v0.3.0

A ground-up overhaul of the desktop app: a new panel-based workspace, a
reimagined gallery and image viewer, a persistent activity dock, a full design
system, and a long list of reliability fixes. Also bumps the bundled
imference-engine to v0.3.2.

Upgrading: no action needed — the app installs/updates the local engine to
v0.3.2 on next launch if you use local generation.

✨ Highlights

• New workspace — the main window is now a set of drag-reorderable, stackable
  panels (Create / Gallery) you can arrange side-by-side or stacked, and resize.
  Your layout persists across launches.
• Activity dock — in-flight runs now live in a persistent bottom-right widget
  showing live cloud/local counts, with a pulse when a run starts, finishes, or
  fails. Click it for the full activity list. No more hunting for a panel that
  scrolled off-screen.
• Reimagined gallery — right-click actions, multi-select, prompt search, and
  drag an image straight onto the Create panel to use it as an img2img source.
• Better image viewer — click-to-zoom magnifier, keyboard navigation, and a
  true fullscreen mode (press F) with a solid backdrop and no side panel.
• Command palette — press ⌘K / Ctrl+K to jump to any action.

🚀 New features

• Drag-reorderable, 2D-stackable, resizable panels with animated transitions.
• Persistent Activity dock + overlay (replaces the old Activity panel).
• Gallery: context menu, multi-select with a floating action bar, prompt search,
  drag-out to the OS, and drag-to-img2img.
• Image viewer: click-to-zoom, arrow-key navigation, F for fullscreen.
• Command palette (⌘K) and a global generate shortcut (⌘/Ctrl+Enter).
• Guided first-run onboarding in the Create panel.
• Manual light/dark theme toggle.
• Local generations are queued through a serial FIFO so requests never collide.
• Cancel button for in-progress local model downloads.

💅 Design & UX

• A shared design system: every native select/checkbox/dialog replaced with
  themed primitives (segmented controls, checkboxes, skeletons, progress, toasts).
• Create panel reworked: mode-tinted surface, cleaner model bar, format moved
  under the prompt (with custom dimensions in local mode), parameters below.
• Settings redesigned as an icon-led paned dialog.
• Payment bar on the main UI shows credits (API key) or USD (x402); switching
  happens in Settings.
• App-wide toasts + a restyled Logs panel for consistent feedback.
• Brand palette re-derived from the logo; loading skeletons and empty states
  throughout; motion polish with full prefers-reduced-motion coverage.
• Window opens at a size that lands in the two-column layout by default.

🐛 Fixes

• Drag-and-drop reliability — in-app drags (panel reorder, gallery→img2img)
  were rewritten on pointer events. WebKit/WKWebView doesn't fire HTML5
  drop/dragend reliably for in-page drags, which left panel drop-outlines
  stuck on and silently broke img2img; both now work everywhere.
• Cloud downloads on slow connections — the finished image download now
  retries transient failures with backoff instead of losing an image that was
  generated fine.
• Local model download — can now be cancelled; the partial file is cleaned up
  and the previous engine restored.
• Cloud model picker — no longer blocked while a local model is downloading.
• Gallery delete no longer leaves stale tiles; format selector overflow fixed;
  Create panel scroll no longer clips the last card or its rounded corner.

🔧 Under the hood

• Bundled imference-engine → v0.3.2.

---

Full changelog: v0.2.3...v0.3.0

v0.2.3

Big release: a redesigned model picker, support for every engine image backend, local single-file checkpoints, automatic engine version management, and the low-VRAM SDXL performance fix.

⚡ SDXL performance on low-VRAM GPUs

Generations that could take 10+ minutes on 8 GB cards (SDXL 1024×1024) now run at normal speed.

• Auto CPU-offload: on startup the app probes GPU VRAM (nvidia-smi) and automatically enables CPU-offload on low-memory cards (≤ 12 GiB), keeping peak VRAM from spilling into slow shared memory. Exposed as a tri-state Auto / On / Off control in Settings — Auto is the default and decides per GPU.
• Ships pinned to imference-engine v0.3.1, which also fixes the root-cause VRAM leak in weighted prompt encoding.

🎴 New model picker

The model dropdown is replaced by a card-grid picker dialog:

• Models grouped by type, with search/filter.
• Works for both cloud and local models.
• A dedicated "My models" tab for manually-loaded custom checkpoints.

🧩 All engine image backends + custom local checkpoints

• Every backend the engine ships is now selectable: SDXL, SD 1.5, Z-Image, FLUX, Chroma, Qwen-Image, and Anima.
• Load your own .safetensors: pick a local checkpoint, choose which backend should load it, and (for transformer-only backends) optionally override the base repo. All seven backends are available in the custom-model flow — including Anima single-file DiT checkpoints (Civitai / ComfyUI form).
• Runtime settings unified into a single Image block that applies across every image backend (instead of per-backend duplication).

