
Kirigami
: How to use it, features, and the business problems it solves
Add bookmark
What is Kirigami?
Kirigami is a web service and API, run by an individual in Japan, that converts images and PDFs into editable PowerPoint (PPTX) files. OCR restores the characters inside an image as text boxes you can rewrite, shapes are vectorised, and related elements are grouped through what the service calls Stack Mode. Multi-page PDFs can be submitted in one go and reconstructed slide by slide, and the same conversion can be called from a REST API, a CLI, MCP or Claude Skills. The interface and output are available in 17 languages. Pricing runs on credits: one image costs 10 credits, and monthly plans start at $12.00 with STARTER (as of September 2026).
Business problems it solves
About "Kirigami"
What is Kirigami
Kirigami is a web service and API, run by an individual in Japan, that converts images and PDFs into editable PowerPoint (PPTX) files. It takes slides produced by image-generating AI, documents that survive only as images, screenshots and similar material, and returns a .pptx in which the text comes back as text boxes you can rewrite and the shapes come back as vector elements.
As of September 2026 the official top page leads with "Turn images generated with GPT Image 2 into editable PowerPoint" and calls the product an "AI PowerPoint agent". The screen is less a file drop zone than an instruction box with an attachment field: you attach an image, a PDF or a note, write what you want, and the converted deck comes back. The second entry point is the API — the same conversion can be called from a REST API, a CLI, MCP, or Claude Skills. The interface and the output are both available in 17 languages, Japanese among them.
A PDF costs pages × 10
STARTER, 360 credits a month
Japanese, English, Chinese and more
After the conversion finishes
(All figures are as published on the official site in September 2026.)
A slide that arrives as an image cannot be edited at all
Slides and infographics made with generative AI often look finished, yet resist any partial change. Because they are drawn as images, clicking a heading selects the picture, not the words. Advancing a fiscal year by one, or making a company name consistent, means rebuilding the whole deck. PDFs whose source file is long gone, and scanned paper documents, sit in the same position.
A deck handed over as images
- Headings and body text are part of the picture, so they cannot be selected or corrected
- Fixing them means rebuilding in PowerPoint, or having the AI regenerate the whole deck
- Diagrams and illustrations are baked into a single image and cannot be moved
After running it through Kirigami
- Text comes out as text boxes you can rewrite
- Shapes are vectorised and can be edited individually in PowerPoint
- Related elements are grouped, so the deck can be handled layer by layer
The official documentation describes the difference as "not a screenshot — the text stays editable, and shapes, layout and background are reproduced faithfully".
A four-stage pipeline rebuilds the image as a PPTX
Conversion runs as an asynchronous job. The official API documentation publishes the four stages and what each one does. The route is the same whether you start from the web screen or from the API.
Send the request
An application, an AI agent or a script posts to the REST API over HTTP
Check it at the door
API key authentication, per-plan rate limits, validation with Zod, and creation of the job
Put it in line
Job metadata goes to Convex and large payloads to R2; QStash processes each image asynchronously
Take it apart and rebuild it
Vectorisation → OCR → text removal → semantic grouping → PPTX generation. The finished .pptx is served through a signed URL
A job moves through pending → waiting_images → processing → completed; if it fails permanently, the credits are returned automatically
The heart of it is the last stage, with its four steps. Shape outlines are first extracted as vectors, OCR reads the characters and their positions, the characters are then removed from the image side, and finally elements that belong together are bundled into one group. The official name for that bundling is Stack Mode, described as grouping related elements into a structure "with no holes left in it". Because the places where text was removed do not remain as gaps, the background is less likely to fall apart when the file is opened in PowerPoint.
How to use it
-
Create an account
Sign up from the "Start for free" button on the top page; authentication is handled by Clerk. The official help centre states that "you can try one conversion for free" and that continued use requires creating an account and purchasing a plan, so ongoing work needs either a monthly plan or a credit pack.
-
Attach an image or PDF and write your instruction
Drag an image, a PDF or a screenshot into the input box, or paste it with ⌘V. A multi-page PDF can be submitted in one go and is reconstructed slide by slide. The credit cost (10c per image) is shown beneath the input box.
-
Wait for the conversion
According to the official help, one image usually takes a few seconds to about a minute. Complex images and large files can take longer. If an error occurs during conversion, no credits are consumed.
