| name | ppt-image-to-editable-converter |
| description | Use when converting a frozen imagegen storyboard image into an element-level editable PowerPoint slide by following image-to-editable-ppt: rebuild only text/simple lines/simple boxes natively, crop all other visual elements as complete independent picture objects, render, audit, and retry before the next slide. |
PPT Image-To-Editable Converter
Purpose
Convert one frozen storyboard image into an element-level editable PowerPoint slide. This is not a new design pass and not a redraw-from-scratch workflow. The accepted image is the visual target.
Use image-to-editable-ppt as the authority. In this orchestrated workflow:
- ordinary text and numbers should be editable PPT text when practical;
- only the simplest straight lines, plain dividers, simple boxes, and simple rounded boxes may be native PPT shapes;
- everything else must be preserved as independent crop-safe picture objects: stylized icons, pictograms, robots, illustrations, logos, decorative marks, styled arrows, badges, charts, tables, plots, diagrams, complex panels, photos, screenshots, and source figures;
- each element should be movable, resizable, reorderable, replaceable, or text-editable where appropriate.
Preferred conversion pattern: create named icon/visual crops with source_bbox and display_bbox, then place those named assets in PPT alongside native text and simple shapes. Keep the crop set human-auditable and avoid dense anonymous fragmentation unless the user explicitly asks for pure raster slicing.
Required Tooling
Load and follow image-to-editable-ppt for the image-to-editable conversion. This skill only narrows how that conversion is used inside the orchestrated PPT workflow.
Do not use Presentations, PowerPoint tooling, artifact-tool, JSX authoring, or direct PPT authoring to replace this conversion step. Those tools may be used only after the converted slide exists, and only for rendering previews, inspection, export, or merge evidence.
Inputs
- accepted
storyboard/slide_XX.png
slide_specs/slide_XX.md
storyboard/visual_style_spec.md
- relevant source figures or externally supplied assets
Required Outputs
manifests/slide_XX_conversion.json
editable_slides/slide_XX.pptx
previews/slide_XX.png
qa/slide_XX_object_audit.json
Conversion Rules
Use a fidelity-first element conversion strategy:
- Preserve the accepted storyboard layout, scale, z-order, visual density, colors, and hierarchy.
- Rebuild ordinary slide text, labels, titles, footers, numbers, and captions as editable text when practical. Fix OCR, wrapping, font size, and alignment drift.
- Rebuild only simple straight lines, plain dividers, simple boxes, and simple rounded boxes as native shapes.
- Crop every other separable visual object as its own picture object when feasible: icon, logo, robot, pictogram, badge mark, styled arrow, illustration, chart, table, plot, diagram, complex panel, photo, screenshot, decorative element, or source figure.
- Name every crop semantically and record it in
assets/slide_XX_crop_inventory.json or an equivalent inventory before PASS.
- Prefer multiple complete object crops over one broad crop.
- Do not use a full-slide image as the final slide structure.
- Do not use a broad crop for a whole section when text, icons, arrows, cards, or labels can be separated.
Crop completeness is mandatory:
- no clipped letters, icons, borders, arrows, shadows, glows, antialiasing, legends, axes, figure edges, or labels;
- use padded
source_bbox for icons or figures with soft edges, shadows, circular strokes, glow, or content near the boundary;
- use asymmetric padding when the object sits next to text or another icon;
- if padding captures neighboring content, split the objects or adjust
source_bbox and display bbox;
- crops must not unintentionally overlap unrelated neighboring content in the rendered PPT preview;
- completeness beats tightness. A slightly loose complete crop is better than a clean but clipped crop.
Text Handling
Text is one of the few things allowed to be rebuilt. Keep it native when practical and visually faithful. Use separate text boxes for separate logical blocks: title, subtitle, bullet group, card title/body, legend, footer, callout, or annotation.
Do not hide ordinary slide text inside a broad crop. A text crop is allowed only when the text is highly stylized or inseparable from a logo, screenshot, chart/table/plot, source figure, or visual mark; record that reason in the manifest and object audit.
Icon, Chart, And Figure Handling
Stylized imagegen icons are visual assets. Preserve them by cropping from the accepted storyboard image; do not replace them with generic icon libraries or newly drawn approximations.
Charts, tables, plots, complex diagrams, external source figures, screenshots, photos, logos, dense plots, trajectory maps, and 3D renders should remain as picture objects unless the user explicitly requests native chart/table editability. Isolate them as independent objects and include full internal text, axes, legends, colorbars, shadows, and figure edges.
If a higher-fidelity external source asset replaces the matching storyboard region, it must preserve the same position, size, crop, and visual role. Do not introduce new visible content that was absent from the storyboard.
Prohibited Shortcuts
- no full-slide screenshot as final slide structure;
- no broad section crop when smaller text/icon/shape objects can be separated;
- no redesign after storyboard PASS;
- no replacing stylized icons with unrelated generic icons;
- no accepting clipped icon or text crops because the first conversion is close enough;
- no accepting unintended crop overlap or crop residue;
- no dense anonymous fragment sets replacing named, reviewable visual crops;
- no rebuilding charts/tables/plots/complex visuals natively unless the user explicitly requests it;
- no adding new facts, labels, icons, or content that were absent from the storyboard.
QA And Retry
The first converted PPTX is a draft. Render the PPTX and compare it to the accepted storyboard image.
Blocking defects require manifest fixes and another conversion pass:
- any icon, logo, pictogram, arrow, shadow, glow, border, text, label, axis, legend, or figure edge is clipped;
- crop padding overlaps, covers, or duplicates unrelated neighboring content;
- ordinary text is missing, wrong, overflows, wraps differently, or is trapped in a broad crop without justification;
- stylized icons were redrawn, substituted, or simplified;
- multiple separable icons were bundled into one crop;
- crop assets are anonymous fragments rather than named semantic objects;
- a broad crop or full-slide image acts as the slide structure;
- chart/table/plot was rebuilt natively without explicit user request;
- visible layout, scale, z-order, color, or density drifts from the storyboard.
The slide may not reach PASS and the next slide may not start until these defects are resolved.
Object Audit Rules
qa/slide_XX_object_audit.json must include:
- total object count;
- editable text object count;
- editable simple-shape object count;
- picture object count;
- largest picture object bbox and percentage of slide area;
- list of picture objects with bbox, source, source_bbox when used, and reason;
- path to the crop inventory and confirmation that crop names/roles are meaningful;
- confirmation that stylized icons are preserved as complete independent crops;
- confirmation that no full-slide or broad crop acts as the slide structure;
- confirmation that crops do not unintentionally overlap neighboring content;
- render method used for
previews/slide_XX.png;
- known risks for visual QA.