How to use it

Behind the scenes each upload runs through:

1. Parse — read every slide's text and images.
2. Plan — decide which sections the page is built from and in what order (always Hero-first).
3. Generate — write each section's content from its slides.
4. Place images & render — drop in deck pictures and show the page.
5. Edit — rearrange, reshape, and regenerate in place.
6. Keep it — save to a shareable link, download the HTML, or print to PDF.

Composing itself is stateless: everything happens in memory for that visit until you click Save. Saving is what writes the page to storage and gives it a URL (see Save & share below); without a save, a refresh starts over.

• Read: slide titles, bulleted/shape text, table cells, SmartArt/diagram text, and speaker notes.
• Reading order & grouping: text is read in visual order and grouped by layout — a multi-column slide is read column-by-column, so each heading stays with the list beneath it (not all headings first, then all lists). Groups are separated by a blank line so the AI sees the structure.
• Source tags: each block is labelled with where it came from and what it is — heading, eyebrow, callout, table, diagram. The tags show as small badges in the parsed-slides panel and are passed to the AI, so a table reads as a table.
• Eyebrow marker: a line that starts with a dot (e.g. .about us) is treated as an eyebrow/kicker — kept verbatim, kept on its own line, and placed in the section's eyebrow.
• Skipped: hidden slides, and slides with almost no text. A slide that is only an image is still kept, so its picture stays available; the count of ignored slides is reported after composing.
• Not extracted: text baked into charts (unless you turn on vision text reading — see below).
• Google Slides: File → Download → Microsoft PowerPoint (.pptx), then upload that.
• Limits: files up to 50 MB; individual images over 12 MB, or in formats a browser can't render (EMF/WMF/TIFF), are skipped.

• Ignore speaker notes (on) — compose from on-slide text only. Untick it when your notes carry real copy rather than presenter asides.
• Keep text verbatim (on) — the generator places your wording as-is instead of rewording or polishing it. Untick it to let the AI write tighter marketing copy from the same facts.
• Keep sections in slide order (on) — the page is built top-to-bottom in deck order. Untick it to let the planner reorder for narrative flow (which is also what lets a Marquee Strip be hoisted to just after the Hero).
• Ignore deck images (off) — compose text-only; every image slot stays a placeholder.

The options you composed with are remembered on a saved page, so a later regeneration keeps the same wording and image policy.

On the upload screen (whole page):

• By default each usable slide becomes one section, in order. Ask to change that — “combine the intro slides”, “aim for about 6 sections”, “split the dense roadmap slide” — and the AI will merge or split accordingly.
• Change emphasis or order — “lead with the metrics”, “drop the team bios”.

In a section's editor (that section only):

• Adjust tone or focus — “use a warmer tone”, “lead with the revenue figure”.
• Pull a specific deck image by its slide — “use the image from slide 3” (see Images below).
• A section's instructions are combined with your page-level instructions.

It's free text, so be specific — vague asks give vague results. The AI tries to honor it but always stays within the facts on your slides.

• Text layer + slide screenshots (default, recommended): reads the file's text and also attaches each slide's screenshot to the planning and generation steps. It's what lets a five-column comparison or a positioned slider be understood correctly. The button shows “Rendering slides…” while the screenshots are produced.
• Text layer only: skip the screenshots and compose from text alone — faster and cheaper when a deck is plain and the text is enough.
• Read text from screenshots (vision): instead of using the file's text layer, the AI transcribes each slide straight from its picture, grouped in visual reading order. Reach for it when a deck's underlying text is messy, out of order, or partly missing (e.g. text baked into charts). The button shows “Reading slide text…”, and the “Parsed slides” panel relabels to “Slide text — read from the slide images (vision)” so you can see exactly what the model read.
• Visual notes: when screenshots are in play, the planner can attach a short note about how a slide looks; it shows under the slide in the parsed panel as Vision: and is passed along when that section is generated or regenerated.
• Trade-offs: vision text is the priciest path (one read per slide) and depends on render fidelity — a rendering glitch can drop text the file's layer had — so it's an opt-in for hard decks, not the default. If a slide can't be read it quietly falls back to the parsed text with a note.
• Availability: the choice appears only when the environment has a slide renderer configured, and the vision option only when an AI key is configured too. If it isn't shown, the tool is composing from the parsed text layer.

Edit content — a form built from that section's own fields, no AI call:

• Edit text directly, pick options from dropdowns (colors, icons, layout variants), toggle switches, and add / remove / reorder list items — bullets, cards, tiles, panels, and the like.
• Image slots show a thumbnail picker of your deck's pictures: click one to fill the slot, or None to clear it (see Images below).
• Item limits: some layouts cap how many items they allow — e.g. Bento holds 2–4 blocks — so Add disappears at the maximum and Remove stops at the minimum.
• Changes apply when you click Save in the editor. If something breaks a section's rules the save is blocked and the reason is shown inline.

Regenerate — let the AI rewrite the section:

• Change its layout to any section type — the content is regenerated into the new layout's shape.
• Jump it to a specific position, and add per-section instructions (combined with your page-level ones).
• Images already placed in other sections are off-limits, so a regenerate can't duplicate a picture that's already on the page.
• This mode needs an AI key — and, on a saved page, the stored deck. When either is missing the tab explains why and content editing still works.

Reorder & remove:

• Drag a section or use the up/down arrows — manual moves aren't constrained, so you can arrange sections however you like.
• The trash icon deletes a section; it's disabled when only one section is left.
• Heads up: on a page you haven't saved, edits live only in the current browser session — a refresh clears them. On a saved page, edits are kept only when you click the page's Save button.

