### v2.2.1 - Fixed: opening a new image inherited the previous image's processing parameters and palette (e.g. every image got the previously selected 4-color palette) — every new image now starts with default parameters and no palette, while the selected algorithm is kept. - Preference learning no longer records or recommends palettes; it only learns your algorithm and parameter preferences (p…
### v2.2.0 - MCP and CLI now support the artisan and classiclab algorithms (the backend already supported them, but the enum lists were not updated, so selecting them was rejected by validation); a new algorithm_list tool/command now lists the public parameter ranges of all algorithms. - Fixed an issue where an algorithm explicitly specified via MCP could be silently overridden by history recip…
### v2.1.9 - Added a switch in Learning settings to turn off automatic parameter adjustment when opening an image — if you prefer full manual control, your current algorithm and parameters are now kept untouched (recipes from your own save history still apply unless preference learning is disabled). - Updated version to 2.1.9.
### v2.1.8 - Transparent backgrounds are now preserved after conversion instead of being filled with a solid color (only fully transparent areas stay transparent; semi-transparent pixels are still made opaque to avoid hollow pixels). - Updated version to 2.1.8.
### v2.1.7 - Fixed batch conversion and GIF processing silently producing blank (fully transparent) images when the conversion worker was interrupted by another operation — interrupted files are now automatically retried instead of being saved empty. - Fixed small images silently producing fully transparent output in auto/batch conversion (an over-aggressive block size collapsed the pixel grid…
### v2.1.6 - Fixed an issue where adjusting parameters (e.g. color count) with SLIC, Pixelizer, or Classic Lab unnecessarily reset the canvas, causing a black flash — the previous result now stays on screen until the new one is ready. - Updated version to 2.1.6.
### v2.1.5 - Learning data can now be exported and imported as files, and organized into multiple named datasets (like palettes) — create, activate, and manage them with the new card-based UI. - Clearing learned data is now tucked into a danger zone with double confirmation. - Steam Cloud now syncs all learning datasets across devices. - Updated version to 2.1.5.
### v2.1.4 - Each user has a unique aesthetic, so the software cannot apply a single universal conversion standard. - Privacy protection: All data is saved locally on the user's computer and in Steam Cloud saves. No information is uploaded to any server. - Preference learning: every save teaches the app your parameters. Similar images automatically get your preferred algorithm, params, and pale…
### v2.1.3 - Performance: pixelizer auto-detection up to 5.4× faster, SLIC up to 3.2× faster, Classic Lab ~20% faster (identical output). - SLIC and Classic Lab now convert in a background worker — the UI stays responsive during conversion. - Updated version to 2.1.3.
### v2.1.2 - Bead pattern export: cells now show per-pattern numbers (1–N) with a color code legend at the bottom, fixing overly long Perler / Perler Mini color codes. - Updated version to 2.1.2.
### v2.1.1 - Added Steam integration: Steam Cloud saves now sync recipes, settings, and custom palettes automatically. - Enabled contextIsolation and sandbox in the renderer for stronger security. - Fixed transparent borders around PNGs exported with the standard algorithm. - Updated version to 2.1.1.
### v2.1.0 - Upgraded to Electron 43 for better security, performance, and compatibility. - API keys are now stored in the OS secure storage (macOS Keychain / Windows DPAPI). - Smaller download packages with platform-specific binaries only. - Unified the pixelation engine across the app, CLI, and MCP. - Improved state management reliability and expanded automated test coverage. - Updated versio…
v2.0.7 - Frequency-weighted Median Cut: high-frequency colors (skin, sky, grass) get more palette slots across all algorithms. - Classic Lab upgraded to use the same frequency-weighted Median Cut for better color fidelity.
v2.0.6 - All palette operations upgraded to CIELAB perceptual color space (Median Cut, palette matching). - New Classic Lab algorithm: clean solid-color blocks, no dithering, pure classical pixel art. - Fixed Standard algorithm canvas sizing and block-edge artifacts when changing BlockSize. - Added Classic Lab option to the algorithm selector.
v2.0.4 - Fixed GIF conversion failure. - 1bit、Pixelizer support GIF conversion. - 1-bit "Color Count" now means actual color count: 2 = Bayer dithering, ≥3 = exact N-level grayscale. - Fixed black pixels turning transparent in 1-bit GIF output. - 1-bit images with ≤3 colors now save with a black background instead of transparent. - Fixed sub-pixel offset artifacts in square-cell rendering.
- Auto-detect optimal target width for Pixelizer algorithm using information elbow method. - Updated recommender to let auto-detection handle target width selection.
v2.0.2 - Fixed Pixelizer Grid alignment to precisely match pixel cell boundaries. - Added CLI "--show" flag to preview converted images directly in the terminal. - Custom palettes now support bead pattern export with auto-generated numeric color labels and a color legend.
- Upgraded app version to 2.0.0. - Added image-to-image support for Doubao-Seedream 4.0/4.5/5.0 when a source image is available. - Improved Save Original behavior to preserve original image extension instead of forcing .png.
- Upgraded the CLI with AI-friendly help, schema, dry-run planning, structured errors, and output controls. - Added convert/batch/tag dry-run support plus limit/page-limit controls for CLI workflows. - Updated version to 1.9.9.
Fixed an issue where the "Save Image" and "Open Image" buttons overlapped after using the text-to-image model. · Fixed issues related to the `debug.log` file. · Fixed an issue where, in the event of an error while using the text-to-image model, the underlying cause was not displayed correctly.
v1.9.7 - Remember the last used algorithm and parameters across app restarts. - Limit automatic algorithm recommendations to early onboarding hints for new users. - Updated version to 1.9.7.
Announcements are the developer’s own posts to the game’s Steam news feed: updates and their patch notes, in-game events, releases and sales. The important ones are marked on the players chart.