gen-infographic

Objective

Generate multiple infographics for the daily report:

1. Cover infographic — top 4-5 updates across all types
2. Per-type infographics — one for each type that has important (7+) items

---

Output Specs

Image	Filename	When to generate	Aspect
Cover	news_infographic_YYYY-MM-DD.png	Always (if image gen is enabled)	9:16
Model	news_infographic_YYYY-MM-DD_model.png	Type has 7+ score items	9:16
Product	news_infographic_YYYY-MM-DD_product.png	Type has 7+ score items	9:16
Benchmark	news_infographic_YYYY-MM-DD_benchmark.png	Type has 7+ score items	9:16
Funding	news_infographic_YYYY-MM-DD_funding.png	Type has 7+ score items	9:16
KOL	news_infographic_YYYY-MM-DD_kol.png	≥1 item with is_kol_voice: true (NO 7+ score gate — KOL voices are scored 4-7 by design)	9:16
Combined	news_infographic_YYYY-MM-DD_combined.png	Always (for report embed)	long

• Cover is always generated first as a standalone image — even if no per-type sections qualify. Do NOT skip or merge the cover into a combined-only flow
• Per-type sections use 9:16 portrait format
• Combined image: Cover + sections stitched vertically — used for embedding in the Markdown report
• Format: PNG

Social Posting (multi-image, no combined)

When posting to social platforms, use the individual images (not the combined long image):

Platform	Max Images	Strategy
X (Twitter)	4 per tweet	Cover + top 3 section images (by item count)
Xiaohongshu	9 per post (carousel)	Cover + all section images

IMAGE_GEN_TYPES config: Controls which per-type images to generate. Default: auto (only types with 7+ items). Options: all (all types with any items), none (cover only), or comma-separated types like model,product.

---

Title Format Specification

Each news title must include three elements:

Element	Description	Example
Entity name	Company/product/organization name	NVIDIA, Cursor, Midjourney
Event subject	Specific model/product/event name	Alpamayo, Agent mode, V7
Core event	Verb phrase describing the event	releases, open-sources, tops leaderboard, closes funding

Title Examples

✅ Correct Examples:

• NVIDIA Alpamayo Releases Open-Source Autonomous Driving VLA Model
• Cursor Closes $900M Series B Funding
• Midjourney V7 Officially Launches

❌ Incorrect Examples:

• Alpamayo 1 (missing entity name and event)
• NVIDIA Alpamayo (missing event description)
• Open-source autonomous driving model released (missing entity name)

---

Content Detail Rules

Determine number of points based on score, but do NOT use [MAJOR]/[MINOR] tags in Prompt or image:

Score	Number of Points	Card Size
>= 7	3-5 detailed points	Large card (priority display)
< 7	2-3 concise points	Small card

⚠️ Score classification is only for guiding content detail level. Do NOT use [MAJOR]/[MINOR] or similar labels in Prompt!

---

Style Presets

Select a visual style via IMAGE_STYLE config (default: classic). Insert the matching style block into {STYLE_BLOCK} in all prompt templates below.

classic — Clean Editorial Magazine

Style: Clean editorial magazine layout. Off-white (#F5F5F0) background with subtle warm gray grid lines. Bold sans-serif header "MorningAI" in black with a vivid accent color underline. Each card is a white rectangle with soft drop shadow (4px blur, 10% black), separated by generous whitespace. Use a refined accent palette: deep navy (#1A2744) for card titles, muted teal (#2A9D8F) for bullet icons, slate gray (#4A5568) for body text. NO gradients, NO textures, NO background patterns — pure flat white space.
Layout: Cards arranged in a balanced grid with equal gutters (24px). Maximize space for card content (titles and bullet points). Do NOT display score numbers or score badges — let card size and content density convey importance.
Card design: Card title in 18pt bold navy sans-serif, subtitle in 12pt gray italic. Bullet points with small teal dot markers, 14pt regular weight. Thin 1px light gray top-border on each card for subtle separation. NO icons, NO illustrations, NO decorative elements inside cards — text only with strong typographic hierarchy. Prioritize readable text and generous line spacing.

dark — Dark Mode

Style: Dark mode editorial layout. Deep charcoal (#1A1A2E) background. Bold sans-serif header "MorningAI" in white (#FAFAFA) with electric blue (#00D4FF) accent underline. Each card is a dark slate (#16213E) rectangle with subtle 1px border in muted blue (#0F3460), separated by generous spacing. Accent palette: white (#FAFAFA) for card titles, soft violet (#7B68EE) for bullet icons, light gray (#B0BEC5) for body text. NO gradients, NO glow effects — clean flat dark surfaces.
Layout: Cards arranged in a balanced grid with equal gutters (24px). Maximize space for card content (titles and bullet points). Do NOT display score numbers or score badges — let card size and content density convey importance.
Card design: Card title in 18pt bold white sans-serif, subtitle in 12pt light gray italic. Bullet points with small violet dot markers, 14pt regular weight. Thin 1px muted blue top-border on each card. NO icons, NO illustrations — text only with strong typographic hierarchy on dark background. Prioritize readable text and generous line spacing.

