+
+Palette, type, edges, texture, depth, motion and sound can reinforce one another. A retro cartoon
+treatment, a precise mathematical diagram and a spatial 3D scene call for different choices about
+weight, path, perspective and rhythm. Their names are creative directions, not renderer or component
+types. Extract the actual relationships that make the chosen style convincing, and carry shared
+decisions into the project's Recipes or ordinary component dependencies when several scenes use them.
+
+Keep the scene's meaningful objects complete through a restyle. If a folder's contents, an editor's
+controls or a comparison's result carry the idea, preserve their job as the look changes. Recoloring
+only the outer panels changes less than the audience may think, while removing a label can require
+the remaining layout to be recomposed. The film's Treatment owns its particular palette, ornament
+and attitude.
+
+Work with source media as part of that language. A moving viewport changes how an existing Take is
+seen; it does not change the Take's performed time. An independently moving image, a drawn border
+and a caption may share a moment without having the same visual owner. When footage and graphics
+must move as one scene, their shared geometry can be directed together. [Performance](../../production/performance.md)
+owns the current Timeline footage, and [component design](../../production/component-design.md)
+owns the implementation boundary.
+
+## Give events and movement different jobs
+
+An explanation, reveal or comparison can respond to a Script Selection or Moment; a silent animated
+passage can use authored time. The reference locates the behavior and its meaningful changes in
+the work; travel, settling, texture and secondary motion can be designed locally around them.
+[Timing](../../production/timing.md) and [component design](../../production/component-design.md#let-meaning-drive-the-behavior)
+own the exact projection and author inputs.
+
+Derive related movement from the relationships it depicts. A pointer arrives at the control it
+presses; a badge follows the object it labels; a rejected drag returns from the place it reached.
+That keeps a layout revision from breaking the visual action. Coordinate entrance, response,
+persistence and exit across a scene, while leaving independent Caption, sound or media contributions
+free to serve their own roles.
+
+Inspect the changing passage with its actual material and sound. Exact frames can show whether a
+shape, overlap or settled state reads; playback reveals weight, pace and whether the handoff lands.
+Judge the effect at delivery size and in the neighboring picture, including the moment after its
+most impressive motion. [Snapshots](../../production/snapshots.md) and
+[review](../../production/review.md) provide those views; [drawing](../../production/component-visuals.md)
+owns how the selected visual behavior is rendered.

package/skills/using-vvicat-ai-short-studio-cli/references/hypit/playbooks/craft/recreation-policy.md

@@ -0,0 +1,51 @@
+# 参考复刻策略与 P0 门禁
+
+复刻项目先记录参考片来源、权利范围、要保留的镜头结构和要替换的人物、场景、道具。原片只用于分析、对白、音轨与字幕时间；最终视频只引用已选资产图、该镜深度视频、必要的音频参考和按明确风险补充的分镜。原片关键帧和上一镜尾帧不得进入最终生成请求。
+
+项目笔记可以按下列结构记录导演决定；它不替代 CLI 的镜头清单、资产清单和审批文件：
+
+```yaml
+source_languages: [ja, en]
+dialogue_policy: preserve_source_per_utterance
+dialogue_segments:
+  - id: line-001
+    source_shot: S001
+    speaker: female_lead
+    language: ja
+    original: "待って"
+subtitle_language: zh-CN
+audio_policy:
+  voice: generated
+  emotion_reference: source
+  ambience: generated
+  music: generated
+text_policy:
+  generated_subtitles: forbidden
+  allowed_text: [countdown]
+  final_subtitles: postproduction
