video-workbench/projects/2026-06-23-在路上/visual-system/style/style-prompt-master.md

458 lines
18 KiB
Markdown
Raw Permalink Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

# Style Prompt Master
## Status
- Document status: active.
- Project: `在路上`.
- Purpose: reusable style prompt master and reference registry for per-shot prompt construction.
- Scope: A/B/C/D style modules, global style kernel, reference roles, forbidden drift, and shot-type mapping.
- Generation status: no image generation is authorized by this file.
This file defines what each shot prompt should inherit. It does not replace per-shot `sNN-vN-prompt.md` files.
## Skill Mode Record
Mode check command:
```powershell
node C:\Users\wangq\.agents\skills\gpt-image-2\scripts\check-mode.js --json
```
Result:
```json
{
"mode": "B-or-C",
"recommendation": "host-or-advisor",
"garden_mode_enabled": false,
"has_api_key": false,
"base_url": "https://api.openai.com/v1",
"model": "gpt-image-2",
"env_flag_value": "(unset)"
}
```
Interpretation:
- Use the `gpt-image-2` Skill as the prompt-engineering workflow.
- Garden local generation is not enabled.
- Do not call Garden generation scripts from this document.
- If later generation is requested, create or update the relevant per-shot prompt/spec first, then wait for user approval.
## Skill Template / Reference Basis
Skill files read for this master:
- `C:\Users\wangq\.agents\skills\gpt-image-2\SKILL.md`
- `C:\Users\wangq\.agents\skills\gpt-image-2\references\prompt-writing.md`
- `C:\Users\wangq\.agents\skills\gpt-image-2\references\scenes-and-illustrations\concept-scene.md`
- `C:\Users\wangq\.agents\skills\gpt-image-2\references\scenes-and-illustrations\minimalist-mood-scene.md`
Template adaptation:
- `concept-scene.md` informs foreground / midground / background / lighting structure for road-film stills.
- `minimalist-mood-scene.md` informs restraint, reduced spectacle, and mood discipline.
- The final per-shot output should be a realistic cinematic documentary still, not concept art, illustration, poster, or generated-label board.
## Project Sources Actually Used
Project control:
- `project.md`
- `execution-plan.md`
- `visual-system/visual-system.md`
- `slides/slides.md`
- `slides/broad-generation-self-audit-v1.md`
Style process sources read before cleanup and then archived:
- `archive/style/style-anchor-brief-v1.md`
- `archive/style/style-anchor-board-review-v1.md`
- `archive/style/style-anchor-c-repair-review-v1.md`
- `archive/style/style-anchor-board-spec-v1.md`
- `archive/style/style-anchor-trial-plan.md`
Reference policy:
- `visual-system/characters/character-reference-registry.md`
Five-shot validation sources:
- `slides/s01/s01-v1-prompt.md`
- `slides/s01/s01-v1-review.md`
- `slides/s10/s10-v1-prompt.md`
- `slides/s10/s10-v1-review.md`
- `slides/s46/s46-v1-prompt.md`
- `slides/s46/s46-v1-review.md`
- `slides/s47/s47-v1-prompt.md`
- `slides/s47/s47-v1-review.md`
- `slides/s51/s51-v1-prompt.md`
- `slides/s51/s51-v1-review.md`
## Active Style References
Active style references are defined by this registry. They do not all need to live in `visual-system/style/`.
| Anchor | Active reference | Role | Source status |
| --- | --- | --- | --- |
| Common | `visual-system/style/style-anchor-board-v1.png` | Human / visual calibration board; overall A/B/C/D continuity check. | Deterministic board assembled from accepted references. |
| A | `intake/imgs/A.风景主导型蓝图.png` | Landscape-led road composition reference. | Accepted intake blueprint; not copied into `style/`. |
| B | `intake/imgs/B.高位观察型蓝图.png` | High-angle city-edge / system-pressure reference. | Accepted intake blueprint; not copied into `style/`. |
| C | `visual-system/style/style-anchor-c-repair-v1.png` | Repaired muddy starlight sample. | Accepted project-side repair sample. |
| D | `intake/imgs/D.夜路篝火型蓝图.png` | Night-road small-fire reference. | Accepted intake blueprint; not copied into `style/`. |
Non-active C problem reference:
```text
intake/imgs/C.泥泞里的星空蓝图.png
```
Use it only when explicitly discussing why C had to be repaired. Do not use it as the default C style reference.
