ningzimu
1378c5ac44
* docs: require syncing all README translations * docs: simplify README language navigation * docs: add multilingual usage documentation * docs: reference multilingual documentation PR
3.6 KiB
Standard Workflow
Overview
The Codex PPT workflow emphasizes staged confirmation. Instead of immediately generating the entire presentation, it first confirms the outline, style, image backend, and sample slide to reduce rework.
Stage 1: Review the Source Material
The agent first determines:
- The topic and central argument
- The target audience
- The presentation objective
- The required slide count
- Content that must be included or excluded
- Whether any image assets are required
Stage 2: Confirm the Outline
The agent generates outline.md, which usually includes:
- Slide number
- Slide title
- 3-5 key points per slide
- The role of each slide, such as cover, agenda, concept explanation, process, comparison, data evidence, or summary
- Optional visual ideas
- Required image assets and how they will be used
No final slide images, speech.md, or .pptx file should be generated before the outline is approved.
Stage 3: Confirm the Visual Style
The agent proposes 2-3 style directions and recommends one. Candidates come from the 12 built-in styles, including clean professional, scientific defense, hand-drawn technical explanation, McKinsey-style, party and government red, and teaching courseware, as well as your personal style library. The agent can also reproduce a style from screenshots, a PDF, or a presentation you provide. See Styles and Personal Style Library for complete style previews.
After a style is selected, the entire presentation should maintain a consistent visual language, while individual slide layouts may vary according to the content.
Stage 4: Confirm the Image-Generation Backend
The Codex built-in image-generation tool is preferred. The workflow switches to scripts/image_gen.py only when the built-in tool is unavailable, the user explicitly requests an API/CLI fallback, or the current capability cannot satisfy the request.
Once the image backend is confirmed, the entire presentation should use the same backend without switching midway.
Stage 5: Generate and Confirm a Sample Slide
Generate one sample slide first and check:
- Whether the text is clear
- Whether the style matches expectations
- Whether the information density is appropriate
- Whether the colors and layout are stable
- Whether the design can scale to the full presentation
Generate the full deck only after the sample slide is approved.
Stage 6: Batch Generation
After the sample slide is approved, the agent generates origin_image/slide_XX.png one slide at a time. In environments that support sub-agents, one sub-agent generates each slide in parallel to speed up multi-slide production. Every slide follows the same style and image-generation backend approved for the sample.
Stage 7: Quality Review and Fixes
Before assembly, the agent reviews every slide for text clarity, consistency with the outline, clipped content, visual consistency, unnecessary page numbers, and overlapping elements. Slides with serious issues are regenerated using stricter prompts, while minor local issues are preferably corrected with targeted image-editing tools.
Stage 8: Speaker Notes and Assembly
The agent generates the speaker notes in speech.md, then uses assemble_ppt.py to assemble the presentation into a .pptx file. The speaker notes are automatically added to the notes section of each slide.
Stage 9 (Optional): Save the Style
If the presentation uses a custom or adjusted style, the agent notes in the final report that you can save it to your personal style library and reuse it by name in the future. See Styles and Personal Style Library.