| name | visual-story-prompts |
| description | Generate visually coherent Lovart/Nanobanana image prompts for multi-card WeChat story posts. Use when the user has a story plan document and needs to create AI image generation prompts that maintain visual continuity across different card styles (character-focused, panel/grid, text-focused). Also use when the user mentions 'out-image instructions,' 'Lovart prompts,' 'visual story,' 'card story,' 'image story continuity,' or 'carousel post prompts.' |
Visual Story Prompts
Generate visually coherent Lovart (Nanobanana) prompts for multi-card image stories, ensuring continuity across different card styles.
Target Tool: Lovart + Nanobanana
Lovart is an AI design agent (not a simple image generator). Users interact through chat-based dialogue: describe what you want, Lovart generates it, then you refine through follow-up natural language instructions without rewriting the full prompt.
Key capabilities that affect how we write prompts:
| Capability | Implication for Prompt Writing |
|---|
| Conversational iteration | First prompt establishes direction (composition, color, character position, text layout). Details (expression tweaks, light adjustments) are refined through follow-up dialogue. No need for a "perfect" first prompt. |
| Touch Edit | Lovart can modify selected regions while preserving the rest. Expression/layout changes don't require full regeneration. |
| Reference images (up to 14) | Screenshot-based cards: upload the original screenshot directly, mark with 【附加参考图】 in the prompt. Lovart interprets layout and content from the reference. |
| Character consistency | Nanobanana maintains consistent facial features, clothing, and character elements across multiple generations. Keep the character description prefix identical across all character-focused prompts. |
| Natural language understanding | Nanobanana responds to detailed natural language — no special syntax needed. But physical/visible descriptions execute better than poetic/abstract ones (see Step 5 Expression guidelines). |
| Resolution | Supports 2K/4K output. Default 3:4 vertical. |
Theory Foundation
This skill draws from two sources. Read references/shot-grammar.md for full theory.
From "Understanding Comics" (Scott McCloud):
- Panel-to-panel transitions: how adjacent images create narrative flow
- Closure: reader's brain fills gaps between panels — smaller gaps = smoother story
- Time-as-space: image size/layout controls perceived time passing
From "Master Shots" (Christopher Kenworthy):
- Shot size controls emotional distance (wide = context, close = intimacy)
- Camera angle conveys power dynamics and vulnerability
- Movement direction creates continuity between cuts
可可角色参考
参考图: references/可可_角色参考.png
给 Lovart 生成含可可的图时，务必附上这张参考图，并在 prompt 中包含以下角色描述前缀：
角色「可可」：圆脸小女孩，浅蓝色尖顶帽兜（帽兜连着衣服），金黄色及肩短发从帽兜两侧露出，
大黑眼睛，粉红脸颊，微笑。穿浅蓝色连帽长裙+同色打底裤+浅蓝靴子。
整体配色以浅蓝为主，风格圆润可爱，扁平插画风。
跨卡一致性要点：
- 帽兜形状（尖顶、贴合头部）和衣服是一体的，不是分开的帽子
- 头发金黄色，从帽兜两侧自然垂下
- 表情默认是温和微笑，根据场景调整（参考 Step 5 Expression 写法）
- 全身配色保持浅蓝色系，不同场景可微调明暗但不改色相
Core Concept: Three Shot Types as a System
Each card style is a stable "shot type." The skill does NOT modify how individual card types are prompted — only how they connect.
| Shot Type | Video Analogy | Narrative Function | Lovart Output |
|---|
| Character-focused (可可为主) | Main footage | Carry emotion, advance story | Full scene + character + text, 3:4 |
| Panel/Grid (分格) | Picture-in-picture / reaction shot | Show evidence + instant reaction | Full manga-style multi-panel page, 3:4 (Lovart generates the entire grid layout with panels, gutters, reactions in one prompt) |
| Text-focused (文字为主) | Title card / voiceover | Breathing room, topic pivot | Background frame or pure text layout, 3:4 |
Default format: All cards are 3:4 vertical. Panel/grid cards are NOT assembled in Canva — Lovart generates the complete multi-panel composition directly.