glassmorphism — Frosted Glass

Style: Glassmorphism editorial layout. Soft gradient background blending from lavender (#E8EAF6) top-left to pale rose (#FCE4EC) bottom-right. Bold sans-serif header "MorningAI" in dark charcoal (#212121) with warm coral (#FF6B6B) accent underline. Each card is a semi-transparent frosted white panel (rgba(255,255,255,0.65)) with backdrop blur effect, rounded corners (16px), and subtle white border (1px, 30% opacity). Accent palette: charcoal (#212121) for card titles, soft indigo (#5C6BC0) for bullet icons, medium gray (#546E7A) for body text. Soft diffused shadows (8px blur, 5% black) behind each card.
Layout: Cards arranged in a balanced grid with generous gutters (28px). Maximize space for card content (titles and bullet points). Do NOT display score numbers or score badges — let card size and content density convey importance.
Card design: Card title in 18pt bold charcoal sans-serif, subtitle in 12pt gray italic. Bullet points with small indigo dot markers, 14pt regular weight. NO hard borders — rely on frosted glass contrast for separation. Clean, airy, modern feel. Prioritize readable text and generous line spacing.

newspaper — Classic Newsprint

Style: Classic newspaper editorial layout. Warm cream (#FFF8E7) background with very faint paper texture grain. Bold serif header "MorningAI" in deep black (#1A1A1A) with crimson red (#B71C1C) thin rule line below. Each card is separated by thin black hairline rules (1px) — NO card backgrounds, NO shadows, NO boxes. Content flows in a column-based newspaper grid. Accent palette: deep black (#1A1A1A) for card titles in bold serif, dark gray (#333333) for bullet text in serif, medium gray (#666666) for subtitles in italic serif.
Layout: Multi-column newspaper grid (2-3 columns). Large stories span full width at top, smaller stories in side-by-side columns below. Separated by horizontal and vertical hairline rules. NO cards, NO boxes — pure typographic layout. Do NOT display score numbers or score indicators — let column placement and headline size convey importance.
Card design: Card title in 18pt bold black serif, subtitle in 12pt gray italic serif. Bullet points with small em-dash markers, 14pt regular serif weight. Feels like the front page of a prestigious broadsheet. Prioritize readable text and generous line spacing.

tech — Terminal / Hacker

Style: Tech terminal aesthetic layout. Near-black background (#0D1117) with very faint dot grid pattern (8px spacing, 5% white). Bold monospace header "MorningAI" in bright cyan (#00FFCC) with a blinking cursor underscore effect. Each card is a dark panel (#161B22) with 1px border in dim cyan (#1A3A3A), rounded corners (4px). Accent palette: bright green (#39FF14) for card titles in monospace bold, amber (#FFB000) for bullet markers as `>` symbols, light gray (#C9D1D9) for body text in monospace. Each card has a subtle top-left label like `// MODEL` or `// PRODUCT` in dim green (#1A4A1A).
Layout: Cards arranged in a balanced grid with equal gutters (16px). Compact spacing, information-dense. Maximize space for card content (titles and bullet points). Do NOT display score numbers or score badges — let card size and content density convey importance.
Card design: Card title in 16pt bold green monospace, subtitle in 11pt gray monospace. Bullet points with amber `>` markers, 13pt regular monospace. Thin 1px dim cyan border. Feels like a developer dashboard or terminal readout. Prioritize readable text and generous line spacing.

---

Prompt Template

9:16 portrait infographic, MorningAI {YYYY-MM-DD}, {LANG} text content.

Total news items for today: {N}

News cards (display EXACTLY {N} cards):

Card 1: {Entity name} {Event subject} {Core event verb phrase}
- {Point 1}
- {Point 2}
- {Point 3}
- {Point 4}
- {Point 5}

Card 2: {Entity name} {Event subject} {Core event verb phrase}
- {Point 1}
- {Point 2}
- {Point 3}

(... list according to actual item count ...)

CRITICAL RULES:
- Each card title MUST include: Entity name + Event subject + Event description
- Display complete titles, do NOT truncate
- Do NOT display any labels like [MAJOR], [MINOR], or importance markers
- Do NOT display score numbers, score badges, or any numerical ratings on the image
- Do NOT invent items not listed
- Display ALL bullet points for each card
- Cards arranged in a grid layout (landscape orientation)
- Maximize content area — card titles and bullet points are the primary focus
- If fewer than 4 items, use more whitespace and decorative elements
- Cover header MUST include the date: "MorningAI {YYYY-MM-DD}"

{STYLE_BLOCK}

{STYLE_BLOCK} — insert the matching style block from the Style Presets section above, based on IMAGE_STYLE config (default: classic).

