Roadmap

Core AI pipeline

RMBG-1.4 and BiRefNet inference — no signup, no upload, no server.

Image editor

Brush erase/restore, edge refinement, feather, smooth, expand, decontaminate.

Video editor

Playback, grade, trim, overlay, and export for processed video.

Video upscaler

Single-pass ffmpeg Lanczos 2× and 4× upscaling for video clips, fully in-browser.

Video filter

Cinematic LUT presets applied frame-by-frame to video.

Video stabilizer

Browser-based camera shake reduction.

AI image upscaler

2× and 4× super-resolution. Runs in your browser.

Image filter

Cinematic LUT presets with intensity control and open-in-editor.

WASM auto-retry

When GPU inference fails, all workers automatically retry with WASM backend.

17 utility tools

Format converter, PNG optimizer, image compressor, resizer, canvas extender, rotate, grayscale, ICO creator, add background, color picker, color extractor, compare formats, metadata remover, check transparency.

Manual image adjustments

Brightness, contrast, saturation, and temperature controls in the image editor — composable with filter presets.

Enhanced edge quality

CPU-side guided filter + hair/fur soft alpha preservation + erosion — fast "Clean Edges" in ~1s without re-running AI inference.

33 image filter presets

Expanded filter library: Sepia, Teal Orange, Bleach Bypass, Cross Process, Neon Noir, Pastel Dream, Infrared, B&W Soft, B&W Hard, Fade Black, Fade White — all selectable by name in the image editor.

Color-key background removal

Automatic solid-color detection for graphic content — faster than AI, no model download required.

Smart edge decontamination

Lab ΔE post-processing pass removes background color spill from subject edges without eroding fine hair or fur detail.

Upscaler quality improvement

CPU-side unsharp mask post-pass for AI-upscaled images — sharper output without a larger model.

Editor session restore

Auto-save every 5 s + "Continue previous session?" banner on return visits. Recent files row with thumbnails in editor empty state.

First-run model caching

Pre-download AI models on first visit with a progress indicator — subsequent sessions start instantly with no wait.

Named projects

Save and reload named editing projects from IndexedDB — with edit history, canvas state, and source image preserved.

Mobile optimization

Pointer-event canvas (touch + stylus), pinch-zoom, collapsible bottom tab bar on mobile viewports.

Transparent material support

Post-pass algorithm for glass, plastic, and acrylic — Lab ΔE background detection, glass/specular preservation, partial alpha smoothing.

Lifestyle scenes in the editor

Product photo → AI removes background → 14+ lifestyle scene templates (oak tabletop, marble studio, dark slate, kitchen counter…) → 6-pass seamless blending → PNG/WebP/AVIF export. Built into the image editor background panel.

Sitewide queue overhaul

Cancel and restart work correctly across all tools — no page refresh needed. Workers are terminated immediately on cancel; next queued item auto-starts.

Image upscaler minimum resolution

2× always outputs at least 2048 px on the longest edge; 4× always outputs at least 4096 px. Tiny inputs are pre-scaled so the AI model always produces useful output.

Video stabilizer duration fix

Fixed output being 2–3× longer than input. Frames are now collected as blobs before encoding — seek overhead no longer inflates video length.

Offline model pre-download

First PWA launch shows a one-time prompt to cache AI models (~600 MB) for offline use via the Cache API.

Video utility tools

Format converter, compressor, resizer, canvas extender, rotate, metadata remover, and format comparison — in-browser video utilities. AudioContext removed from all play-through tools to eliminate freeze-on-large-files.

ORT heap-corruption fix

Reclassified WASM heap-corruption ORT errors as non-recoverable. Per-job worker isolation now works correctly — Best Quality mode retries spawn a clean WASM session instead of looping in a corrupt heap.

Stabilizer over-smooth fix

Smoothing window reduced from 5–60 frames to 3–25 frames at max strength (80% default → ~20 frames vs. old ~49). Scene-cut detection resets the trajectory accumulator at hard cuts.

Image upscaler instant-mode sharpening

Unsharp mask (strength 0.45) applied after Lanczos instant mode — same post-pass as AI mode. Per-tile 90 s health check terminates stalled workers early.

BiRefNet heap-corruption fix

Definitive WASM heap fix: session.dispose() in all reload paths including the stale-session detection block. Deterministic ONNX input key (session.inputNames[0]) replaces brute-force key loop. Best Quality mode now reliable.

Edge quality — fast mode +1px erosion + tighter bg kill

RMBG-1.4 erosion radius increased by 1px; near-white background kill threshold tightened 60→40. Less white halo fringe on light backgrounds.

Per-tile upscaler timeout

RealESRGAN tile loop now has a 30s per-tile Promise.race() guard. Stalled tiles surface as errors rather than hanging indefinitely.

Per-tool how-it-works pages

22 individual tool guides at /how-it-works/[slug] — 3-step technical walkthroughs, FAQ, related tool links, and HowTo structured data for rich results.

Landing pages upgraded

All tool and editor landing pages now include the 4-point "Why it's different" section and a contextual CTA — consistent standard across all 30+ pages.

Video utility unmount cleanup

All 6 video utilities abort their RAF draw loops when the user navigates away — no wasted CPU from background processing after page exit.