-
Collect the PPTX and finish it in PowerPoint
Download the generated .pptx and open it in PowerPoint to edit, present and share. Because the text arrives as text boxes and the shapes as vector elements, your usual PowerPoint habits apply.
-
If you do this repeatedly, move it to the API, CLI or MCP
There is no need to run the same conversion by hand every time. For the CLI,
npm install -g @kirigami/cli, thenkirigami loginto authenticate through the browser, thenkirigami convert slide.png --output deck.pptx. MCP-capable clients such as Claude Code, Claude Desktop and Cursor connect with a single line:claude mcp add --transport http kirigami https://kirigami.ai/api/mcp.
The conversion can be handed from a person to an agent
What most distinguishes this service from other converters is that the conversion itself is published in a form other programs can call. The public API was released as v1 on 22 April 2026, and v1.1.0 on 26 April added endpoints for layer decomposition and HTML-to-PPTX, along with remote MCP.
- REST API —
POST /api/v1/convertsubmits a job that merges 1 to 50 images into a single PPTX, andGET /api/v1/jobs/:jobIdreturns its status. The specification is served as OpenAPI 3.1 from/api/openapi.json, with an interactive reference through Scalar - CLI — install
@kirigami/cliglobally and receive the key through the browser. Because the API key need not be hard-coded, CI/CD only needs an environment variable - MCP — two routes, remote HTTP (
/api/mcp) and local stdio. The remote route runs OAuth 2.0 (PKCE) automatically and exposes three tools:kirigami_decompose_image,kirigami_html_to_pptxandkirigami_get_usage. If MCP requests must not leave your network, the bundled Node.js script can be launched over stdio instead - Skills — two
.skillpackages for Claude Code and Claude Desktop.kirigamiconverts an image straight to PPTX (10 credits per image), whilekirigami-htmlhas Claude design the slides in HTML first and then turns them into PPTX (5 credits for PPTX generation, 5 for image decomposition). The two share~/.kirigami/config.json, so one login covers both
An API key is 64 characters beginning with `ip2p_` and is shown only once, when it is issued. Only a SHA-256 hash is stored on the server, and a leaked key can be revoked immediately from the dashboard. Each user may hold up to five keys.
Features
01Converting images and PDFs into PPTX
The core conversion. Every monthly plan includes all of the core features.
- OCR text restoration — characters inside the image come out as text boxes you can rewrite later
- Vectorisation — shapes are extracted as vectors and become elements you can edit individually in PowerPoint
- Layer decomposition (Stack Mode) — related elements are grouped into a structure with no holes left in it, which makes editing easier
- PDF import — a multi-page PDF is submitted in one go and reconstructed slide by slide
- Batch conversion — one job can merge 1 to 50 images into a single PPTX
- 17 languages — interface and output are both available in 17 languages, including Japanese, English and Chinese
02Calling it from external tools and agents
Routes for running the same conversion from a program, without going through the web screen.
- REST API — endpoints for upload, conversion, job status, publishing and CLI authentication, with an OpenAPI 3.1 specification
- CLI — `@kirigami/cli`, with `convert`, `status` and `publish` commands and browser-based login
- MCP — remote HTTP and local stdio, reachable from Claude Code, Claude Desktop, Cursor and others
- Skills — two of them: `kirigami` (image to PPTX) and `kirigami-html` (HTML design to PPTX)
- Rate limits — 10 conversions per minute, 20 uploads per minute and so on; exceeding them returns HTTP 429 with a Retry-After header
03Around the conversion
Features that support what happens before and after.
- HTML/CSS to PPTX — `POST /api/v2/html-to-pptx` turns slides built in HTML into an editable PPTX (5 credits)
- Layer decomposition API — `POST /api/v2/decompose` splits an image into layers and returns them through a signed URL (5 credits per image)
- Credit refunds — if a job fails permanently, the credits it consumed are returned automatically
- Marketplace — a place to find AI-generated slides and templates for free, reachable through the CLI and API as well. The creator monetisation programme was still in preparation as of September 2026
Supported input and output
Pricing
Pricing runs on credits: one conversion (one image) costs 10 credits. A PDF costs its page count × 10, so a ten-page PDF consumes 100 credits. Uploading, polling a job and browsing the marketplace cost nothing.