---

Usage Instructions

1. Determine Images to Generate

From the report, identify:

• Cover: Top 4-5 items across all types by score
• Per-type: For each type (Model/Product/Benchmark/Funding), check if it has 7+ score items. If yes, generate a per-type image.
• KOL: If ≥1 item has is_kol_voice: true, generate a KOL section image (_kol.png). NO 7+ score gate — KOL voices are scored conservatively (4-7) by design, so applying the standard gate would suppress the section image even when the report's ## KOL Voices block is populated.

2. Build Prompts

Cover prompt: Use the Cover Prompt Template above.

Per-type prompt: Use the Per-Type Prompt Template below.

Per-Type Prompt Template

9:16 portrait infographic, MorningAI — {Type} Updates, {LANG} text content.

Total news items: {N}

News cards (display EXACTLY {N} cards):

Card 1: {Entity name} {Event subject} {Core event verb phrase}
- {Point 1}
- {Point 2}
- {Point 3}

Card 2: {Entity name} {Event subject} {Core event verb phrase}
- {Point 1}
- {Point 2}

(... list according to actual item count ...)

CRITICAL RULES:
- Each card title MUST include: Entity name + Event subject + Event description
- Display complete titles, do NOT truncate
- Do NOT display any labels like [MAJOR], [MINOR], or importance markers
- Do NOT display score numbers, score badges, or any numerical ratings on the image
- Do NOT invent items not listed
- Display ALL bullet points for each card
- Cards arranged vertically (portrait layout), one below another
- Maximize content area — card titles and bullet points are the primary focus
- If fewer than 3 items, use more whitespace and decorative elements
- Do NOT display dates in the header or title — per-type images use "MorningAI — {Type} Updates" only, without any date

{STYLE_BLOCK}

Adjust the style header text to "MorningAI — {Type} Updates" when using the preset.

Combined Prompt Template — not used

Removed. All reports now use Cover (9:16) + Per-Type Sections (9:16) + stitch for the combined report image. Social platforms use the individual images directly.

---

Image Strategy

Image generation produces individual images + a combined long image:

1. Cover (9:16 portrait): Top 4-5 items across all types — always generated first as a standalone file
2. Per-type sections (9:16 portrait): One for each type with 7+ score items
3. Combined (long vertical): Cover + sections stitched vertically — for Markdown report embedding

Important: The cover image must always exist as a separate file (news_infographic_YYYY-MM-DD.png). Do NOT skip cover generation even when producing a combined image.

Social posting uses the individual images directly (cover + sections), NOT the combined long image. See Output Specs > Social Posting for platform limits.

If only 1-2 qualifying items exist, the cover still generates with extra whitespace. Per-type sections are skipped for types with no 7+ items.

Manifest example:

[
  {"prompt": "<cover prompt>", "output": "news_infographic_YYYY-MM-DD.png"},
  {"prompt": "<model section>", "output": "news_infographic_YYYY-MM-DD_model.png"},
  {"prompt": "<product section>", "output": "news_infographic_YYYY-MM-DD_product.png"}
]

---

3. Generate Images

Follow the Image Strategy section above — generate cover + per-type sections + stitch combined.

Option A — Native tool (if supported): Generate each image using your tool's built-in capability, then stitch.

Option B — Python script (default, recommended): Build a manifest JSON and run with --stitch:

cd {SKILL_DIR} && python3 skills/gen-infographic/scripts/gen_infographic.py --batch {CWD}/manifest.json --stitch

All images are 9:16 portrait. Requires pip install Pillow. Produces individual images + news_infographic_YYYY-MM-DD_combined.png.

4. Post-generation Verification (required)

For each generated image:

• Check card count equals N
• Verify each card title includes complete three elements
• Check if titles are truncated
• Confirm no unprovided content
• Confirm no [MAJOR], [MINOR] or similar labels displayed
• Confirm no score numbers or score badges are displayed
• If issues found, regenerate or manually correct

5. Insert into Report

• Cover image: at report beginning
• Per-type images: at the top of each type section

---

Handling Few Updates

When valid news items < 4, do NOT pad with fake content:

Item Count	Layout Strategy
1 item	Centered large card + "Today's Highlight" label + whitespace and decorative elements
2 items	Left-right symmetric or top-bottom stacked
3 items	Three-column even or pyramid layout
4-5 items	Standard grid layout

Core principle: prefer whitespace over fabrication.

---

Notes

• Title uses "MorningAI", news content in the target language (default: English, follows --lang parameter)
• Date display rule: Only the cover image header includes the date (MorningAI YYYY-MM-DD). Per-type section images must NOT include any date — use MorningAI — {Type} Updates only.
• {LANG} — substitute with the full language name: "English" (default), "Chinese", "Japanese", etc., based on the --lang parameter
• Strictly generate based on actual item count, do not force 4-5 items
• Do NOT use [MAJOR]/[MINOR] labels in Prompt
• Use card size and point count to reflect importance differences
• Insert image into report Markdown beginning after generation
• Popular Model Blocklist: unless explicitly in report, do not include models/products not in the report