iml-pptx

name

iml-pptx

description

Create Chinese PPT pages through a prompt-to-image-to-editable-PPT workflow. Use when the user wants 中文PPT图片提示词, PPT image prompts, 16:9中文PPT页面图片, 直接生成PPT图片, 根据提示词生成图片, 把图片/PDF/截图复刻成可编辑PPT, 转成可编辑PPTX, 按图片完全复刻PPT, or full production from topic/source material to generated page images and editable PowerPoint slides with editable text, cards, tables, lines, callouts and high-quality icon/image assets.

IML PPTX

Production Goal

Use this skill to make real Chinese PPT pages through any part of this chain:

1. write prompts for Chinese 16:9 PPT page images;
2. generate the corresponding PPT page images;
3. convert generated or supplied images into visually matched editable PPT slides;
4. assemble or continue a deck page by page.

Core rules:

• Treat a supplied/generated page image as the visual source of truth.
• Do not redesign a different deck when the user asks for replication.
• Keep text, boxes, tables, lines, arrows, labels, callouts and layout frames editable whenever practical.
• Icons, small logos, photos, screenshots and generated bitmap visuals may be non-editable high-resolution assets.
• Preserve the page image's layout, hierarchy, density and business-PPT style.

Default visual system unless the user provides another style: 16:9 Chinese business-training content page, white background, dark gray/black text, technology blue accent, readable Chinese typography, PingFang SC / 苹果苹方 for recreated PPT. For important content pages, default to rich professional executive business-PPT image prompts before editable PPTX replication; for drafts or minor pages, a simpler minimal infographic style is acceptable.

Reference Index

Load only the references needed for the current task:

Task	References
Decide page type, layout rhythm, or avoid repetitive decks	references/page-planning.md, then references/prompt-writing.md
Write PPT image prompts	references/page-planning.md, references/design-draft-prompts.md, references/image2-visual-quality.md, references/prompt-writing.md, optional references/examples.md
Generate PPT page images	references/page-planning.md, references/design-draft-prompts.md, references/image2-visual-quality.md, references/prompt-writing.md, optional references/external-visuals.md
Replicate image/PDF/screenshot into editable PPT	references/editable-replication.md, references/qa.md
Product/software/website training PPT	references/external-visuals.md, references/prompt-writing.md, references/editable-replication.md if making PPTX
Icon-heavy or connector-heavy page	references/icons-connectors.md
Unsure how to execute a common request	references/workflow-recipes.md
Need command examples	references/scripts.md

Start Brief

Before choosing a mode, create or infer a short production brief unless current context already contains one. Include:

制作简报
- 主题：
- 受众：
- 交付模式：prompt / image / editable-ppt / full
- 样式主题：
- 页面节奏：
- 图标/logo策略：
- 图片策略：
- 外部取材策略：
- 字体：
- 来源约束：
- 页面类型策略：
- 版式多样性约束：
- 视觉野心：
- 文本安全：
- 可编辑边界：
- 首个确认页：

Use python3 scripts/start_brief.py ... when a generated checklist would help.

Default choices:

• 样式主题: white background, dark text, technology blue accent, business-training infographic.
• 字体: PingFang SC.
• 图片策略: use photos/screenshots/generated visuals only when they clarify the subject, scenario, product, industry or evidence.
• 外部取材策略: when a public product, software, website, company, person, place, device or real object is central, acquire official/verified visuals unless forbidden or sensitive.
• 页面类型策略: choose page type from content intent first; do not start from a reusable PPT template.
• 版式多样性约束: no more than two adjacent pages may use the same layout family unless the source explicitly requires it.
• 视觉野心: for important content pages use executive; for quick drafts use polished or standard.
• 文本安全: generated images and editable PPTX must keep text inside cards, bands, table cells, labels and callouts.
• 可编辑边界: text/cards/tables/lines/diagrams editable; icons/logos/screenshots/photos may be image assets.

For detailed external visual acquisition rules, read references/external-visuals.md.

Mode Router

Decide the user's real intent from the latest request:

• Prompt mode: user asks for 提示词, prompt library, page image prompts, or reusable prompt rules. Read references/page-planning.md, references/design-draft-prompts.md, references/image2-visual-quality.md, and references/prompt-writing.md.
• Image mode: user asks to 根据提示词生成图片, 直接生成图片, or continue generating PPT page images. Read references/page-planning.md, references/design-draft-prompts.md, references/image2-visual-quality.md, and references/prompt-writing.md; if real visuals are needed, also read references/external-visuals.md.
• Editable PPT mode: user asks to turn images/photos/PDF pages/screenshots into editable PPT, 复刻, 转成可编辑PPT, or continue later pages. Read references/editable-replication.md, references/qa.md, and references/icons-connectors.md when icons/lines are present.
• Full production mode: user asks to go from topic/PDF/source to prompts, images and editable PPT. Read references/page-planning.md first, then all relevant references as the workflow reaches each phase. Full mode is visual-first by default: make or obtain page images before building editable PPTX unless the user explicitly asks for a code-native deck or the image-generation/preview tool is unavailable.

