# ai_ppt_new

**Repository Path**: dkb778/ai_ppt_new

## Basic Information

- **Project Name**: ai_ppt_new
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-07-03
- **Last Updated**: 2026-07-07

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# ai_ppt_new

New AI PPT architecture built around mainstream template-driven generation:

```text
Uploaded PPTX
  -> ai_ppt.template          # low-level PPTX -> TemplateSchema
  -> ai_ppt.template_profile  # Theme + Layout + Component + Content Pattern
  -> planning / layout_engine / rendering / quality
```

The uploaded PPT is treated as a **style and layout reference**, not as a raw
shape-filling target. Strict shape filling should be implemented later as a
separate manifest-based mode.

## Current Modules

- `ai_ppt.template`: deterministic PPTX analyzer and low-level schema compiler.
- `ai_ppt.template_profile`: high-level profile builder for:
  - Theme Template
  - Visual Asset Template
  - Layout Template
  - Component Template
  - Content Pattern Template
- `ai_ppt.content_planner`: prompt-to-DeckIR planner with deterministic fallback and LLM JSON contract.
- `ai_ppt.layout_engine`: deterministic Layout Pattern library, planner, and autofit layer.
- `ai_ppt.rendering`: PPTX writer for `LayoutSpec` based generation.

## Example

```bash
python -m ai_ppt.template.loader \
  --input-dir /path/to/pptx_dir \
  --output-dir /tmp/schema \
  --cache-file /tmp/template_cache.json \
  --force

python -m ai_ppt.template_profile.demo_build_profile \
  --template-schema /tmp/schema/风格4.schema.json \
  --output /tmp/profile/风格4.profile.json \
  --asset-output-dir /tmp/assets/风格4
```

The profile builder exports reusable images from the uploaded PPTX into
`theme.background_assets` and `theme.decorative_assets`. LayoutEngine then
selects background and decorative assets by slide role before rendering.

Generate a 10-slide demo using all built-in layout patterns:

```bash
python -m ai_ppt.layout_engine.demo_generate \
  --template-profile examples/profile/风格4.profile.json \
  --output examples/output/layout_engine_demo.pptx
```

Generate a prompt-driven deck:

```bash
python -m ai_ppt.content_planner.demo_generate_from_prompt \
  --prompt "生成一份智能制造升级方案汇报，重点包括背景、核心能力、关键指标、实施路径和总结建议。" \
  --template-profile examples/profile/风格4.profile.json \
  --page-count 10 \
  --output examples/output/content_planner_demo.pptx
```

Generate with the configured OpenAI-compatible LLM endpoint:

```bash
python -m ai_ppt.content_planner.demo_generate_from_prompt \
  --use-llm \
  --llm-base-url http://111.203.213.68:3092/v1 \
  --llm-model qwen3-32b-fp8 \
  --prompt "生成一份智能制造升级方案汇报，重点包括背景、核心能力、关键指标、实施路径和总结建议。" \
  --template-profile examples/profile/风格4.profile.json \
  --page-count 10 \
  --output examples/output/llm_content_demo.pptx
```

Generate with Qianfan web search context:

```bash
python -m ai_ppt.content_planner.demo_generate_from_prompt \
  --use-llm \
  --enable-search \
  --search-config /Users/liuxiaoyu/PycharmProjects/web_search/3d_fengzhi/config.json \
  --llm-base-url http://111.203.213.68:3092/v1 \
  --llm-model qwen3-32b-fp8 \
  --prompt "生成一份北京两日博物馆游学计划，包含行程框架、四个博物馆安排、实用贴士、教育价值总结。" \
  --template-profile examples/profile/风格4.profile.json \
  --page-count 10 \
  --output examples/output/search_llm_content_demo.pptx
```

Built-in layout patterns:

- `cover_center`
- `section_divider`
- `title_bullets`
- `three_cards`
- `four_cards_grid`
- `metrics_grid`
- `comparison_columns`
- `process_steps`
- `timeline`
- `summary_actions`
- `agenda_grid`
- `two_column`
- `problem_solution`
- `quote_focus`
- `big_number_focus`
- `swot_matrix`
- `table_matrix`
- `case_study`
- `risk_matrix`
- `contact_closing`