Back Home
Product Updates

What changed recently in Image2PPT

The latest improvements to conversion quality and everyday use.

Start converting
2026-11-09

One conversion engine for every slide

Every slide is now rebuilt by a single engine, so there is no conversion mode to pick when you upload. People and artwork on illustrated and poster-style slides are cut out whole along their outlines, the area behind them is filled in, and everything on the original appears exactly once.

  • More frames, colour blocks and labels become editable shapes, with straight edges drawn as lines and rounded corners as arcs, so you can change their colour and outline directly
  • Downloaded PPTX files are about a third smaller, so they are quicker to send and open
  • Printed Kai-style Chinese text stays editable with one consistent typeface per slide, and table text follows the original font
2026-10-08

Every element stays whole and clean when you move it

Cards, icons, photos and connectors on business slides are each delivered as one clean piece. Move any of them and it brings no fragments of its neighbours and leaves no ghost behind, and everything visible in the original appears exactly once.

  • Teardrop and other irregular cards and frames become editable shapes, so you can change their colour and outline directly
  • Grids of separate cards come out as cards, and text in the same table column or header row keeps one size and weight
  • Large headlines on illustrated slides no longer stay baked into the background, so you can move or edit them on their own
2026-10-04

More editable text, sharper small icons

More of the text in your converted deck is now real, editable text, including chart axis labels, English handwriting and smaller decorative lettering, and icons, badges and logos stay crisp when enlarged.

  • Each paragraph keeps one font size, so units next to numbers no longer blow up to the size of the numbers
  • Large brush lettering and designed headlines keep their exact look and can be moved on their own
  • While converting, each slide shows eight steps, time elapsed and an estimate of time remaining
2026-10-01

API: completion webhooks and submit by URL

Integrating the API takes less glue code: we notify your server when a job finishes, and files already online can be submitted by link instead of being downloaded and re-uploaded first.

  • Convert only selected PDF pages and pay for those pages alone, which also lets you pick pages from PDFs longer than 50 pages
  • Retrying the same submission after a network drop creates one job and reserves credits once
  • A new job list shows credits used and refunded per job by date, for reconciliation
2026-09-30

Everything on the slide comes through, once

Each slide is now fully separated into its parts before anything is replaced: what matches the original becomes an editable line, arrow, or shape, and everything else keeps its original look, so nothing goes missing and nothing appears twice.

  • Large brush and display lettering on posters and illustrated slides is cut out along its strokes as a transparent image, keeping brush texture and outlines
  • Japanese slides use Japanese fonts, so kanji display in their Japanese forms
  • In your conversion history you can rename a job or permanently delete it along with its files
2026-09-26

Next-generation conversion model

Conversions now run on a next-generation recognition model that reads text, shapes, and layout more accurately, with more consistent results when the same slide is converted again.

  • Calligraphy and display lettering on posters and lesson slides is matched to fonts by glyph shape, and Song, FangSong, italic, and vertical text keep their original look
  • Subscripts and superscripts in formulas and chemical equations convert as editable text, without letters and digits getting mixed up
  • Gradient banners, bold arrows, small icons, and numbered circles are preserved more faithfully
2026-09-21

More slide content converts as editable

Arrows, tables, and chart labels on business slides now convert into elements you can edit directly.

  • Arrows and connectors become editable lines you can recolor and reshape
  • Detected tables convert as editable tables instead of flat screenshots
  • Chart ticks, legends, and labels stay editable where possible
2026-09-15

Improved business slide conversion

Business slides with complex layouts now convert more completely and cleanly.

  • Panels with cut corners or angled edges become editable shapes
  • Overlapping images merge into one, without ghosting
  • Text on dark backgrounds stays readable
2026-09-08

Shapes, lines and photos reproduced more faithfully

Unfilled shapes, arrows and thin lines are rebuilt from their real outline, and photos and illustrations embedded in the PPT keep their full colour.

  • Unfilled circles, diamonds, triangles and trapezoid tiers follow their real outline
  • Arrows and closely spaced thin lines each come out intact
  • Photos and illustrations keep smooth, full colour
2026-09-04

Dense slides convert more completely

Dense pages that mix tables, arrows and icons now convert more completely and closer to the original.