When the user says 不是重新构建一套PPT, 完全按照刚刚的图片, 图片对应的PPT复刻, or 先执行一张我确定下, switch to Editable PPT mode and treat the page image as canonical.

Workflow

1. Identify topic, audience, slide count, source constraints and delivery mode.
2. Run the Page Type And Visual Diversity Gate before writing prompts or building PPTX. Produce a page decision card for every page: page type, content relationship, layout family, visual ambition, visual anchor, icon/connector plan and why this layout fits.
3. Identify public visual entities that may require screenshots/photos/logos. If present, run the External Visual Acquisition Gate in references/external-visuals.md before final prompts or PPT layout.
4. If source is PDF/image, render or inspect pages first. Extract title, core conclusion, modules, diagram type, visual hierarchy, icon/logo needs, image asset needs and likely editable objects.
5. If writing prompts, read references/design-draft-prompts.md and references/image2-visual-quality.md, then create concise copyable Chinese prompts with page type, visual ambition, title, core conclusion, modules, layout, visual system, icon/image strategy, text safety constraints and avoid-list.
6. If generating images, create one image per prompt, save with stable page numbers such as P01_标题.png, inspect each image, and regenerate if text containment, layout, visual ambition, icons or connectors fail.
7. If making editable PPT from a new topic/source in full mode, treat the generated or supplied page image as the canonical visual source. Rebuild the PPTX from that image; do not skip directly from outline to a generic coded deck.
8. If replicating to PPT, rebuild page by page or in small similar batches. Recreate editable objects first; insert high-quality icons/photos/screenshots as image assets only where appropriate.
9. Run mandatory QA before reporting completion.

Mandatory Gates

Page Type And Visual Diversity Gate

Before prompt writing, image generation, or full PPTX production, classify every slide by content intent and page type. Do not pick a visual template first. Use references/page-planning.md.

Required:

1. Write a page decision card for each page.
2. Choose a page type and layout family from the content relationship, not from habit.
3. Vary layout families across the deck; if three adjacent pages share the same structure, revise the page rhythm unless the source demands repetition.
4. Include the page type and layout family in every image prompt.
5. For full production, generate or obtain a page image before editable PPTX replication. If this cannot be done, say so explicitly in the final response and describe the fallback used.

Image2 Visual Quality Gate

For image-generation steps, do not underprompt the image model into generic minimal cards. Use references/image2-visual-quality.md.

Required:

1. Set a visual ambition level for every generated page: standard, polished, or executive. Default to executive for important business decks unless the user asks for a rough draft.
2. Include concrete composition devices, such as numbered section headers, central model diagrams, screenshot evidence panels, bottom value bands, capability maps, process loops, comparison matrices or branded title strips.
3. Use stronger visual hierarchy than a plain code-native PPT skeleton: scale contrast, section rhythm, icon systems, subtle shadows, pale panels and purposeful accent color.
4. Keep generated text short enough to render cleanly. If a card needs more than 2-3 lines of body text, split or summarize before generation.
5. Reject and regenerate images where text escapes boxes, text overlaps borders, tiny text becomes unreadable, or the page looks like a generic template rather than a designed business slide.

Visual Asset Gate

Use real or verified visuals when they materially improve a product, software, website, case, public person, place, device, UI or industry-context slide. Save selected assets and source notes in the project asset folder. Put useful visuals into the PPT, not only into notes or prompts. Details: references/external-visuals.md.

Icon Gate

For every slide with 3 or more modules/cards/steps, create an icon plan or manifest before building the PPT unless the source image is intentionally text-only. Use scripts/icon_picker.py, scripts/icon_narrative.py, scripts/render_icon_manifest.py, or scripts/render_lucide_icon.py as needed. Details: references/icons-connectors.md.

Connector Gate

Treat connectors as semantic objects. Before drawing lines, define purpose, necessity, type, anchors, avoid zones and layer. Use scripts/connector_narrative.py for complex pages and scripts/connector_guard.py after building decks with connector lines. Details: references/icons-connectors.md.

Text Containment Gate

Every generated or editable text object must live inside an explicit safe text zone. Do not place body text by eye over a card or callout; compute or reserve inner bounds first.