Workflow
Input: Story plan document (with card sequence, content, emotions, style types)
↓
Step 1 — EXTRACT: Parse card sequence into structured list
Step 1.5 — ASSESS: Evaluate trimmability, pacing, and reorder opportunities
Step 2 — ANCHOR LINE: Define visual continuity thread (color curve + recurring props + lighting)
Step 3 — TRANSITIONS: Tag each card-to-card transition and apply bridge rules
Step 4 — SHOT PLAN: Assign shot size + angle per card for rhythm
Step 5 — GENERATE: Write Lovart prompts with bridge annotations
Step 6 — VERIFY: Check adjacent pairs for continuity
Step 1: Extract Card Sequence
Parse the story plan into a table:
| # | Style | Content Summary | Emotion | Notes |
|---|
| ... | ... | ... | ... | ... |
Step 1.5: Coherence Pre-Assessment
Before defining visual anchors, evaluate the extracted card sequence for trimmability, pacing, and reorder opportunities. This prevents generating prompts for cards that will be cut during production.
1. Trimmability Rating — Add a column to the Step 1 table:
| Rating | Meaning | Criteria |
|---|
| Essential | Cannot be removed | Carries unique narrative function (opening, climax, CTA) |
| Trimmable | Could be removed if needed | Narrative pattern overlaps with an adjacent card, or serves a pure transition function that neighbors could absorb |
| Recommend trim | Should probably be removed | Duplicates an adjacent card's style AND content pattern (e.g., two consecutive panel cards showing similar feedback types) |
2. Pacing Review — Check the overall rhythm:
- Are there 3+ consecutive same-style cards? If so, consider inserting a breathing card or trimming one.
- After evidence-dense sections (multiple panel/text cards), is there a digestion space (character card)?
- Would reordering improve momentum? (e.g., moving a group-evidence card earlier to establish breadth before individual evidence)
- After any proposed reorder, re-derive the color curve — reordering invalidates the original color interpolation.
3. Style Switch Frequency — Check transitions are neither too dense nor too sparse:
- 2+ consecutive cross-style switches (e.g., Character→Panel→Text) need enough bridging space between them
- Long same-style runs (4+ character cards in a row) may lose visual variety — consider if a style switch would improve rhythm
Output: Updated Step 1 table with "Trimmability" column + a brief assessment paragraph noting any recommended trims, reorders, or pacing adjustments. Present to user for decision before proceeding to Step 2.
Step 2: Define the Visual Anchor Line
Three continuity threads that run through ALL cards regardless of style:
Color Curve — Plot the entire story's color progression as a continuous gradient. Each card is a point on this curve. When style switches happen, the new card's background color must be derivable from the curve, not an arbitrary default.
Recurring Props — Choose 2-3 small props that appear across card types:
- Primary prop (appears in most cards)
- Secondary prop (appears occasionally)
- Micro-detail (subtle)
Lighting Direction — Maintain consistent light source direction throughout. If character cards have warm light from the upper-left window, panel cards should show the same directional warmth as edge glow.
Step 3: Tag Transitions and Apply Bridge Rules
For each adjacent card pair, determine the transition type and apply matching bridge rules.
Transition Types:
| Transition | When | Bridge Required |
|---|
| Same-style | Character → Character | Light (color + emotion progression) |
| Cross-style entry | Character → Panel/Text | Heavy (color bridge + element traversal + gaze link) |
| Cross-style return | Panel/Text → Character | Heavy (return signal + spatial anchor) |
| Breathing | Any → Text | Medium (color bridge + breathing space) |
Bridge Rules — Read references/bridge-rules.md for full examples.
Core rules summary:
-
Color Bridge: Next card's dominant color = transition point on the color curve. Panel/text card backgrounds derive from adjacent character card colors, never from an independent default palette.