  • Arrows are not doubled, and small arrows are kept
  • White text on dark cards stays visible
  • Auto-cropping keeps titles, logos and page numbers
2026-08-30

The full account menu on mobile

On phones and narrow windows, whatever does not fit in the top bar folds into one menu, so history, purchases and sign-out are always reachable.

  • Mobile shows your full sign-in email
  • Pricing and blog stay available as the window narrows
2026-08-26

Sign-in no longer gets stuck on "sending"

Some people were getting stuck on "sending" at sign-in and never received a code. That is fixed — if it kept you out, just try again; there is no need to switch browsers or clear your cache.

  • Sign-in getting stuck on "sending" without ever delivering a code is fixed — just try again.
  • Two verification boxes occasionally stacking on top of each other is fixed too.
2026-08-23

Local files and PDFs work directly in Codex

Using Image2PPT inside Codex used to mean handing it a public image URL. Local files and PDFs can now go straight in, with PDFs counted by their real page count. When an upload is too large, you are told to split it into batches, and the web and bulk uploads check the size locally before sending anything. The language switch has changed too: the button shows the language you are using and opens a menu, and pages like the blog and the technology page have a language entry for the first time.

  • Hand Codex a local file or PDF directly, with no need to host it somewhere first.
  • An oversized upload now says so and tells you to split it — being under the page limit does not mean being under the size limit.
  • The language switch opens a menu, and the blog and technology pages gained a language entry.
2026-08-12

Convert slide images to PPT inside Codex

Install Image2PPT, drop a slide image into a Codex conversation and say "convert to PPT": the conversion runs there, progress and all, and hands you a download link. It uses the credits in your own account, and the website and the conversation show the same job history. Signing in always happens on image2ppt.com, so Codex never sees your email code or your session on the site.

  • Say "convert to PPT" in a Codex conversation and get a download link back, billed to your own account.
  • Signing in always happens on image2ppt.com; Codex never sees your email code or your session.
  • To disconnect it, use "sign out of all devices" on the website.
2026-07-25

Signing in got easier: Google sign-in

You can now sign in with Google in one click, with no waiting for a code by email (Gmail and Workspace accounts). Either method works at any time, and switching between them does not create a second account — your credits and history stay where they were. Your account also gained a "signed-in devices" list, so you can see what is currently signed in and sign out anything you do not recognise.

  • Sign in with Google in one click, with no waiting for an emailed code.
  • Either sign-in method works at any time without creating a second account; credits and history are unaffected.
  • A new "signed-in devices" list lets you sign out anything you do not recognise.
2026-07-16

High fidelity, another step further: sharper details, steadier conversions

Version 6.0 pushes high fidelity another step further: shape borders and positions are now calibrated against the original image's actual pixels, text colors are sampled directly from the source, and arrows and connectors are rebuilt as draggable, editable native elements wherever possible. Background cleanup on dense pages is much cleaner, with no more gray ghosting. Stability got a major boost too: a bug causing widespread timeouts on multi-page jobs is fixed, auto mode classifies business slides more accurately, and conversions in progress survive service upgrades — they resume automatically once the service is back.

  • Finer details, sharper again: shape boundaries calibrated against the original's actual pixels, and text colors sampled straight from the source.
  • Arrows and connectors are rebuilt as editable native elements wherever possible — easy to move and recolor.
  • Cleaner background removal on dense pages — no more gray clouds or ghosting.
  • Fixed widespread timeouts on multi-page jobs, and auto mode now classifies clean business slides more accurately.
  • Service upgrades no longer interrupt your work: in-progress conversions resume automatically once the service is back.
2026-07-05

A major conversion-engine upgrade, built for high fidelity

We've given the underlying conversion engine a major upgrade with one goal: make the PPT you get back as close to the original as possible. It now automatically picks the best reconstruction approach for each page, rebuilding text, shapes, tables, and icons as editable elements — with position, color, and layout all pushed toward being nearly indistinguishable from the original.