• To place a specific picture, name its slide in an instruction — “use the image from slide 3”. The slide numbers match the parsed-slides list.
• Or place it by hand: in a section's Edit content mode, every image slot has a thumbnail picker of the deck's pictures — pick the one you want (or None). Each thumbnail is labelled with its slide.
• Add your own: use “Additional images” on the upload screen to bring in custom pictures (up to 30). They join the picker in every section (labelled Uploaded); the AI won't auto-place them — you place them yourself. Removing one from the thumbnail strip also resets any section that used it back to a placeholder.
• An unfilled picture slot renders a dashed placeholder box naming what belongs there, so you can see where an image would land before you pick one.
• If nothing fits a deck image, it's left out and you'll see a note; change a section's layout to make room for it.
• Tick “Ignore deck images” to compose text-only — every image slot stays a placeholder.
• Before you save, images ride inside the page itself, so an exported HTML file is fully self-contained. Saving uploads them to storage and the saved page points at those URLs instead.

• The link is the key. The id is random and unguessable, and there's no login — anyone with the /pages/{id} link can view and edit. Send the /shared/{id} variant when you want view-only.
• Saving is explicit. Composing never writes anything; edits on a saved page are kept only when you click Save there. The button reports Saved. when it's done.
• Two people at once: if the page changed elsewhere since you opened it, your save is rejected with a message asking you to reload rather than silently overwriting the other version.
• The deck travels with the page. Saving also stores the parsed slides and the whole deck image inventory, so on reopening you still get the image picker and AI Regenerate. Slide screenshots aren't kept, so regeneration on a saved page works from text and visual notes.
• Availability: the Save button appears only when the server has storage configured. Without it the tool simply stays stateless — compose, edit, export, print.

• Both export and print include only the rendered sections — never the sidebar, the upload controls, or the per-section hover tools and editor.
• The section styles are always inlined. Export before saving and the pictures are embedded too, so the file opens standalone with no network access; export from a saved page and its images load from their stored URLs.
• Both work the same on a composed page, on /pages/{id}, and on /shared/{id}.

• Hero — the opening banner: a title over a full-bleed photo, with an optional subtitle, button, and 3+ cards. Part of the title can carry the green accent gradient.
• Image & Bullet Mosaic — the about-us / “why us” section: short bullets interleaved with photos and award badges.
• Panel Grid — repeated large panels, each a mini-section with its own heading plus a checklist and/or a paragraph. Best when the slide is a set of labelled columns/panels that each carry their own list.
• Tile Grid — a flat field of compact plates, each one atomic idea: a short icon + caption tile, or a big stat plate.
• Marquee Strip — a moving band of awards, logos, or ratings.
• Bento — an intro beside a few icon + text benefit callouts.
• Checkerboard Plates — a description column beside a grid of mixed proof plates (stats, logos, images).
• Achievements — a titled awards/certifications grid where each card names the awarding body and adds a detail line (level, rating, year).
• Testimonial — one verbatim client quote attributed to a named person (role + company) with their portrait, plus a case-study link.
• Industry Grid — the industries or domains you serve, as photo cards each naming a domain and describing the work there.
• Level Pyramid — an ordered hierarchy of levels, tiers, or stages (a maturity model or ladder), each with a rank code and a short description.
• Comparison Columns — 2–4 named options side by side, each a column with an optional logo, a short intro, and the same attribute rows so the columns line up.
• Statistics — a grid of quantified metrics: a label plus a figure, range, or multiplier, optionally closing with a heading and CTA.
• Text & Image — a titled passage of running prose beside a single supporting picture.
• Matrix Table — a real table: category columns crossed with labelled rows, each cell holding a short checklist or one value.
• Process Timeline — a left-to-right sequence of stages, each with a phase marker over a card naming and describing that step.
• FAQ — a numbered accordion of questions and their answers, one row open at a time. Only for copy that is actually phrased as questions.
• Case Studies — named client projects, each with its tech tags, a short write-up, an optional quote, a mockup, and a case-study link. Either paged (one at a time) or stacked as full-width bands.

The planner keeps a page readable — the Hero always comes first, and with “Keep sections in slide order” off it may also reorder for flow and hoist a Marquee to just after the Hero. Your manual reordering overrides all of that.

Picking between look-alikes: a flat set of one-idea plates is a Tile Grid; per-column headings each with their own list is a Panel Grid; standalone labelled figures are Statistics, but the moment the content is laid out as a table — column headers crossed with row labels — it's a Matrix Table, whatever the cells hold. A bare scrolling row of logos is a Marquee Strip; described award cards are Achievements.

• Generate (/generate) — paste a single slide's text (or load one of the samples) to get one section, handy for iterating without a whole deck. It shows how it classified the text, how confident it was, the runner-up sections, and the token cost.
• Gallery (/gallery) — see every section type rendered from built-in sample content. Sections with more than one layout variant get ‹ › arrows to cycle through them, and Edit opens the same content form for a live preview — a fast way to feel out what a section can hold before using it on a real page.

• No AI key configured? The app still runs, but the AI steps fall back to stubs and every input renders as a Hero — a banner says so. Manual content editing still works; Regenerate doesn't.
• No storage configured? The Save button is simply absent and the tool behaves as a stateless processor — compose, edit, export, print.
• Warnings, not failures: after composing, notes about skipped slides, unplaced images, or a vision read that fell back appear above the page. The token cost of the run is shown next to the section count.
• Sidebar: collapse it to an icon-only rail with the Collapse control at its foot (desktop); your choice is remembered.