+shot_policy:
+  preserve_micro_edits: true
+  storyboard: risk_triggered
+  generated_video_repair: regenerate_failed_shot
+```
+
+`allowed_text` 只指导提示词和人工审片，不触发自动 OCR 或阻塞。
+
+## 逐镜制作
+
+1. 用 `hypit recreate init <source> --project <project>` 导入原片并检测镜头，再由导演观看、修订镜头清单，执行 `hypit recreate shots <project>`。镜头按时间顺序连续覆盖整段参考片，不能漏段或重叠；每个提交 Comfly 的镜头必须独立成片，最长 15 秒且不超过 100 MiB。超限镜头先按动作边界人工修订，不按固定秒数盲切。微小剪辑和叙事功能另记在时间线，不把像素切点当成叙事判断。
+2. 先确认对白转写的厂商、模型和费用。[StarRouter 官方文档](https://starrouter.io/docs/zh-CN.md)列出 `/v1/audio/transcriptions`，示例模型为 `qwen3-asr-flash` 和 `whisper-1`。原生复刻已接入 StarRouter `qwen3-asr-flash`：对已切好的镜头先执行 `hypit recreate transcribe <project> --shot <id> --runtime <profile>` 看厂商、模型、语言、音频哈希与预计付费请求；用户确认后分别执行 `--approve` 和 `--confirmed`。结果保存为待审核文本，不含逐词时间戳；必须人工校对原文和说话人，并另行确认时间位置，不能把它当作已对齐字幕。提交结果未知时会阻断重试；查验 StarRouter 日志及费用后，才能用 `--approve --acknowledge-unknown` 显式重新授权。需要逐词时间戳时仍可选 `hypit transcribe` 的 WhisperX 对齐能力，但它可能下载本地模型。原文、译文和字幕分开保存，不从中文字幕倒推对白。
+
+   Hypit 对该模型按[上游规格](https://help.aliyun.com/zh/model-studio/asr-model)施加单次 10 MB、5 分钟的本地限制；StarRouter 文档没有声明自己的文件大小或时长上限。复刻仍须先切镜，再按镜头转写。
+
+   重做同镜转写时显式使用 `--replace`；成功替换会把旧文本保存为 `.replaced.json`，输出 `previous_transcript` 路径。审批会绑定原文本哈希；审批后编辑文本会使审批失效。若付费请求执行期间原文本发生变化，新结果会另存为 `.pending.json` 供人工审核，不覆盖编辑后的文本，也不自动重试。`hypit recreate status <project>` 会列出参考片完整性、当前主转写文件、待审核候选和结果未知的镜头；参考片被移走或改动时仍可只读查看状态，但制作命令会继续阻断。存在待审核候选时不能再次批准同镜付费请求。审核候选后，可用 `hypit recreate transcribe <project> --shot <id> --adopt-pending <文件> --replace` 在本地采用它；命令会核对镜头与音频来源，备份原文本，仍将新文本标为待审核，且不会调用远端模型。
+3. 切镜确认后就设计新人物、场景和道具资产，在 `.hypit/recreation/required-assets.json` 为当前镜头清单的每一镜显式审阅人物、场景、道具需求，并填入已选资产键；没有某类资产也要写空数组。资产生成涉及付费时先核对厂商、模型、参数和费用并取得用户确认。镜头实际重切后逐项重新选版；只关联已删除镜头的旧候选可用 `hypit recreate assets <project> --discard <candidate-id>` 从清单移除，原图片文件保留。倒计时、设备状态、手部接触、精确姿态等有具体视觉风险的镜头，可按所需状态制作分镜板；普通镜头不默认制作分镜板。分镜板只作为经审核的语义参考，成片仍须是单幅全屏视频。
+4. 执行 `hypit recreate preprocess <project>`，为每镜生成独立 Author/Run。逐镜核对 `hypit plan <shot.svrun> --runtime <profile>` 的厂商、经 StarRouter 调用的 Comfly 1194 去字幕和 1776 深度工作流、参数与付费请求；先确认所选账户已开通 `comfly/1194` 和 `comfly/1776`，否则阻断制作。用户确认后才执行 `plan --approve` 和 `build --confirmed`。1194 下载后会拦截与原片明显无关的画面；即使通过自动检查，也须人工核对画面与字幕残留。两次远端处理之间及之后恢复原镜头时间基准，不能把整片直接交给去字幕工作流。资产设计和选版可在深度任务运行期间继续推进，不必等全片深度完成。
+5. 审核深度结果的首尾覆盖、动作对应、字幕残留、视频流、无音轨和时长漂移。漂移超过 5%、空文件或不可解码时阻断；提交回执丢失视为结果未知，只查询原任务，不自动重提。通过后用 `hypit get <build-id> --output depth-map.video --to <project>/references/<shot-id>-depth.mp4 --workspace <project>` 导出，再用 `hypit recreate depth <project> --shot <shot-id> --build <build-id> --file references/<shot-id>-depth.mp4` 校验本地文件与该镜 Build Result 完全一致。实际重切后的旧 Build 不能重新登记。
+6. 按镜头写最终提示词和 Run，并在 `recreate references` 中绑定该镜深度、选定资产与必要音频。逐句保留 `说话人 [语言代码]: "原文"`，明确对白只存在于音频，不显示字幕、翻译、标题、水印、Logo 或 UI。执行 `hypit recreate plan <project> --run <build.svrun> --runtime <profile>` 核对实际媒体绑定、厂商、模型、参数和付费请求；用户确认后才 `--approve`，再 `recreate build --confirmed`。
+
+`clean-text` 已禁用。生成视频中的数字、人物、动作、字幕或局部画面错误，要拒收并重做最小失败镜头；不得用原片像素拼贴或局部覆盖掩盖问题。最终审片和交付另记录结果与剩余风险。
+
+逐镜远端转写如果留下“结果未知”，先核对 StarRouter 原请求和账单，再重新审批并显式使用 `--acknowledge-unknown`。如果进程异常退出时还留下 `transcription-control/<shot>.lock`，先确认没有正在运行的 Hypit 转写进程，再按报错给出的路径移除锁文件；不要在请求运行中清锁重提。
+
+时间基准恢复只能校正帧率与时长，不能找回远端丢失的画面。StarRouter 的 `metadata.input_video` 沿用直连工作流字段，当前接口不返回输入 Schema。去字幕结果是深度生成的参考：自动检查拦截完全无关、空文件或不可解码的输出；局部首尾差异只写入证据，不单独阻断后续制作。3 秒真实样本曾缩为约 2.67 秒，超过 5% 漂移门禁；另一段 4.4 秒样本的输出为 4.333 秒，恢复时间基准后已生成对应的深度视频。人工仍须审核动作、字幕残留和时间基准；超出 5% 的时长漂移不能仅靠拉伸放行。StarRouter 返回的公开任务 ID 必须用于查询和下载；提交结果未知时保留本地业务凭证，不得换业务 ID 自动重提。旧版直连 Comfly 返回 `success, files=[]` 时保留凭证并记为 `COMFLY_OUTPUT_MISSING`。TempFile 上传默认有效期一小时，凭证从 Credential Store 读取。旧 RunningHub 或直连 Comfly 配置仅供读取历史证据，不能继续发起制作；历史故障现象不能直接推断为当前故障率。

