PPT-Forge|Image-Generated Templates and Canva Editable Rebuild
Transform the content and advanced image-generated visuals from a source PPT into a presentation that has been rebuilt with Canva layers and is fully editable. This Skill provides a reusable production pipeline; the project name, business data, color scheme, and total number of pages are all determined by the current input.
Startup and Scope
- Read the user's source PPT, materials, and reference images, and select or copy a sample configuration. A full 20-page example is available in another configuration. In the example,
source_pptx: nullindicates that a real source file must be provided. - Use the current user's request as the task boundary.
target_slide_countis the total number of target slides. Whenmode: pilot, visual drafts, reconstruction, and export are only generated forpilot_pages(default is the first 5); the full source structure is still parsed first, and the full template mapping is planned. Full mode works based on the actual total number of pages, not hardcoded rules like 24 or 20. - The original file is read-only; first create a backup copy and record its SHA-256 hash. All modifications occur on the working copy or a new Canva design. If no source file is available, you can create an input checklist and a blank ledger, but you cannot claim that parsing, image generation, or reconstruction has been completed.
- Confirm that image generation capability, Canva decomposition/editing, and PPTX export paths are available for this task. Canva is a mandatory step in this workflow. If capabilities are missing, save a handoff package and explain the specific blocker; do not implement a workaround that bypasses Canva. See Canva Operations Guide for capability verification.
Fixed Order and Stage Artifacts
Proceed strictly according to the table below. Pass the stage check to continue; only pause for confirmation at user-requested review points or when the current tool explicitly requires a save/merge confirmation.
| Stage | Required Work | Evidence to Proceed to Next Stage |
|---|---|---|
| A Source Parsing & Prefilled Planning | Extract original text, notes, structure, charts, images, placeholders, layers, and reference image relationships page by page. Determine each page's purpose, complete prefilled text, layout, and image positions; record the mapping from original pages to target pages. | page-plan.json, text/fact ledger, source page previews, and complete image index; do not generate backgrounds first and decide content later. |
| B Templates & Allocation | Based on the content plan, generate a series of 3–5 text-free background templates that cover the cover, content pages, end page, and any necessary variant layouts. Lock colors, light direction, materials, margins, title areas, and content areas. | Actual template files, style specification, unique template mapping for each target page; even in pilot mode, do not reuse the "two backgrounds" rule from other projects. |
| C Full-Page Visual Drafts per Slide | For each slide, input its template, full prefilled text, corresponding reference images, and real image assets to generate a complete visual design. Save prompts, inputs, and versions per slide. | Complete visual images and traceable inputs for this batch. The full-page image serves as a design reference at this stage; text may appear in the image, but it cannot be treated as the final editable slide. |
| D Canva Decomposition & Reconstruction | Use Canva to decompose the full-page image into editable elements. Individually verify and reconstruct independent text, cards, connectors, charts/tables, and transparent decorative components. Use the prefilled text to correct OCR errors. | Per-slide design/ID, object ledger, independent editing verification, and reconstruction preview; primary text and data should not remain in the full-page bitmap. |
| E Batch QA & Static Integration | For this batch, compare text, facts, layout, template consistency, and editability page by page. Merge into a new Canva working draft, export, and check the static PPTX. | Correct page order/count; missing characters, duplicate background text, overflow, rasterization, and missing resources have been fixed; per-slide QA records exist. |
| F Unified Animations & Final Delivery | After all target slides in the set have been reconstructed and static QA passed, apply animations/GIFs uniformly, test playback, and perform the final export. | Full set gate evidence, animation test results, final PPTX, and handoff index. Pilot mode defaults to no animations/GIFs. |
For complete stage gate and rework rules, see workflow-and-gates.md. Do not start animations midway through Stage C, and do not use "one full-page image per slide" to pretend Stage D is complete.
Visual and Editability Judgments
- Infer the style from the source PPT and current reference images. Information density is determined by the configuration; use clear titles, evidence images, cards, and connector relationships to convey density—do not rely on endlessly shrinking text to fill space.
- Backgrounds and complex illustrations can be bitmaps. All business titles, body text, numbers, labels, and source attributions must be independent text. Card bodies, simple geometric structures, connectors, and data charts should be reconstructed as Canva native objects whenever possible, and verified after export.
- When complex lighting effects, illustrations, or glass decorations require bitmaps, create transparent PNGs (without business text) individually and place them independently; record both the ability to edit the component itself and to move it as a whole. PNGs are not editable shapes or editable charts.
- First try Canva's
image_to_designtype capabilities, then inspect actual objects. Capability names, parameters, and export support depend on the current tool. If connectors lack operations like creating shapes, complete them within the Canva editor and record the actions; do not assume the conversion will perfectly restore all layers. - Plan navigation structures and decorative components based solely on the current source draft and current requirements; do not pre-populate components exclusive to other projects.
Traceability and Real Data
Use the source materials as the text baseline, but "the source PPT wrote it" does not equal "fact verified." For every key number, award, patent, customer, and business performance metric, record the source location and status: verified, provided by materials but not independently verified, estimated, or pending verification. Mark missing data with 【Needs Verification】 and do not fabricate it. AI conceptual images cannot be presented as real product photos; numbers and brands from reference images cannot be transplanted as facts for this project. See Data and File Contracts.
Use stable IDs for each page, template, and component. Record relative paths, hashes, sources, and Canva IDs in asset-index.json, and list the project absolute root path separately during handoff. For each round, save completed pages, missing items, and the next action. When resuming a task, first read the index and handoff; do not regenerate assets that have already been accepted.
On-Demand Resources
- Quick Start: Controller Prompt, Kickoff Example.
- Stage A: Parsing and Prefill Prompt, Page Plan Example.
- Stage B: Template Prompt.
- Stage C: Per-Slide Image Generation Prompt.
- Stage D: Canva Rebuild Prompt, On-Demand Transparent Component Prompt, Object Ledger Example, Capabilities and Export Boundaries.
- Stage E/F: Batch QA and Motion Prompt, Acceptance Checklist.
- Networking, paths, breakpoints, and handoffs: Operations Guide, Handoff Template.
- Package scope and limitations: Source Scope.
- Optional offline helper: project_manifest.py only performs configuration checks, original file backup, ledger initialization, and path indexing; it does not generate images, call Canva, create PPTX, or replace visual acceptance.
Actual Deliverables
Deliver the in-scope editable PPTX, per-slide previews, Canva design links and page mappings, templates and independent components, complete visual drafts, fact and asset index, QA records, and handoff records. For trial productions, clearly state "Completed first 5/N pages" and do not claim full set completion. Mark any untested conversion, editing, or playback capabilities as unverified; when blocked, accurately deliver the stages already completed.