  • A major upgrade to the underlying conversion engine, built for high fidelity to the original.
  • Automatically picks the best reconstruction approach per page, rebuilding text, shapes, tables, and icons as editable elements.
2026-06-11

Downloads now use your browser's native downloader, with resume and real progress

Previously, clicking download silently fetched the whole file inside the page before showing a save dialog — on slow networks nothing seemed to happen, and a dropped connection meant starting over. Downloads are now handed to your browser's native downloader: the download bar appears immediately with real progress and speed, and interrupted downloads can resume from where they left off. Batch zip downloads also show an accurate percentage instead of an unknown total.

  • The download bar appears immediately with real progress and speed.
  • Interrupted or paused downloads resume from where they stopped.
  • Batch zip downloads show an accurate progress percentage.
2026-06-08

Your conversion history and downloads are now easier to find

Finding your converted files and re-downloading results used to mean clicking a small, easy-to-miss button in the bottom-right corner — many people never noticed it. Now, once you're signed in, there's a "History" entry right in the top bar on desktop, and at the top of the account menu on mobile, so you can get back to past results anytime. The bottom-right button now stays visible while signed in and spins to show progress when a conversion is running.

  • A "History" entry now sits in the desktop top bar and in the mobile account menu, making past conversions easy to reach.
  • When a conversion is in progress, the entry shows a live progress indicator.
2026-06-08

Text-heavy cards, lists, and tables stay more editable after conversion

Text-driven regions on a slide — callout and KPI cards, numbered or bulleted lists, data tables, numbers inside circular badges — were sometimes extracted as a single flat image, leaving the text inside uneditable. We've tightened the recognition rules so these regions are reconstructed as editable text plus shapes or tables wherever possible, letting you edit the wording directly instead of recreating it from a picture.

  • Callout cards, lists, tables, and other text-driven regions are kept as editable text rather than images more often.
  • Numbers and letters inside circular badges are rebuilt as a shape plus editable text.
2026-06-04

Cleaner backgrounds where text is removed from images

When an image on your slide has text on top of it, the conversion removes that text and restores the underlying picture so you can edit freely. This step used to sample-and-blur, which sometimes left a smudge like correction fluid; it now uses smarter image inpainting that fills the gap from the surrounding picture, so backgrounds come out cleaner and closer to the original. If inpainting ever fails it automatically falls back to the old method, so your conversion is never affected.

  • Text removed from image backgrounds leaves fewer, more natural traces.
  • If inpainting hits a problem it falls back automatically; the conversion still completes.
2026-06-04

Fixed a case where a finished job stayed stuck and showed no result file

In rare cases your conversion had actually finished and produced a file, yet the job stayed stuck as processing with no downloadable result. This is now fixed at the root: when the service restarts in that state it recognizes these already-finished jobs and restores their result instead of leaving them stuck, and several concurrency race conditions that could mis-rewrite job status are now closed. Affected jobs recover automatically with no action needed.

  • Jobs that already converted successfully no longer get stuck or lose their result file.
  • On restart, the completed status of such jobs is restored automatically — no re-conversion needed.
2026-06-01

Fixed conversion failures on text-heavy slides

Slides packed with text — long paragraphs, dense tables, full pages of business or legal boilerplate — could previously fail to convert and refund your credits. This is now fixed on two fronts: the model is given its full output space so long content is no longer cut off midway, and when a page of text is refused by the model's copyright filter, the content is automatically rebuilt via text recognition. These slides now convert normally instead of wasting an attempt.

  • Long-content slides no longer fail from truncated output.
  • When a full page of text is refused, content is rebuilt automatically via text recognition.
2026-05-29

Fixed a rare case where credits stayed locked and conversions reported insufficient balance

If the service restarted while one of your conversions was still running, the credits that job had reserved could fail to unlock — leaving you with a balance that still reported "insufficient credits" on every later conversion. This is now fixed at the root: on restart, credits held by interrupted jobs are returned directly instead of relying on a callback that wasn't wired up yet at that moment. Affected accounts recover automatically, with nothing for you to do.