## Global Style Kernel
Every shot prompt should inherit this kernel unless a later approved revision changes the visual system.
Core visual definition:
> Real documentary texture + restrained poetic surrealism + landscape-led road-film composition.
Chinese working definition:
> 现实纪实质感 + 克制的诗性超现实 + 公路纪录片式风景叙事。
Core composition rule:
> The world is larger than the person, the road is longer than the person, and the traveler is only the coordinate that keeps moving.
Chinese working rule:
> 世界比人更大,路比人更长,人只是持续前行的坐标。
Default prompt kernel:
```text
Photorealistic cinematic documentary road-film still. Real documentary texture with restrained poetic realism. Environment first, road / mud / night / fire / city pressure before the person. Low saturation: cold gray, earth brown, dark khaki, coal black, wet dark blue-gray. Warmth appears only as a small emotional signal, usually firelight. No victory, no tourism, no fashion, no poster heroism.
```
Default negative drift:
```text
Avoid travel advertising, scenic tourism, motivational poster, portrait MV, heroic centered stance, fashion photography, clean commercial city, cyberpunk, neon, sci-fi UI, code, camping lifestyle, group reunion, victory tone, fantasy magic, glowing portal, shoe advertisement, broad sunrise, broad sunset, high saturation, readable text, labels, logos, watermarks, UI.
```
## Five-Shot Validation Backfill
The accepted five-shot trial validates the usefulness of the A/B/C/D prompt language, but it does not validate actual image-reference control.
Evidence:
- `S-01` used A road/world grammar and no character reference. The accepted result confirms the opening no-person rule.
- `S-10`, `S-46`, `S-47`, and `S-51` read character and style references locally for prompt construction, but were generated through a standalone prompt path.
- Therefore, this master may reuse the successful wording and review boundaries, but it must not claim that the style board or character references have been proven as actual image inputs.
Validated first-read rules:
| Shot | Module | First read must be | Reusable boundary |
| --- | --- | --- | --- |
| `S-01` | A | cold muddy road and large empty world | no visible traveler, silhouette, vehicle, or animal; do not attach character references when the shot must be empty. |
| `S-10` | B | dark steel-concrete overpass / city-edge structure | traveler is secondary, about `1/10` to `1/6`, and identity mostly reads through silhouette, coat, backpack, and walking scale. |
| `S-46` | C | wet mud, dark water, puddle edge, grit, rough ground | broken star reflection must be readable but not too clean or cosmic; boot / lower-body trace stays peripheral under about `15%`. |
| `S-47` | D transition | night road and wet ground, not camp portrait | traveler about `1/8` to `1/6`, fire about `3%` to `6%`, road continues beyond the fire. |
| `S-51` | D final wide | road, night, sky, wilderness, weak fire points | traveler about `1/20` to `1/12`; far fires signal others without showing a group. |
Validated scale ladder:
```text
empty opening: no person
city pressure: traveler about 1/10-1/6
muddy starlight trace: boot / lower-body trace under about 15%
near-fire transition: traveler about 1/8-1/6; fire about 3%-6%
final wide: traveler about 1/20-1/12
```
Validated watch points to keep in future prompts:
- A can allow a pale dawn band, but it must not become postcard sunrise.
- B can show distant skyline only as secondary pressure, not glossy city promotion.
- C needs clear stars, but too many clean stars may become cosmic or decorative.
- D transition can show distant fire points, but should not preview the final wide too strongly.
- D final can show attractive stars, but the night must remain documentary and not too pretty.
## Module A: Landscape-Led Road
Working title:
```text
A清晨泥泞旷野世界先于人出现
```
Reference:
```text
intake/imgs/A.风景主导型蓝图.png
```
Reference role:
- style reference;
- composition reference;
- world-before-person reference;
- not an edit target;
- not a strict requirement to preserve the source horizon color.
Use for:
- opening road / wilderness shots;
- long road and continuation shots;
- environment-first shots with tiny or absent traveler;
- cold morning, frost, mud, wind, empty road.
- empty-establishing shots where character references would increase accidental person insertion risk.
Reusable prompt module:
```text
Landscape-led muddy road in a large cold wilderness. Long worn road, wet ruts, frost, rough earth, puddles, stones, sparse terrain, cold dawn or pre-dawn haze. The road and world dominate the frame before any person. If the traveler appears, keep him small and secondary. Low saturation, cold gray and earth-brown palette, documentary realism, quiet continuation.
