Imagegen PPT Workflow
Use this skill to produce PPT decks where every final slide is a raster image and the .pptx is only the final container.
Overview
Follow 4 phases in order:
- Plan the deck from user materials and requirements.
- Confirm the plan and visual direction with a style preview and then a whole-deck thumbnail sheet.
- Generate each final slide as a complete PNG.
- Package the ordered PNGs into a
.pptx.
Treat slides_plan.md as the source of truth. Do not skip approval gates.
Inputs To Collect
Collect or infer these anchors before writing the plan:
- Source content: paper, report, notes, thesis, outline, transcript, or existing slides.
- Deck goal: defense, research report, creative pitch, class talk, investor deck, product intro, or other presentation scenario.
- Audience and speaking context.
- Page count target or approximate duration.
- Required utility pages: cover, agenda, chapter divider, summary, outlook, appendix, thank-you page.
- Style intent: creative, defense, research, business, editorial, or another clear direction.
- Any style reference images, templates, school or brand identity, and required colors.
- Whether the user wants strict chapter naming, speaker-friendly phrasing, or visual-heavy condensed text.
If some anchors are missing, make reasonable assumptions and state them in the plan summary.
Phase 1: Outline And Style Preview
1. Write the outline first
Create slides_plan.md in the working directory before generating the deck.
Read references/outline-format.md before writing it.
The plan must include:
- Slide order and count.
- Page type for each slide.
- Exact chapter or section titles.
- The intended purpose of each slide.
- The visible text hierarchy to show on the slide.
- Any image or figure requirement.
- Notes about whether the slide is a utility page such as agenda or chapter cover.
Keep the plan concise and presentation-facing. Do not dump raw paper text into every slide.
2. Generate a single style preview image
Use $imagegen to produce one preview image that expresses the confirmed direction.
The preview should usually be one of these:
- a cover slide preview,
- a chapter divider preview,
- or a body slide preview that best shows the intended style.
Read references/prompt-recipes.md before writing the prompt.
Always show the user:
- the concise deck structure,
- the main assumptions,
- the style preview image,
- and the parts that still need confirmation.
Stop for confirmation after this phase.
Phase 2: Revision Loop And Whole-Deck Thumbnail
1. Revise until the user is satisfied
If the user requests changes:
- update
slides_plan.md, - regenerate the style preview if the style changed,
- and summarize what changed.
Repeat until the user explicitly approves both content structure and style direction.
2. Generate a whole-deck thumbnail sheet
After plan approval, generate a low-fidelity review surface that shows nearly the whole deck at once.
Use one of these patterns:
- Generate low-fidelity slide previews first, then assemble them into one board.
- Generate a single storyboard-style board directly if that is faster and still readable enough.
The review board is for rhythm, density, and style consistency only. It does not need final-quality text or figures.
If you generate individual low-fidelity previews, assemble them with scripts/make_contact_sheet.py.
Expected output:
preview_slides/slide-01.png... rough previews when neededdeck_overview.pngor another clearly named overview board
Show the overview board to the user and ask for one more approval before final slide generation.
Phase 3: Final Slide Image Generation
After the user approves the overview, generate the final slide images.
Hard rules
- Every final slide must be a complete PNG image.
- Do not rely on native PPT text boxes, shape construction, HTML screenshots, or code-drawn slide layouts as the main output path.
- Keep the visual system consistent across slides: typography, spacing, color, banner treatment, and chart style should remain coherent.
- Use imagegen prompts that contain both the fixed global style block and the slide-specific content block.
- Generate in slide order and save with stable filenames such as
final_slides/slide-01.png.
Quality guidance
- Prefer concise visible text, especially for research and defense decks.
- Convert long prose into presentation language instead of copying paragraphs.
- When a slide is figure-heavy, prioritize clarity and layout over decoration.
- If the user provided must-keep figures or screenshots, either use them as image references during generation or clearly separate those pages so the user can replace them later.
- If Chinese text fidelity is critical, keep wording short and grouped into clear title or card structures rather than dense paragraphs.
Phase 4: PPTX Packaging
Once the final PNG slides are approved, package them into a .pptx.
Use scripts/images_to_pptx.py to place each ordered image on a full-slide blank layout.
Expected output:
final_slides/slide-01.png...slide-NN.pngdeck-name-images-only.pptx
The .pptx is a delivery container for the rendered images. It is not the primary editing surface.
Standard File Set
Use these names unless the user has a better convention:
slides_plan.mdstyle_preview.pngwhen you copy the chosen preview into the workspacepreview_slides/for rough deck previewsdeck_overview.pngfinal_slides/slide-01.pngonward<deck-name>-images-only.pptx
Resource Loading
Read these files only when needed:
- references/outline-format.md: read before writing or revising
slides_plan.md - references/prompt-recipes.md: read before prompting style previews, overview boards, or final slides
scripts/make_contact_sheet.py: use when assembling many preview PNGs into one overview boardscripts/images_to_pptx.py: use when packaging final PNGs into a.pptx
Completion Checklist
Before finishing, confirm that you have:
- a reviewed
slides_plan.md, - an approved style preview,
- an approved whole-deck overview board,
- final slide PNGs in stable order,
- and a packaged
.pptx.
Scan to join WeChat group