  • Restarts no longer freeze credits held by in-flight jobs; credits reserved by interrupted jobs are returned automatically.
  • Accounts that were already stuck recover on their own — no need to repurchase or contact support.
2026-05-27

Task panel thumbnails no longer flicker

When converting multi-page jobs, the thumbnails in the task panel briefly disappeared on every progress update — the whole row looked like it was flashing white. Image access URLs are now cached in the browser, so cache hits render instantly and skip the few hundred ms it took to mint a fresh token. Multi-page jobs and batch uploads no longer flicker; thumbnails stay put while progress updates.

  • Thumbnails stop disappearing and re-rendering as conversion progress updates — the panel reads much steadier.
  • Live connection drops auto-refresh the access URL, so thumbnails don't end up broken after long-running sessions.
2026-05-27

Card checkout reopens Apple Pay / PayPal and surfaces Alipay for overseas users

The card-checkout path had been temporarily locked down to credit / debit only while we debugged a Paddle error, with Apple Pay and PayPal both disabled. The English UI also hid the Alipay tab entirely, so overseas users with a Chinese card couldn't pick Alipay even if they wanted to. Both are restored: the card tab now offers Apple Pay and PayPal (and shows an "Apple Pay · PayPal" subtitle), and the English UI surfaces the Alipay tab so overseas users can pick Alipay checkout (settled in CNY).

  • The card tab now carries an "Apple Pay · PayPal" subtitle so the alternative methods are discoverable.
  • The English UI now shows the Alipay tab, letting overseas users pick Alipay checkout.
2026-05-25

Done card redesigned — icon-anchored file rows with real file sizes

The conversion-done card used to cram multiple files together with truncated names and no icons, making it hard to tell how big each output was. The card has been redesigned: one file per row, anchored to a leading icon, full filenames, and a real file size pulled from disk (no more estimates). When you convert a batch or upload multiple images, every output's size is now visible at a glance.

  • Each output gets its own row with the full filename and a leading icon for stable left alignment.
  • File sizes come straight from disk, not from upload-time estimates.
  • Multi-file and batch conversions render with a consistent look so you can compare sizes at a glance.
2026-05-25

Aspect-ratio picker buttons are no longer cut off

On some screen heights, the aspect-ratio picker modal could grow too tall and the bottom action buttons (Start / Cancel) would get clipped by the viewport, forcing you to scroll inside the modal to reach them. The modal body is now independently scrollable and the action buttons are pinned to the bottom — they stay visible and clickable no matter how tall the screen or how long the content.

  • The Start / Cancel buttons stay pinned to the modal's bottom edge at any viewport height.
  • When the modal content exceeds the viewport, only the body scrolls — no more whole-page scroll jank.
2026-05-25

PPTX rendering fixes: element overlap regression + merged table headers

Some slides occasionally exported with two adjacent elements drifting into each other and overlapping their text; tables whose source image had merged header cells (e.g. a Q1 2024 label spanning three columns) used to lose the merge and split into separate cells. Both are fixed in this release: the layout geometry regression is restored to correct alignment, and merged header cells survive the round trip so they stay merged when opened in PowerPoint.

  • Adjacent elements in exported PPTX no longer overlap each other — text and cards return to their designed positions.
  • Header cells that span multiple columns in the source image remain merged in the exported table instead of splitting into separate side-by-side cells.
2026-05-25

Text and icon edge-cleanup fixes on busy backgrounds

The recently shipped preserved-background and text-mask pipelines still left visible flat blocks or icon halos in a few edge cases. Three of them are fixed in this release: text masking on photo backgrounds now does a fill-then-blur pass rather than a hard erase; icon regions sitting on high-variance backgrounds go back to the blur path instead of revealing a flat color patch; and when a foreground icon overlays a large background image, the background is also patched under the icon so dragging the foreground no longer reveals a duplicate icon outline underneath.

  • Text regions on photo or busy-illustration backgrounds clean up more naturally — no more visible flat-color blocks that don't match the source.
  • Icon regions on high-variance backgrounds use the blur path so they no longer leave a flat-color rectangle behind.
  • When small foreground icons or images sit on top of a large background image, the background is patched under them — moving the foreground no longer exposes a duplicate icon outline underneath.
2026-05-24

Slides with a large unified visual background are no longer fragmented

Slides with a continuous decorative background — nebulas, photographs, illustrations, or 4+ stop irregular gradients — used to be shattered into dozens of rectangles and gradient shapes, leaving seams, white halos, and decorative debris. This release detects those regions and lays down the original pixels as a single bottom layer of the slide; text and real foreground icons are still extracted and rendered on top, so they remain editable. Ordinary business slides without a large unified background are unaffected.