Required:

1. For image prompts, specify that all Chinese text must remain inside its visible card, band, table cell, label or callout with clear inner padding.
2. For editable PPTX, define an inner text box for every card/callout/table cell. Default minimum padding: 0.14 in horizontal and 0.10 in vertical.
3. Text boxes must not extend outside their parent card, badge, table cell, note band or caption frame.
4. For generated PPTX, enable shrink/wrap behavior where the library supports it, but do not rely on autofit alone. Also reserve enough height.
5. If text overflows during image inspection or PPTX preview, repair in this order: shorten copy, add a line break, widen/raise the text box, increase container height, lower font size by 0.5-1 pt, then rebalance the grid.
6. Run scripts/text_fit_guard.py when possible before screenshot QA. Treat warnings as repair targets, not as a substitute for visual preview.

Screenshot QA Gate

For every task that generates, edits, repairs or replicates a PPT/PPTX page, create and inspect a visual preview unless the user explicitly says not to preview.

Required:

1. Produce a screenshot/thumbnail after building the PPTX.
2. Inspect the screenshot with the available visual tool.
3. If only one page changed, preview that changed page.
4. If the preview tool only renders the first slide, create a temporary one-slide preview for the changed page or use another renderer.
5. Do not treat XML checks, file size, layout_guard.py or connector_guard.py as substitutes for visual preview.
6. In the final response, mention the preview path or which preview image was inspected.

In Codex desktop, prefer:

scripts/preview_pptx.sh output/deck.pptx output/preview

Detailed QA and repair rules: references/qa.md.

Output Modes

• Single Prompt: short opening line plus one polished prompt block.
• Prompt Library: short system summary plus prompts grouped by page type.
• Theme-Adapted Prompts: state what visual system stays fixed; per-page prompts should say 统一风格，不统一死版.
• Generated Images: generated image files, page numbers and filenames; no long prompt explanation unless requested.
• Editable PPT From Images: .pptx files, preview status, completed pages, and any non-editable icon/logo/photo/screenshot assets.
• Full PPT Production: prompt plan, generated page images, editable PPT replication, final checked PPTX or per-page PPTX files.

Scripts

Use helper scripts when they reduce repeated manual work. Invoke Python scripts with python3.

• scripts/smoke_test.py: run a non-destructive smoke test for core helper scripts.
• scripts/start_brief.py: create a production brief template.
• scripts/icon_picker.py: search offline/lucide icon candidates by English or Chinese business meaning.
• scripts/icon_narrative.py: draft page-level icon logic and manifest skeleton.
• scripts/connector_narrative.py: draft connector strategy from design pattern and modules.
• scripts/render_lucide_icon.py: render one lucide icon to SVG/PNG.
• scripts/render_icon_manifest.py: render a page icon manifest to SVG/PNG assets.
• scripts/build_icon_resources.py: rebuild offline icon resources after pack/library changes.
• scripts/ppt_replicate_helpers.py: reusable low-level OOXML helpers for editable PPT builders.
• scripts/layout_guard.py: heuristic PPTX overlap preflight; inspect warnings manually.
• scripts/text_fit_guard.py: heuristic text-density and text-box fit preflight; repair warned text before final preview.
• scripts/connector_guard.py: scan likely accidental diagonal/unstable connector lines.
• scripts/preview_pptx.sh: create Quick Look preview PNGs for visual QA.

See references/scripts.md for command examples and references/workflow-recipes.md for common end-to-end recipes.

Page Types

Infer page type when missing:

• 总览页: chapter overviews, frameworks, maps, taxonomy.
• 概念页: defining a term or explaining a judgment.
• 对比页: A/B, before/after, can/cannot, old/new.
• 流程页: steps, methods, lifecycle, workflow.
• 方法论页: rules, principles, practices, playbooks.
• 场景页: applications, use cases, industry examples.
• 问题页: common mistakes, risks, diagnosis.
• 数据页: indicators, market size, metrics, trends.
• 案例页: company examples, benchmark samples, implementation stories.
• 总结页: chapter close, key takeaways, action points.

Style And Boundaries

• Reply in concise Chinese for Chinese PPT workflows.
• When the user asks for action, produce files rather than only advice.
• Keep final image prompts copyable and avoid poster/webpage/marketing-page language for content slides.
• Do not drift into long-form slide copywriting unless requested; compact page design drafts and module copy are allowed when needed for high-quality image prompts.
• Do not assume every page uses the same left-right split.
• Do not claim every object is editable if icons/logos/photos/screenshots are inserted as image assets.
• Do not promise exact official brand logos unless actually sourced or available.