```
Avoid for A:
```text
No travel-ad scenery, no beautiful postcard sunrise, no hero backshot, no clean inspirational road, no large portrait subject, no polished tourism landscape.
```
## Module B: High-Angle City Edge
Working title:
```text
B高位观察系统边缘里的小人物
```
Reference:
```text
intake/imgs/B.高位观察型蓝图.png
```
Reference role:
- style reference;
- high-observation / city-edge composition reference;
- system-pressure reference;
- not an edit target.
Use for:
- bridge, overpass, underpass, city edge, steel canyon;
- shots where city pressure should make the traveler small;
- high-angle or distant observational compositions.
- shots where the traveler should be read mainly by silhouette, coat, backpack, walking posture, and scale rather than face detail.
Reusable prompt module:
```text
Cold city-edge observational frame. Massive concrete, steel, bridge shadow, wet gray ground, puddles, retaining walls, drain lines, rust traces, outside edge of the city rather than its bright center. The built system is larger than the person. The traveler, if visible, is a small moving coordinate in the lower or side area of the frame. Documentary, utilitarian, low-saturation, no technology spectacle.
```
Avoid for B:
```text
No clean CBD promotion, no glass-tower advertisement, no futuristic UI, no code, no cyberpunk, no neon, no sci-fi city, no commercial skyline as the main subject.
```
## Module C: Muddy Starlight
Working title:
```text
C泥泞里的星空低处看见宇宙
```
Reference:
```text
visual-system/style/style-anchor-c-repair-v1.png
```
Reference role:
- primary C style reference;
- mood / composition / texture / acceptance reference;
- repaired sample replacing the original C blueprint;
- not an edit target unless a future approved edit task says so.
Use for:
- muddy puddle / water / reflection shots;
- soul-image shots;
- broken star reflection inside real mud;
- low ground-level shots where poetry must come from real texture.
- shots where any traveler presence should be only an edge trace, not a full person.
Reusable prompt module:
```text
Dark muddy puddle at night or deep pre-dawn, low ground-level view. Real wet mud, grit, stones, broken shoe marks, dirty water, uneven puddle edge. Broken cold-white star points reflected in the puddle, fragmented by ripples, mud, stones, and water texture. The viewer should first see real mud and dark water, then discover starlight inside it. Any traveler trace stays peripheral and under about 15% of the frame.
```
Avoid for C:
```text
No sunrise, no sunset, no broad warm horizon, no clean mirror lake, no pretty postcard reflection, no fantasy portal, no magic glow, no cosmic spectacle, no shoe advertisement, no full-body traveler, no face, no portrait, no only-mud-without-stars, no only-stars-without-mud.
```
## Module D: Night Road Fire
Working title:
```text
D夜路小火远方也有人在走
```
Reference:
```text
intake/imgs/D.夜路篝火型蓝图.png
```
Reference role:
- style reference;
- restrained firelight reference;
- distant fellow-traveler signal reference;
- not an edit target.
Use for:
- night road;
- small roadside fire;
- late-section warmth;
- distant fire points;
- ending shots where the traveler continues rather than arrives.
- C-to-D transition shots where cold star reflection becomes small real roadside fire.
- final-wide shots where far fire points mean "others are on the road" without group presence.
Reusable prompt module:
```text
Dark wet wilderness road at night or deep pre-dawn. Road and darkness dominate. Warmth comes only from small local orange firelight or weak distant fire points. Fire is a signal, not a destination. The traveler, if present, is road-worn, small or medium-small, upright, tired but capable, and still oriented toward the continuing road. The emotion is unfinished but less alone, not triumphant.
```
Transition variant:
```text
For near-fire transition shots, use a medium-wide road-film frame: environment first, road second, traveler third. Traveler about 1/8-1/6 of the frame. Fire about 3%-6% of the frame, near the roadside, not centered as a camp scene. The road must continue beyond the fire.
```
Final-wide variant:
```text
For final-wide shots, use a very wide road-film landscape frame: environment first, road second, distant fire points third, traveler last. Traveler about 1/20-1/12 of the frame. Fire points are sparse signals in darkness, not a destination, campsite, reunion, or victory.