  • Whole-slide or partial regions with photographic, illustrated, or complex-gradient backgrounds keep the original pixels instead of being rebuilt from shape mosaics.
  • Titles, subtitles, bullets, and icons stay extracted as editable text and icons — the new pixel background underneath doesn't fold the foreground into a single flat image.
  • Detection is automatic in the preprocessing turn and aims at roughly a third of incoming slides. Ordinary business decks (charts, tables, flow diagrams, flat backgrounds) skip the mechanism entirely and behave exactly as before.
2026-05-20

Dark, glass, and neon panels no longer turn into white blocks

Slides containing dark cards, glass-style panels, or cards on neon / gradient backgrounds used to occasionally render as solid white blocks that obscured the title or numbers inside, breaking the overall look. This release samples the actual colors from the source image first: when the panel's source pixels are clearly dark, the white default is overridden with the sampled dark color; the white default only kicks in when the source really is light. A text-contrast safeguard is added on top so a flipped fill never produces unreadable dark-on-dark text. Visual fidelity on busy backgrounds is significantly better.

  • Dashboards, KPI cards, and chart panels with dark or glass styling keep their original colors instead of getting flipped to plain white.
  • Cards on neon or gradient backgrounds match the source colors — no more white slabs sitting awkwardly inside a bright background.
  • When a panel's fill is corrected to a darker color, text contrast is re-checked automatically so you never end up with dark text on a dark background.
2026-05-19

Outline-only frames no longer paint phantom panels

Slides containing outline-only rounded frames (the kind used to group a few labels or pills with just a border) used to render as an opaque black block on dark slides and an opaque white block on light slides, occluding the text and labels inside. This release corrects the rendering: the interior of outline-only frames is now fully transparent as intended, the stroke keeps its original color, and whatever sits behind the frame (background image, pills, text) is visible again.

  • Dashed or solid outline-only frames on dark slides no longer get filled with a black panel.
  • Outline-only frames on light slides no longer get filled with a white panel — text and labels inside are visible again.
2026-05-18

Cleaner download filenames + cross-platform fixes

Downloaded PPTX and ZIP files used to inherit junk from the original upload — random prefixes, URL-encoded symbols, zero-width characters — and Chinese filenames occasionally degraded to `download` in older browsers or strict proxies; Windows would sometimes refuse to save files whose names matched reserved variants (COM1, NUL, including fullwidth or superscript forms). This release rebuilds the entire download filename pipeline: upload-side junk is cleaned by a unified rule set, Chinese / Unicode names are encoded strictly per RFC 5987, and Windows reserved-name variants fall back to a safe name. Modern browsers on Windows, macOS, and Linux now save the original filename cleanly; in pathological cases the filename falls back to `presentation-<short-id>.pptx`, so a download never fails to save.

  • Chinese and Unicode filenames save correctly across Chrome, Firefox, Safari, and Edge — Windows no longer fails to save due to encoding issues.
  • Random prefixes, URL-encoded characters, and zero-width characters left over from upload are stripped on download.
  • Windows reserved-name variants (COM1, NUL, and their fullwidth or superscript forms) fall back automatically to `presentation-<short-id>.pptx`, so even pathological filenames save successfully.
2026-05-18

Icon crop precision significantly improved

Icon boxes used to be loose, sometimes including stray horizontal text, decorative lines, or neighboring labels — exported icons looked fuzzy or unclean. This update reworks both icon detection and cropping: icon boxes now align to the actual pixel edges; stray nearby text, labels, and decorative lines are automatically trimmed out; circular icons are shape-validated before being rendered as true circles, falling back to rectangular crops when needed so critical content is never cut off.

