Version History
Development timeline and changes
Latest Release
Version 2.1.3 (November 2026)
Projects
- A project is now required. Everything happens inside a project, which keeps your drawings, the results and the history of every run. Until one is open, only the Projects and Hardware Check tabs can be used
- One library per project. Drawings are added once (files, a folder, or dragged onto the page) and copied into the project; the same image is never copied twice. Preprocessing, Model Diagnostics, Batch Processing and Cloud Inking all pick their drawings from the library, as a grid of thumbnails with filters (To ink with these settings, Already inked, All)
- Preprocessing keeps the original. It saves a corrected version next to it, and inking uses that version when Use the preprocessed version when there is one is on. Statistics files you add or calculate are saved in the project
- One folder for every combination of settings, local and cloud (for example
6h-MCG-Model_p512_o64_c1_x1, with_prefor the preprocessed version). The same settings replace only their own earlier results, so different models or values never overwrite each other - Run history with the parameters, folders, failed drawings and, for cloud runs, the cost; an optional external copy of every run (for example on a shared drive); projects are included in the Launcherβs Backup
Batch Processing
- Full-screen progress view that blocks the app while the model works, with the drawing in progress, patch progress, time elapsed and estimated time left, and a Stop button: what is finished is kept and the drawings not started stay in To ink
- The app asks before inking again drawings already done with the same settings (ink them again, ink only the new ones, or cancel) instead of doing nothing
- A prominent Open the results folder button when the run ends, and the reason for any failure is now shown instead of a success message
- Clearer model list: only the model name in the menu and its description in a note below; no more horizontal scrolling in the side menu
- Redesigned tabs, with the drawings on the left and the settings and the main button always in view on the right
Cloud Inking (optional, beta)
- New tab that inks with a general-purpose image model through OpenRouter (GPT Image 2 by default): no GPU and no download, about 0.05 $ per drawing. Choose the Fine or Coarse stippling, 1 to 4 versions per drawing and a cost limit; the cost of every image is shown and saved in the log
- Optional reference drawing to show the style you want: a pencil drawing with its inked version, a sheet with both, or just one inked drawing to copy the style from
- Your drawings leave the computer in this workflow, so the tab asks you to confirm it; the API key is kept in memory only and is never saved. Registration of the ink onto the pencil drawing is off by default
Fixes
- A batch with FP16 failed on every drawing, without a message, after a run in full precision (the model was reused whatever the precision); this is fixed and failures are reported
Version History
Version 2.1.2 (September 2026)
Interface
- New look shared with the other PyPottery applications, with locally served fonts and icons in place of the emoji in messages and buttons
- The Info window now shows the hardware the app detected (CPU cores, and the GPU or CPU Only) and has a Copy button for the citation;
Esccloses it - Batch Processing has an Open Output Folder button, like the diagnostics, and the previews of a large selection of images stay smooth: small thumbnails decoded a few at a time, at most 60 tiles (the rest is counted as β+N moreβ) and removal of a single image without reloading the others
- Hardware Check redesigned with clearer verdicts and messages, a card per component, and updated tips. Processors with 8 or more cores are now rated excellent
Fixes and stability
- Apply Preprocessing did nothing when clicked, because the page still looked for a checkbox that had been removed. It works again, and it always needs a statistics file (
.npy) - The browser now opens on the port the app is really running on, also when you change it with the
PORTvariable - The app shuts down by itself a few seconds after you close its browser tab
- Console messages are written as UTF-8 on Windows, so log lines with special characters no longer risk an encoding error
- Crashes are reported with a timestamp and the name of the thread that failed, which makes problems easier to diagnose
Version 2.1.1
- Real download progress bars for both the small style models and the large
sd-turbodiffusion backbone (~5GB), so first-time setup no longer looks stuck - Comparison images are now clickable and open fullscreen
- Removed Dark Mode
- Updated
diffusers/peft/transformersto their latest compatible versions
Version 2.1.0
- Modified System Requirements Check
- Added Pixel
- Minor UI Improvements
Version 2.0.1 (12/12/2025)
- Diffusion model downloaded in local directory (models)
Version 2.0.0 (December 2025)
π A PyPotteryInk major release π
Key Features:
- Flask Web Interface: New user-friendly web interface with multiple tabs:
- Hardware Check: Verify system compatibility
- Model Diagnostics: Preview processing with different settings
- Preprocessing: Analyze and optimize images before processing
- Batch Processing: Process multiple images with progress tracking
Version 1.0.0 (September 2025)
π A PyPotteryInk major release π
Key Features:
- Multi-GPU Support: Added support for Apple Silicon (M1/M2/M3) GPUs via Metal Performance Shaders
- Gradio Web Interface: New user-friendly web interface with multiple tabs:
- Hardware Check: Verify system compatibility
- Model Diagnostics: Preview processing with different settings
- Preprocessing: Analyze and optimize images before processing
- Batch Processing: Process multiple images with progress tracking
- Automated Installation: One-click installation scripts for all platforms
- Improved Compatibility: Fixed compatibility issues with newest packages
- Statistics Visualization: New detailed statistics analysis with:
- Summary tables showing all key metrics
- Interactive distribution plots
- Histogram and KDE visualizations
- Box plots for metric overview
- Advanced Export Options (Experimental):
- Element extraction: Automatically identify and extract individual pottery elements
- SVG export: Convert processed images to scalable vector graphics for publication
with contributions from Francesco di Filippo and Enzo Cocca
Version 0.0.3 (July 2025)
Update to the PyPotteryInk package, with a few improvements
Key Features:
- π Added new model
6h-MCmodel as default π - Added a
test_imgto visualize the performance of each model - Reworking of the
test.pymodule: now it test the library on thetest_img - Added downscaling options in
process_single_imageandprocess_folder
Models:
- Fine-tuned example for Monte Cimino assemblage (6h-MC)
- Fine-tuned model for historical and painted pottery (4h-PAINT)
Version 0.0.2 (April 2025)
A little update to the PyPotteryInk package, with a few bug fixes and improvements.
Key Features:
- Added an AI disclosure reminder when running the
process_folderfunction - Added a Google Colab notebook for easy testing of the package
- Fixed some errors in the documentation
Version 0.0.1 (January 2025)
Initial public release of PyPotteryInk.
Key Features:
- Image-to-image translation using pix2pix-turbo architecture
- Adaptive patch-based processing for high-resolution drawings
- Comprehensive preprocessing and postprocessing tools
- Web documentation
Models:
- Base model trained on 500 protohistoric pottery drawings (10k)
- Fine-tuned example for Monte Croce Guardia assemblage (6h-MCG)