| Plan | Monthly (USD) | Credits per month | Conversions | What it includes |
|---|---|---|---|---|
| STARTER | $12.00 | 360 | Up to 36 a month | Text editing, batch PDF conversion, shared across Web / API / Claude Code |
| PRO 20 | $20.00 | 600 | Up to 60 a month | The above plus priority support |
| PRO 40 | $40.00 | 1,200 | Up to 120 a month | Top-priority support and API access |
If you would rather not subscribe, or you have used up the month's credits, you can buy an extra credit pack.
| Credit pack | Price (USD) | Credits | Conversions |
|---|---|---|---|
| 10-conversion pack | $4.00 | 100 | 10 |
| 30-conversion pack | $10.00 | 300 | 30 |
| 100-conversion pack | $33.00 | 1,000 | 100 |
Prices and billing are in US dollars. The official pricing page notes that Japanese cardholders are billed in yen at the card issuer's exchange rate. The Japanese commercial disclosure page (updated 10 December 2025) states that all prices are tax-inclusive.
- Free usage — there is no free plan on the pricing page. The top page does carry a "Start for free" button, and the help centre says "you can try one conversion for free". The number of credits granted at sign-up is not published (as of September 2026)
- Payment — credit card through Stripe (Visa, Mastercard, American Express, JCB and others). Subscriptions are charged automatically on the renewal date; credit packs are charged immediately at purchase
- Cancellation — cancel any time from "Subscription management" in the dashboard; service continues until the end of the current billing period
- Refunds — as a digital service, refunds are not offered as a rule. Serious defects, or a payment error that leaves credits unallocated, can be raised through the refund request form in the dashboard
How uploaded images are handled
Before putting internal documents or proposals through an outside service, it is worth checking retention and third-party processors first. The official privacy policy (updated 16 March 2026) and security policy (updated 10 December 2025) set this out.
If you are adopting it for a company, start from the fact that it is run by a sole proprietor and settle early whether you need invoice billing or a separate confidentiality agreement. Payment is by Stripe credit card only; no invoice billing is advertised.
How it differs from other converters
There are several ways to make an image-based deck editable again: convert it with Kirigami, convert it with a different service, fix it with a feature of whatever created it, or rebuild it with a slide-generating AI. Which fits depends on what you want to preserve, and on whether you want to automate the conversion.
| Aspect | Kirigami | Swift-Slide | NotebookLM's own export | Rebuilding with GAMMA or Canva |
|---|---|---|---|---|
| Input | Images (JPG, PNG, GIF) and PDF | PDF, PNG, JPEG, WebP | — | Generated afresh from text or your own material |
| Text | Read by OCR and output as text boxes you can rewrite | Read by OCR; the background is restored and the text placed back where it was | Even a PPTX export puts each slide in as an image | Generated as editable slides from the start |
| Shapes | Vectorised, with related elements grouped (Stack Mode) | Diagrams and photos stay as images; object separation can cut them out | Left as images | Built anew along a theme or template |
| Calling it from outside | REST API, CLI, MCP, Claude Skills | Not stated on the official site | — | Depends on the service |
| Free trial | No free plan; the help centre mentions one free conversion | Free plan at ¥0: 5 pages a month plus a 5-page sign-up bonus | Within your NotebookLM usage | Depends on each service's free plan |
| Price | From $12.00 a month (tax included, billed in USD) | From ¥980 a month (tax included) | — | Depends on the service |
If you want the conversion called from an AI agent or an internal script rather than run by hand each time, Kirigami — with its API, CLI and MCP all in place — is the option to look at. If instead you want to adjust wording and positions carefully in the browser before exporting, a service with a built-in editor such as Swift-Slide takes fewer steps. If what you really want is editable slides from the outset, GAMMA and Canva are built for that, and for turning prose into diagrams there is Napkin.
Frequently asked questions
The information on this page is based on what was published on the official site as of September 2026. Please check the official site for the latest specifications and pricing.

![How to Edit NotebookLM Slides [September 2026]: Why You Can't Fix the Text Even in PPTX, and Five Methods by What You Need to Change How to Edit NotebookLM Slides [September 2026]: Why You Can't Fix the Text Even in PPTX, and Five Methods by What You Need to Change](/_next/image?url=%2Fimages%2Farticles%2Fnotebooklm-slide-edit.png&w=3840&q=75)