  • Icon boxes align to actual pixel edges — no leftover background around the crop.
  • Stray horizontal text, adjacent labels, and decorative lines that crept into an icon's bounding box are automatically trimmed out.
  • Circular logos, avatars, and badges are shape-checked before rendering — passing shapes render as true circles, failing shapes fall back to rectangular crops so nothing critical is cut off.
2026-05-17

Ghost text on complex backgrounds is gone

Slides with messy backgrounds used to produce a ghosting problem: when large gradients, dashboards, maps, or glow-decorated cards sat behind titles or numbers, PowerPoint would show every line doubled — one copy baked into the image, another editable on top, impossible to clean up. This update eliminates the ghosting entirely. The slide looks identical to the original, but every word is directly editable.

  • Layouts with text over busy backgrounds — dashboards, KPI cards, maps, colored banners, glow-decorated visuals — no longer show any ghosting.
  • Titles, numbers, and labels are all editable text — change the wording, color, or size directly in PowerPoint.
  • Pure decorative background text (when no editable text takes over) is kept as-is in the image, never erased by mistake.
2026-05-17

Task cards use source filenames, list is grouped, downloaded state shown

Task card titles switched from short hashes (`#a1b2c3d4`) to the source filename you uploaded (e.g. `quarterly-report`), and the downloaded `.pptx` filename matches; the task list is split into 'In progress' and 'Recently finished' groups with per-group counts; previously downloaded tasks show an outline-gray button with a ✓ check, clearly distinct from the solid-green button for tasks you haven't downloaded yet.

  • Task cards use the uploaded filename; multi-file uploads show the primary name plus a count (e.g. `quarterly-report +2`); the download filename matches and works across Windows / macOS plus Chinese / Unicode characters.
  • Tasks are grouped into 'In progress' and 'Recently finished' with per-group counts in the header.
  • Previously downloaded tasks show an outline-gray button with ✓ 'Download again', clearly distinct from the solid-green button on tasks you haven't downloaded; this state is tracked per device.
2026-05-17

Circular icons, background images, and PPTX size all cleaned up

A focused round of visual fidelity fixes against business deck samples: circular logos and avatars no longer leave ghost corners from the rectangular crop; when you drag a foreground icon aside in the exported PPT, the larger background image no longer reveals a duplicate copy underneath; CLI and PDF paths now produce the same compact output as the web — a 5-image sample dropped from 28.8MB to 2.0MB.

  • Circular logos, avatars, and circular badges render with a true circular mask — no leftover corners.
  • When a foreground image covers part of a larger background image, the matching area in the background is automatically blurred — moving the foreground aside no longer reveals a duplicate.
  • CLI and PDF conversion paths now produce significantly smaller output, matching the web (test sample: 28.8MB → 2.0MB).
  • Occasional Gemini hard timeouts now retry once before giving up — fewer rare failure pages.
2026-05-16

PDFs up to 50 pages, exported PPTX files significantly smaller

Looser upload limits: PDFs can now have up to 50 pages (was 30), and the per-file size cap is up to 35MB (was 30MB). Server-side PDF page rendering has also been switched to JPEG, so the exported PPTX is much smaller — one test deck dropped from 19.62MB to 2.42MB, making downloads, sharing, and email attachments easier.

  • PDF page limit raised from 30 to 50.
  • Per-file size limit raised from 30MB to 35MB.
  • PPTX files generated from PDFs are noticeably smaller — easier to download, share, and email.
2026-05-14

Recognition quality much improved: card bodies stop disappearing, colored numbers keep their color

A focused fix pass on a batch of business-slide samples: in multi-card layouts every card's body text and KPI numbers come through (no more silently dropped cards); alert / callout bars render their full background fill instead of just an accent stripe; heading / body / footnote font sizes stay distinct instead of being flattened to one size; in mixed number+label lines only the number takes the accent color and the label stays dark, instead of the whole line getting recolored.

  • Card grids and KPI dashboards extract more completely.
  • Alert and callout bars keep their full background, so readability matches the source.
  • Heading / subheading / body font hierarchy is preserved instead of being flattened.
2026-05-13

Slide dimensions follow each image's true ratio; mixed-ratio batches no longer get squashed

Previously the backend rounded each image's ratio to a standard bucket (16:9 / 4:3 / etc.), so a 1:1 image dropped into a 16:9 slide turned every round icon into an ellipse. Slide dimensions now follow the image's TRUE ratio, so circles stay circles and rectangles stay rectangles. Mixed-ratio batches are supported too: one image's ratio sets the deck size and the rest are letterboxed with proportional font/line scaling so nothing spills outside the image area.

