version: "1.0.0" name: tvc-ai-shot-prompter description: TVC AI shot prompter for image and video generation. Use when Codex needs to create structured image prompts, Seedance-style video prompts, cinematic long-take video prompts, node-asset reference prompts, negative prompts, visual continuity rules, actor/product consistency, camera movement, lighting, color, duration, ratio, sound cue notes, regeneration instructions, or prompt tables for TVC storyboard shots.

TVC AI Shot Prompter

Use this skill to convert storyboard shots into executable image prompts and video prompts.

The image prompt and video prompt use different methods:

• Image prompt: use the 8-layer still-image structure.
• Video prompt: use the Seedance-style video structure based on referenced materials, segmented timing, camera movement, audio, and product continuity.
• For complex single-shot cinematic action prompts, use the cinematic long-take structure in references/video-prompt-method.md: base assets, scene, sound, atmosphere/quality, visual tone, color, shot design, action chain, ending event, and negative constraints.

Workflow

1. Create continuity anchors:

• Brand/product identity.
• Actor identity, wardrobe, makeup, hair.
• Location, props, art direction.
• Lighting, color, lens, camera behavior.
• Reference assets and how to cite them with @ when a video model supports material references.

1. Load prompt methods when needed:

• Read references/image-prompt-method.md for still image prompts.
• Read references/video-prompt-method.md for video prompts.

1. For each storyboard shot, write:

• Image prompt using: Subject + Action + Environment + Composition + Camera + Lighting + Color + Style.
• Video prompt using: reference material + duration/ratio + time-segmented action + camera movement + product/actor continuity + sound/music + negative constraints.
• For prompts with multiple referenced characters, props, vehicles, creatures, or node-asset IDs, first create a 基础设定 block that locks each asset's identity, costume/material, behavior, and allowed expression/state changes.
• Negative prompt.
• Continuity lock.
• Regeneration notes.
• If the user asks for Excel/表格/xlsx, preserve storyboard columns and append prompt columns so the table can be exported directly.

1. Keep prompts executable:

• Use concrete visual facts instead of abstract emotions.
• Write actions, not feelings.
• Separate lighting from color palette.
• Put style reference last.
• Do not overload one shot with multiple unrelated events.

Output Format

全片一致性设定：
素材引用规则：
镜头 Prompt 表：
重生成策略：
下游交接：

Prompt table columns:

| 镜号 | 图片 Prompt | 视频 Prompt | 负面约束 | 连续性锁定 | 重生成备注 |
| --- | --- | --- | --- | --- | --- |

Excel-ready combined table columns:

| 镜号 | 时间码 | 时长 | 景别 | 画面内容 | 镜头运动 | 产品/logo | 声音/字幕 | 图片 Prompt | 视频 Prompt | 负面约束 | 连续性锁定 | 重生成备注 |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |

Image Prompt Method

Write still-image prompts in this order:

1. Subject.
2. Action.
3. Environment.
4. Composition.
5. Camera.
6. Lighting.
7. Color Palette.
8. Style Reference.

Do not start with style. Do not write abstract identity only. Describe visible traits, visible actions, and concrete cinematic grammar.

Video Prompt Method

Use Seedance-style video prompt logic:

• State referenced assets with @ when available, such as @product_reference, @character_reference, @scene_reference.
• Preserve native asset tags such as <node-asset>...</node-asset> when the target platform uses them. Describe what each asset controls immediately after the tag.
• Specify duration and aspect ratio.
• Break motion into time segments when the shot has a beginning, middle, and end.
• Describe camera movement, subject movement, product movement, and ending frame.
• Add sound, music, SFX, or silence cues when useful.
• For product shots, protect package shape, logo, label, material, and scale.
• For one-take action shots, separate 分镜, 景别, 构图, 运镜手法, and 画面内容 so the model gets both cinematography and event order.

Prompt Quality Bar

• State what must remain identical across shots.
• Put product shape, logo visibility, and hand interaction rules in the prompt when relevant.
• Avoid impossible camera moves, contradictory light, and multiple time jumps in one generated shot.
• Negative prompts must block extra fingers, warped logos, unreadable text, face drift, product deformation, random subtitles, and unwanted style shifts.
• If a prompt says "失落", "高级", "清爽", or "电影感", translate it into visible action, composition, camera, light, color, or sound.