```
Avoid for D:
```text
No tent, no camping chairs, no cookware, no picnic setup, no cozy campsite, no group gathering, no reunion, no victory pose, no large theatrical campfire, no warm sunset horizon, no arrival feeling.
```
## Shot-Type Mapping
| Shot type | Style module | Style reference behavior | Notes |
| --- | --- | --- | --- |
| Opening empty road / world before person | A | Name A reference and inherit A module. | Character reference should be `none` if no person should appear. |
| Long road with tiny traveler | A, optionally D if night/fire appears | Name A and any secondary D reference. | Use character registry for visible traveler if identity matters. |
| City pressure / bridge / concrete canyon | B | Name B reference and inherit B module. | Avoid all tech/UI/cyber drift. |
| Mud / puddle / reflection / soul image | C | Name repaired C reference and inherit C module. | Use original C only as problem reference if explaining repair, not as active style reference. |
| Fire / distant warmth / late road | D | Name D reference and inherit D module. | Fire must be small and local. |
| Transition from C to D | C + D | Name repaired C and D references; optionally cite previous accepted shot as sequence continuity. | Keep mud/night continuity; do not become camping or final-wide too early. |
| Character detail with road texture | A or C plus character registry | Style module controls environment; character registry controls identity/prop. | Prompt must list both style and character references. |
| Medium or close identity shot | Style module by setting plus character registry | Style reference is not enough. | Do not generate text-only identity-critical shots. |
## Per-Shot Prompt Requirements
Every future `sNN-vN-prompt.md` should include a `Reference Inputs` section.
Minimum fields:
```text
Reference Inputs
- Style module: A / B / C / D / mixed.
- Active style reference(s): exact file path(s).
- Reference role(s): style, composition, mood, palette, texture, sequence continuity, or avoid-drift.
- Character reference(s): exact file path(s) from character-reference-registry.md, or `none` with reason.
- Non-active references: only if used as problem reference / repair context.
- Accepted prior shot reference(s): optional, only for sequence continuity, not as a default style source.
- Generation path note: whether actual image-reference attachment is available, or whether the prompt is standalone.
```
Rules:
- The common style board is a calibration board. It does not need to be attached to every generation.
- A/B/D stay in `intake/imgs/` as accepted source blueprints; do not copy them into `style/` unless a future reference-capable generation package requires a colocated upload set.
- C uses the accepted repaired project-side file in `visual-system/style/`.
- Style inheritance does not replace character reference handling.
- Identity-critical shots must select active character image references from `visual-system/characters/character-reference-registry.md`.
- If the active generation path cannot attach character references, do not generate identity-critical shots as text-only.
- When the output is based only on a standalone prompt, record that explicitly. A standalone success can validate wording and composition, but it does not validate actual image-reference control.
- Each prompt should include the intended first-read order, for example `road/world before person`, `mud/water before boot`, or `night road before fire`.
- Each prompt should include the relevant scale band when a traveler, boot trace, or fire appears.
## Broad-Generation Repair Rule
The repaired lightweight production rule is:
```text
per-shot brief + per-shot prompt + explicit reference list + no per-shot human review + batch-level user review
```
Not:
```text
batch log only + text-only anchors + no traceable per-shot prompt
```
## Common Style Phase Closure Rule
The common style phase is considered closed after these two actions:
1. Extract the reusable style prompt master / registry.
2. Clean `visual-system/style/` so only active style assets remain.
After closure, do not create more common style briefs, board specs, board reviews, trial plans, HTML board sources, or contact sheets by default.
Future work should move to per-shot execution:
- create or repair `sNN-vN-brief.md`;
- create or repair `sNN-vN-prompt.md`;
- select a style module from this master;
- select character references from `character-reference-registry.md` when identity matters;
- record actual generation path and whether image references were truly attached.
Reopen the common style phase only if one of these happens:
- user explicitly requests a new common style anchor or style-board revision;
- A/B/C/D no longer cover a real shot class;
- an accepted active reference is rejected or replaced;
- a reference-capable generation workflow requires a colocated upload package.
## Active Directory Policy
Keep these files active in `visual-system/style/`:
```text
style-anchor-board-v1.png
style-anchor-c-repair-v1.png
style-prompt-master.md
```
Process files such as prior brief/spec/review/trial/HTML/contact-sheet artifacts belong under:
```text
archive/style/
```
Archived style files remain process evidence. They are not default prompt references unless a repair task explicitly reopens them.