Documentation Index
Fetch the complete documentation index at: https://mintlify.com/ollm/opencomic-ai-training/llms.txt
Use this file to discover all available pages before exploring further.
OpenComic AI Training drives Krita programmatically over D-Bus using the kra-remote plugin, which means its system requirements are shaped by both the Node.js generator and by Krita itself. Review the sections below before setting up the pipeline to avoid compatibility issues mid-run.
Windows and macOS are untested. The pipeline has been developed and tested exclusively on Linux (Ubuntu). D-Bus availability, AppImage compatibility, and filesystem path handling may behave differently on other operating systems. Contributions and bug reports for Windows and macOS support are welcome, but you should expect to make adjustments if you run the pipeline on those platforms.
Operating System
| Platform | Status |
|---|
| Linux (Ubuntu and derivatives) | ✅ Tested and supported |
| Other Linux distributions | ⚠️ Should work; untested |
| Windows | ⚠️ Untested — may require adjustments |
| macOS | ⚠️ Untested — may require adjustments |
Krita
| Requirement | Details |
|---|
| Minimum version | Krita 5.3.0 |
| Recommended delivery | AppImage (Linux) |
| Plugin required | kra-remote (see below) |
The Krita AppImage is the recommended way to install Krita on Linux for use with this pipeline. The AppImage is a self-contained executable that does not require system-wide installation, avoids conflicts with any distro-packaged Krita version, and is straightforward to pass directly to the --krita argument. Download the latest AppImage from krita.org. kra-remote Plugin
The kra-remote plugin must be installed and enabled inside your Krita instance. The plugin exposes a D-Bus interface that the generator uses to issue drawing commands, apply filters, and export image frames.
Install the plugin by following the instructions in the kra-remote repository. After installation, launch Krita at least once manually and confirm that the plugin loads without errors before starting a generation run.
Node.js and npm
| Requirement | Details |
|---|
| Node.js | Any recent LTS release (Node 18, 20, or 22 recommended) |
| npm | Included with all official Node.js distributions |
| Package manager | npm (used by the build and generate scripts) |
npm run prepare) requires tsc (TypeScript compiler) and rollup, both of which are installed as devDependencies and do not need to be present globally on your system.
Key runtime dependencies include:
| Package | Purpose |
|---|
sharp | Image processing and format conversion |
@napi-rs/canvas | Procedural canvas rendering |
yaml | Parsing YAML preset option files |
dbus-next | D-Bus communication with the kra-remote plugin |
cli-progress | Terminal progress bar during generation |
paper | Vector graphics and geometry used in procedural drawing routines |
pure-rand | Deterministic random number generation for reproducibility |
Disk Space
Dataset generation can produce a very large number of image files. Plan your storage accordingly:
- The artifact removal models in the project were trained on 400,000 image pairs.
- Descreen models used 120,000 image pairs.
- Upscale models used between 25,000 and 100,000 image pairs depending on scale factor.
Each image pair consists of a clean PNG and a degraded PNG. At typical comic panel resolutions, a full artifact-removal dataset run of 400,000 pairs can occupy several hundred gigabytes of disk space. Use a fast local disk (NVMe or SSD) for the datasets/ output directory to keep I/O from becoming a bottleneck during long generation runs.
Hardware
There are no strict GPU requirements for dataset generation — all rendering is performed by Krita on the CPU. However, generation is a CPU- and I/O-bound workload, so faster processors and faster storage will directly reduce total generation time. GPU acceleration is relevant only during the downstream model training phase, which is handled by a separate framework such as traiNNer-redux.