PWA icon fix

Added <link rel="apple-touch-icon"> and <link rel="icon"> to <head> — installed app on iOS, Android, and desktop now shows the correct NSS logo instead of a generic browser icon.

Before/after comparison on key pages

Interactive drag-to-reveal comparison slider on the homepage and image upscaler pages. Demo SVG image pairs at /demo/.

All-in-one landing page (/all-in-one)

New comprehensive landing surfacing all editors + tools alongside the existing BG remover landing at /. Header link added; SEO preserved on /.

Shared layer architecture

Image editor and video editor both use the same <LayersPanel> with multi-file drop, opacity, blend modes (12), reorder, lock, duplicate, keyboard shortcuts (Ctrl+J / Shift+]/[ / Delete), and project round-trip.

3D mode (image + video editors)

Three.js installed and dynamic-imported. Flat 3D plane + depth-displaced relief mesh (reuses existing depth model) + lighting/material/background controls + click-to-select gizmo + 4-second 360° orbit recorder → WebM. Available in both editors.

AI subject targeting (R-SAM)

Press S in the image editor and click any subject — colour-space flood-fill with edge cleanup isolates that region. Tolerance shared with Magic Wand slider.

Video object removal (R-VO V1)

Corner-region cleanup panel in the video editor — watermark / channel logo / timer cleanup via temporal median fill across surrounding frames. Brush-region UI coming in V2.

Project export / import (.nss-project)

Save the full editor state — original image, mask, filters, background, full layer stack — as a single ZIP. Reopen on any device with original layer ids preserved.

Collaborative sessions (R-COLL)

Start or join a session; edits stream peer-to-peer over WebRTC. Pixel data never crosses a server — only operation descriptions.

Custom tuning profiles (R-TUNE)

Upload 5–10 reference images of your product category; the analyzer derives ΔE / erosion / guided-filter overrides applied in the inference worker. No model retraining, all client-side.

Batch video processing ZIP (R-BVID)

The Video Upscale queue now shows "Download all as ZIP" when 2+ jobs are complete. Single click downloads all results.

More languages (R-LANG)

Spanish, German, Portuguese, Japanese — 75+ translated UI strings per locale via next-intl. Runtime resolver: cookie > Accept-Language > en. Footer LocaleSwitcher.

PWA icon unification (A1)

src/app/icon.svg + icon.png + apple-icon.png all unified on the purple-sparkle brand mark. Manifest URLs versioned (?v=4) to bust OS install cache. SW cache bumped v3→v4; icon paths served network-first.

Floating quick-actions toolbar

Photoshop-style draggable toolbar above the canvas (image + video editors). Undo / redo / zoom / filter dropdown / background dropdown / adjust sliders / export buttons / shortcuts. Position persists in localStorage.

Layer-aware export (EXPORT-V2)

Export panel now has Pipeline mode (clean mask + filter + background) and Canvas-snapshot mode (captures everything visible including overlay layers). Per-layer inclusion checklist with Select all / Select none.

Tools utility expansion in editor

UtilityToolsPanel in the image editor gained palette extraction, transparency check, colour picker, PNG optimizer, metadata stripper, and one-click PNG/WebP/AVIF/JPG export — every utility callable from inside the editor without leaving.

Real photography template library (R-PHOTO)

All 14 lifestyle templates ship with themed backdrops generated per-slug via scripts/fetch-lifestyle-photos.mjs (image.pollinations.ai, deterministic seeds). LifestyleComposerZone HEAD-checks each photoUrl and falls back to the SVG mockup automatically. Drop a curated studio shot of the same filename into /public/templates/lifestyle/photos/ to override any individual backdrop.

INT2 V2 — direct brush-to-layer write path

V1 ships with a "Bake strokes onto active layer" button; V2 rewrites the brush stroke pipeline so strokes write directly to the active layer's mask in real time. Plan at docs/INT2-PLAN.md.

R-VO V2 — brush-region object removal

V1 ships corner presets (watermark / logo / timer cleanup); V2 adds an interactive brush UI so users can paint arbitrary regions to remove.

Curated lifestyle photo swap

Themed backdrops ship for all 14 slugs via the fetch script; the optional follow-up is replacing any of them with licensed studio shots that exactly match the template's intended product context. Drop a JPG of the matching filename into /public/templates/lifestyle/photos/ to override.

Video utility worker migration (E3+X1)

7 single-file video utilities (compressor / format-converter / resizer / rotate / canvas-extender / metadata-remover / format-comparison) currently run on the main thread. Migration to OffscreenCanvas workers is deferred unless users start requesting multi-file batching.

AI video background replacement

Place a custom image or video behind subjects in processed video — same compositing pipeline as the image editor background panel, applied per-frame.

Real-time video effects

Apply colour grades, overlays, and blur effects to video output in real time via WebGL shader passes.

Smart guides + layer alignment

When dragging overlay layers, snap to edges of other layers / canvas centre. Visual guide lines on hover.

Group + clipping mask layers

Bundle several layers under a group with shared opacity / blend mode / lock. Adjustment-layer clipping masks for non-destructive corrections.

Mobile editor responsive layouts

Beyond the current MobileEditorBar — full responsive layouts for narrow viewports across all editor panels.