package/skills/using-vvicat-ai-short-studio-cli/references/hypit/playbooks/craft/screen-demonstrations.md

@@ -0,0 +1,51 @@
+# Directing screen demonstrations
+
+Read this when a website, app, terminal, editor or other screen is part of what the viewer must
+understand. A screen can document an actual state, show an action with its consequence, or stage
+an explanation of a relationship. Decide which of those jobs this passage needs before choosing
+its material or visual treatment.
+
+## Choose the evidence the claim needs
+
+A screenshot makes wording, arrangement or a particular result available to read. A recording
+preserves an interaction and the interface's actual response. An authored component can isolate a
+concept, choreograph an example or keep changing labels and states editable. These choices can
+coexist in one passage: an actual product recording can establish the behavior, while surrounding
+graphics explain what to notice. A still can also move under a viewport to guide attention across
+its existing contents; that motion expresses a view of the still, not a recorded interaction.
+
+When exact wording, data or behavior is the claim, use material that establishes that fact and keep
+its source identifiable in the project. When the screen is an illustration, direct the invented
+states and actions so the viewer understands the idea they stand for. The film may move between
+evidence and illustration, as long as their roles remain clear in the argument. In one making-of
+piece, actual Studio footage showed the working interface while an authored chat-and-pointer scene
+explained how a request becomes an attached reference. Their different jobs guided the choice of
+material in that work.
+
+## Stage the useful view
+
+Choose the part of the screen that carries the fact or action. The audience may need the whole
+interface for orientation, then a closer view to read a control or notice a change. A crop, zoom,
+cursor or highlight should guide attention while preserving enough context to know *where* the
+action happened. Judge type and controls at the video's delivery size; a sharp full-resolution
+capture can still be unreadable when shrunk into a phone-sized inset.
+
+Give an interaction a comprehensible before and after. Its useful states might include a target,
+an action, visible response and enough time to register the result. An edit can begin with the
+result and return to the cause; the important relationship is what the viewer can infer. When the
+page scrolls, distinguish the moving page or inner list from movement of the viewport, device or
+whole screen card. Those motions say different things about the interface. A real recording can
+carry the interface's own pace; an authored demonstration can place its meaningful changes on the
+work's semantic events.
+
+Treat the screen's surrounding picture as a directing choice. A bare crop can give the interface
+authority and reading room; a photographed device can locate it in a person's world; a graphic
+frame can connect it to the film's visual language. Preserve the readability of the actual screen
+while deciding what the presenter, Caption and other graphics contribute beside it.
+[Graphic composition](graphic-compositions.md) owns that shared hierarchy; [motion graphics](motion-graphics.md)
+owns the animation around or within an authored demonstration.
+
+[Browser capture](../../production/browser-capture.md) owns saving a real page or recording,
+[Media presentation](../../production/media-presentation.md) owns its playback and crop, and
+[component design](../../production/component-design.md) owns an editable screen scene. Choose the
+realization that serves this passage's screen role.