-
Element Traversal: When switching styles, at least one visual element "crosses over" — props, light, particles, texture, or frame elements.
-
Gaze Direction: Character's gaze/body orientation in the current card points toward the next card's visual center of gravity.
-
Return Signal: When returning from Panel/Text back to Character, include a spatial anchor proving we're in the same world, plus a progression signal showing time passed.
Step 4: Shot Size Planning
Vary shot sizes across character-focused cards for emotional rhythm. Do NOT use the same shot size for every card.
| Shot Size | Emotion | Use For |
|---|
| Wide/Full | Establishing, calm, context | Opening, breathing moments, scene transitions |
| Medium | Following, narrative | Most story-telling cards, default |
| Close-up | Intensity, intimacy, impact | Emotional peaks, key reactions |
| Detail | Emphasis, symbolic | Props, hands, specific objects |
Rhythm Rule: Alternate between at least 2 shot sizes. Never use 3+ consecutive same-size shots.
Step 5: Generate Lovart Prompts
For each card, generate the prompt following its style-specific template. The templates below are structural guides — adapt the content to match the story.
Character-focused prompt structure:
[Art style base]. [Character description].
Scene: [Scene description with spatial anchors from Step 2].
[Bridge annotation]: [Color from curve], [recurring prop placement], [lighting direction].
Expression (physical): [REQUIRED. Physically visible markers that Lovart can directly execute:
eyes (open/closed/half-closed, pupil size, gaze direction),
mouth (corners up/down, lips parted/pressed, teeth showing),
eyebrows (raised/furrowed/relaxed),
cheeks (blush intensity),
body language (posture, hand position, shoulder tension)]
Expression (mood note): [OPTIONAL. Poetic/atmospheric supplement for intent, not execution.
e.g., "the quiet of digesting something heavy" — helps convey mood direction
but Lovart's rendering relies on the physical field above.]
Color palette: [Exact colors from color curve position].
Shot: [Shot size from Step 4], [angle].
Hand-drawn text:
[Text content with position, size, color, emphasis marks]
[Style closing]. [Aspect ratio]
Expression Writing Guidelines: Physical descriptions execute far better than poetic ones in Nanobanana. "Eyes wide open, pupils slightly dilated, mouth slightly ajar" produces a consistent shocked expression. "Struck by something she hasn't processed yet" produces unpredictable results. Always lead with physical markers; add mood notes only when the emotional intent isn't obvious from the physical description alone.
Panel/Grid prompt structure (Lovart generates entire multi-panel page):
Make a [art style] page of manga/comic panels, 3:4 vertical format.
Layout: [Panel arrangement description — e.g., "left large panel (2/3) with right column split into two smaller panels"].
Large panel: [Content description — screenshot placeholder area, or main scene].
Small panel 1: [Character reaction — specific expression and pose].
Small panel 2: [Character reaction — specific expression and pose].
Background color: [Color derived from color curve, NOT arbitrary default].
Panel borders: [Style — e.g., white gutters, rounded corners].
[Bridge annotation]: [Shared elements with adjacent character cards — edge glow, particles, props].
[Style closing]. 3:4 vertical.
Lovart's agent (Nanobanana) handles the panel composition internally. Describe the layout in natural language and let the agent interpret it. Include specific panel ratios (e.g., "left 2/3, right 1/3 split top/bottom") for consistency.
Text-focused prompt structure:
[Layout style]. [Aspect ratio]. Hand-drawn text, [background color from curve].
[Layout description with text zones].
[Bridge annotation]: [Corner decorations matching recurring props], [directional light matching character cards].
[Style closing].
Step 6: Continuity Verification
After generating all prompts, verify each adjacent pair:
If any check fails, revise the specific prompt with the bridge rule annotation.
Resources
references/
- 可可_角色参考.png — 可可角色设计参考图（生成含可可的图时必须附上）
- shot-grammar.md — Theory from "Understanding Comics" and "Master Shots" adapted for card stories
- bridge-rules.md — Detailed bridge rules with before/after prompt examples