🔄 Engine version management

• The app version is now shown under the logo.
• On startup the app checks the installed engine against the pinned version and auto-syncs the venv (force reinstall) when it's outdated — no manual pip step.
• The installed engine version is shown in Settings under the engine section.
• Cloud-only setups are respected: if no engine is installed (e.g. no GPU), the app does not auto-install one.

🛠️ Fixes & internals

• Install torchvision alongside torch — required by the Anima / Cosmos forward pass.
• Fix the sidecar for the engine 0.3.x RuntimeConfig field rename (enable_cpu_offload → enable_offload).
• Pin sd_embed to a fixed commit for reproducible installs.

Upgrade

Download the v0.2.3 build. On first launch it will sync the bundled engine to the pinned version automatically (if an engine is installed).

v0.2.2

Full Changelog: v0.2.1...v0.2.2

v0.2.1

What's new in v0.2.1

✨ New branding

• New app icon — the Imference "Im" monogram now appears in the window
  title bar, taskbar/Dock, and installer, replacing the default Wails icon.
  Crisp at every size, including 16px.
• Refreshed in-app header — cleaner look with the new logo.

📄 Docs

• Fully reworked README focused on downloading and getting started:
  direct download links, first-launch steps (SmartScreen / Gatekeeper),
  and requirements at a glance.

---

First launch reminder — the app isn't code-signed yet:

• Windows: SmartScreen → More info → Run anyway
• macOS: right-click the app → Open

Verify your download with checksums.txt below.

v0.2.0

Migrates Imference Desktop to Wails v3 — a modernized app foundation with a new build system and cleaner Go↔frontend architecture. Functionality (cloud & local generation, wallet, gallery) is unchanged; this is an infrastructure upgrade.

Downloads

Platform	File
Windows (installer, recommended)	imference-desktop-go-windows-amd64-installer.exe
Windows (portable)	imference-desktop-go-windows-amd64.exe
macOS (universal)	imference-desktop-go-macos-universal.dmg

First launch (unsigned builds)

• Windows: SmartScreen → More info → Run anyway.
• macOS: right-click the app → Open (or xattr -dr com.apple.quarantine "Imference Desktop.app").

Verify downloads against checksums.txt (SHA-256).

Full Changelog: v0.1.0...v0.2.0

v0.1.0

A desktop app (Windows & macOS) for AI image generation — run models in the
cloud or locally on your own GPU, from one clean interface.

✨ Highlights

Two ways to generate

• ☁️ Cloud — generate on imference.com. Pay with an API key (credits) or
  with x402 (USDC on Base) — switch anytime; the app shows the active
  method's status and links you straight to its settings.
• 💻 Local — install the engine in one click and run SDXL / Z-Image
  models on your own GPU. The engine starts/stops on demand, so cloud-only users
  spend no GPU/RAM.

Model catalog & parameters

• Live model catalog with per-model credit cost for cloud runs.
• Per-model parameters from the catalog: format/ratio, steps, CFG, seed,
  editable quality tags (pre-prompt) + negative prompt.
• Image-to-image in local mode.

Gallery

• Every generation is saved with its metadata (prompt, model, params, seed).
• Aspect-aware masonry with a fullscreen viewer, filters (model / engine
  / source), adjustable columns, infinite scroll, and delete.

Settings

• One tidy window with a side nav; changes save automatically.
• Per-engine machine tuning (device, VAE, offload…).

📥 Install

Windows

• Recommended: imference-desktop-go-windows-amd64-installer.exe (installer), or
  the portable imference-desktop-go-windows-amd64.exe.
• Unsigned → SmartScreen may warn on first launch: More info → Run anyway.

macOS (universal — Intel & Apple Silicon)

• Open the .dmg, drag Imference Desktop to Applications.
• Unsigned/un-notarized → Gatekeeper blocks it on first open: right-click the
  app → Open, or run xattr -cr "/Applications/Imference Desktop.app".

Verify downloads against checksums.txt (SHA-256) if you like.

🧩 Requirements

• Windows 10/11 (x64) — WebView2 runtime (bundled by the installer).
• macOS 12+ (Apple Silicon or Intel).
• Local generation: an NVIDIA GPU (CUDA 12.4) on Windows, or Apple Silicon
  (MPS) on macOS. The one-click installer sets up an isolated Python environment;
  each model is ~6–7 GB, downloaded on demand.
• Cloud generation: an imference.com API key (credits) or a funded x402
  wallet (USDC on Base).

⚠️ Notes

• Unsigned builds — SmartScreen / Gatekeeper prompts as above. Signing +
  notarization are planned.
• No in-app auto-update yet — grab new versions from the Releases page for
  now (auto-update is coming next).
• Windows is the most-tested platform; macOS is fresh — please report issues.

Full Changelog: https://github.com/Publikey/imference-desktop/commits/v0.1.0