  • Round icons, badges, and avatars stay perfectly round; rounded rectangles keep their original aspect.
  • When uploaded images differ in aspect ratio, a modal explains which image's ratio will set the deck and offers a fixed-ratio override.
  • The font-size floor follows the letterbox scale, so panoramic images (10:1) placed in a 16:9 slide keep all labels inside the image area.
  • PDF + image mixed batches now show a proper "≈16:9" hint in the auto-ratio dropdown instead of a broken label.
2026-05-12

Photographed slides now crop to the slide region

Photos of meeting-room screens with surrounding ceiling, floor, desks, or audience are now automatically cropped to just the slide region before extraction, instead of feeding the ambient background to the model. Sideways photos that need rotation are oriented correctly before cropping.

  • Head-on photos with ambient surroundings are cropped too — perspective distortion is no longer required to trigger cropping.
  • Clean screenshots are not mis-cropped — multiple guards combine to decide when to crop.
  • Photos that need both rotation and cropping are handled in the correct order, so the crop targets the right region.
2026-05-11

Chinese font recognition is more accurate

Chinese fonts are now recognized between common PPT families (KaiTi, SimSun, SimHei, FangSong, DengXian, YaHei) instead of collapsing to a single default, so the converted slides look closer to the original.

  • Headings and body text are classified independently by visual cues — titles keep SimSun/KaiTi while body text uses SimHei/YaHei.
  • Server-side rendering falls back to safe fonts instead of tofu boxes when a glyph is missing.
2026-05-11

Task progress and thumbnails more reliable

The conversion page now refreshes task status in real time without manual reloads. The task list only renders thumbnails that actually exist, so broken-image placeholders are gone.

  • Live status updates while converting — no F5 required.
  • The task list no longer requests thumbnails that were never generated.
2026-05-10

Paste images or web snippets to upload

You can paste images directly from the clipboard, or paste a web page snippet that contains image tags. Lazy-loaded HTML now resolves to the real image URLs more reliably.

  • Paste images from the system clipboard.
  • Paste HTML snippets with <img> tags; lazy-load attributes like data-src and srcset are picked up automatically.
2026-05-09

Multi-image upload and credit hints are clearer

The home page now clearly says you can upload multiple images or PDFs at once. After adding files, you can keep appending more without clearing the batch.

  • The pending-file list now has a visible "+ Add more" action.
  • When credits are insufficient, the page shows the shortfall before conversion and offers a top-up path.
2026-05-07

Complex pages and table-heavy slides are more stable

The latest quality fixes focus on tables, graphic labels, and timeout recovery for complex slides, reducing blank tables, duplicate text, and failed pages.

  • When a table has too little structure to rebuild safely, the original visual region is preserved more often instead of outputting an empty table.
  • Complex pages that hit a timeout retry with a steadier recognition setting.
  • Connector lines, numeric callouts, and compact labels received rendering fixes.
2026-04-28

Output PPT size can be selected manually

During upload, you can choose 16:9, 4:3, A4 landscape/portrait, and other sizes, or keep automatic detection from the first image.

  • Useful when converting PDF pages, portrait screenshots, or A4 materials into a specific PPT size.
2026-04-10

Multi-page jobs now support partial success

When one page fails inside a PDF or image batch, it no longer fails the whole file. Successful pages still export, and failed pages become placeholder slides.

  • Credits are charged only for successful pages; failed pages are refunded.
  • After payment, a confirmation modal now shows payment and credit status instead of leaving the user guessing.
2026-03-23

Home examples and download feedback improved

The home page now includes before/after examples, and downloads use the browser's native download behavior for clearer feedback.

  • Examples can be opened as original and converted previews, with PPTX files available for comparison.
  • Task progress is more detailed than before, so long recognition steps no longer look like a static progress bar.