Improved the homepage launchers, descriptions, trust promises and brand details, with a stable two-line mobile layout for longer English task names.
What shipped, what comes next, and why
A verifiable product record: released capabilities are dated, while future work is labelled separately. ImgIng never presents a plan as an available feature.
Multi-layer layout, professional filters and skin-aware beauty have shipped; next comes crop, text, graphics, selections and local edits.
Shipped
Every item below is already in the codebase and public build; concepts are excluded.
Clearer launcher typography and mobile readability
Increased task titles, format descriptions, NEW / BETA badges, trust promises, expandable guidance and brand contact text, together with card height and spacing. The two-column mobile layout now allows long task titles to wrap to two lines without clipping, crowding or misalignment.
Smaller format-aware defaults by content type
Retuned recommendations for regular and smooth photos, texture, graphics and soft alpha around the official Squoosh and codec baselines. Native browser JPEG retains extra headroom, WebP alpha stays lossless at 100, common phone-size AVIF images use a more efficient effort tier, and same-format compression still keeps the source whenever re-encoding is not smaller.
Clear home hero and direct task launchers
Kept the existing two-column structure and full converter while changing the information order to task first, explanation second. Six launchers open conversion, compression, AI cutout, Animation Studio, Online Design and Video Background directly. Local processing, no-signup and no-watermark promises remain visible, with technical detail available on demand across English, Chinese and mobile layouts.
Local WebP processing and mobile layout fixes for iPhone WeChat
WebP capability checks now validate the actual MIME type so WebKit cannot silently return PNG. When Canvas WebP encoding is unavailable, a self-hosted libwebp WASM encoder loads on demand across conversion, same-format compression, background removal, Online Design and animation export without uploading pixels. Mobile text scaling, header controls, mode icons and floating launchers now have stable sizing, baselines and safe-area spacing.
Video background colour selection and instant preview fix
Fixed background changes hiding an already-matted frame and revealing the white source video. Colour, image, blur and background-mode changes now reuse the cached matte for instant recomposition without rerunning AI, while the control shows both a real swatch and its hex value.
Professional filters, skin-aware beauty and open projects
Added per-image-layer and full-canvas scopes with 12 presets and 18 light, colour, skin-aware beauty and detail controls. Edge-preserving smoothing never expands transparent contours, alpha is retained pixel for pixel, large images use overlapping memory-safe bands, and .imging stores all parameters, original assets and a filtered preview.
Video first-frame hair, aspect and blank-intro fixes
Fixed a fallback-mask payload bug and low-contrast hair loss in the experimental tier. RVM ResNet50 now calibrates the first frame before MatAnyone2 temporal memory; narrow gaps are closed in linear time without replacing the original soft alpha. Landscape, square, portrait and ultrawide video keep their aspect ratio, while blank intros output the chosen background and continue looking for a later subject.
Unified light/dark editor themes and scrollbars
Online Design, Video Editor and Professional Refinement now share the site design tokens across toolbars, panels, controls, canvases, cards and exit confirmation. Each workspace also gets an in-editor theme control and matching scrollbars.
Ultimate video model status overlap fix
Video model cards now use an adaptive two-column title and status layout. The long MatAnyone2 title and Cached for ratio state no longer overlap, while the description retains the full row width.
English home launcher overlap fix
Online Design and Video Editor launchers now share an auto-flow container instead of separate fixed vertical coordinates, preserving stable spacing for wrapped English copy and responsive layouts.
Bilingual image and video shortcut routes
Added /img, /video, /en/img and /en/video direct launchers. Home and bilingual task pages now use the clean routes while authoritative task pages remain indexable to avoid duplicate SEO / GEO content.
Video export size and bitrate fix
Fixed oversized MP4 and WebM output after background replacement. Video bitrate now adapts to the source, resolution, frame rate and alpha channel; compatible audio is copied directly and only necessary transcoding uses an adaptive audio bitrate.
Online Video Editor Beta
Opened local video background replacement with two RVM models and aspect-adaptive MatAnyone2 landscape, square and portrait variants. Models use ModelScope CDN first with site fallback; original aspect ratio and audio are retained, with MP4 or transparent WebM export.
Animated WebP ghosting fix and frame-optimization control
Fixed price and text ghosting caused by lossy animated WebP delta-frame blending. Restored the Delta frames / Full frames control and routed WebP, APNG, GIF and AVIF through their matching encoding paths.
Frame scaling and arbitrary-angle image rotation
Shipped Keynote-style eight-way scaling and rotation handles, precise inspector angles and ±90° shortcuts. Preview, undo/redo, .imging project round-tripping and final image export all preserve rotation.
Online Design and an independent multi-layer workspace
Shipped a local design workspace with up to 24 layers: multi-image import, move, scale and order, transparent or solid backgrounds, canvas presets, undo/redo and full-size local PNG, WebP or JPG export.
Local multi-background compositing after AI cutout
Professional refinement supports transparency, a solid colour or up to 12 local background layers. The subject and every background layer can move, scale, position and reorder independently, with full-resolution local compositing on export.
English site and bilingual SEO / GEO
Shipped the complete English tool and eight English task guides, reciprocal hreflang, a bilingual sitemap, structured data, English social previews and AI-readable documentation. AI model names and runtime states are localized too.
Professional AI cutout, real export and large-image performance
Unified full-resolution alpha across preview, refinement and download. Added a true 1 px brush, multi-step undo/redo, Space-to-pan, black/white edge inspection and a fixed-divider Squoosh-style comparison.
Three local AI tiers and fast colour removal
Added on-demand ISNet INT8, BEN2 FP16 and BiRefNet HR-Matting FP16 models. The fast path removes only similar colour connected to the image edge, preserving matching colour inside the subject.
End-to-end animation and isolated workspaces
Completed compression, creation and frame editing for GIF, APNG, animated WebP and animated AVIF, including 100 ms–3 s timing, loops, playback direction, drag ordering and isolated workspace state.
Capability boundaries and tool comparison
Documented live browser capability detection, device/server boundaries and a transparent comparison methodology for ImgIng, Squoosh, TinyPNG and cloud converters.
Next release: deeper local image editing
Build frequent refinement tools on the shipped multi-layer workspace, with preview/export parity as a release requirement.
Canvas and geometry
Crop, flip, free transform and additional composition guides.
Local editing
Brush, eraser, local adjustments and selections connected to AI masks.
Text and graphics
Text, basic shapes, colour, stroke, opacity and layer ordering.
Professional history
Non-destructive operation history, multi-step undo/redo and original/result comparison.
Device-side export
Keep real encoded-Blob verification and export common formats locally.
Planning note: this scope is a public target, not a currently available feature. Every item must pass image-quality, memory, browser-compatibility and large-image performance checks; incomplete work will be delayed instead of shipped prematurely.
Roadmap principles
Every new capability must pass four product rules.
Local first
If a browser can complete the work reliably, image pixels stay on-device.
Results you can trust
Preview, comparison and download share the same real-pixel pipeline.
Large images remain usable
Zooming, panning and editing must stay responsive at high resolution.
Boundaries stay explicit
Planned, device-side, on-demand WASM and server features are labelled separately.
Research pool
Directions without a committed date stay in research, separate from the next release.
Batch workflows and reusable presets
Apply one set of dimensions, quality, naming and export rules to many files.
Masks and finer selections
Extend the shipped multi-layer composition with selections, masks and more precise local control.
Colour management and professional metadata
Evaluate ICC, EXIF, HDR and cross-browser colour consistency.
More local models and plugin-style capabilities
Add only when download size, licence, speed and quality are acceptable.
Release and roadmap questions
The same clear answers for people and search engines.
Can I use the future features shown here now?
No. Multi-layer design, professional filters and skin-aware beauty are shipped, while crop, text, shapes, selections and local editing remain planned. Only the Shipped section describes current capabilities.
Will the local image editor upload images?
The shipped Online Design filters keep assets local. The goal is to keep future crop, text, graphics, brush, selections and common-format export in the browser too. Any server-only operation will be labelled before use.
Why publish target dates instead of invented version numbers?
ImgIng currently follows continuous delivery, so dates are easier to verify. Semantic versions can be introduced when the formal release process needs them.
Can the roadmap change?
Yes. Browser support, memory, large-image performance and export parity can change scope; updates will remain visible on this page.
Reviewed by the DataDance product and engineering team ·