Repository navigation
Releases
Krea 2 LoRA Studio v1.0.2
Krea 2 LoRA Studio v1.0.2
This release polishes the connected-state label, adds a Studio overview screenshot and clear usage notes, and records approximate T4 timing observations from maintainer-reported Colab runs.
Approximate T4 observations
- Uninterrupted cold start: about 8 minutes (user-corrected estimate; earlier longer logs included Colab reconnects).
- One 100-step segment: about 14 minutes 52 seconds.
- Training speed in separate logs: roughly 7–8 seconds per step.
- Two Turbo previews: about 82 seconds including model setup/loading; about 44 seconds for image generation in that run.
These are indicative observations, not controlled benchmarks or time guarantees. Colab reconnects, caches, dataset settings, and preview settings affect elapsed time.
Verification
- Local CPU regression: 28 cases, 27 passed; one optional private-dataset case skipped.
- GitHub Actions: Python 3.10, 3.11, 3.12, and 3.13 all passed.
- Maintainer-reported Colab/T4 end-to-end test completed successfully.
See README, approximate timings, and testing notes.
Krea 2 LoRA Studio v1.0.1
What’s included
- Portable Google Colab notebook for Krea 2 LoRA training
- Lightweight Studio interface, checkpoint management, and optional Turbo previews
- Updated contributor/testing instructions and SHA-256 manifest
- CI regression matrix for Python 3.10–3.13
- MIT license for original project code and documentation; model terms remain separate
Verification
- Local regression suite: 27 passed, 1 skipped (the optional private-dataset fixture is not bundled)
- GitHub Actions: passed on Python 3.10, 3.11, 3.12, and 3.13
- GPU training and a clean Colab/T4 launch are not exercised by CI; please run the notebook in your own Colab session for that final environment check
The attached ZIP is a portable source bundle. The notebook is also available directly from the repository’s README.
Krea 2 LoRA Studio v1.0.0
First public release of the Colab-first Krea 2 LoRA training Studio.
Includes the standalone notebook, lightweight browser Studio, staged cache/train/preview workflow, resumable checkpoints, individual LoRA downloads, optional compact Turbo previews, and local regression tests.
Open the notebook from the repository README. Training requires a Google Colab T4 runtime and Hugging Face access to the gated krea/Krea-2-Raw model.
Local verification: python -m unittest -v test_krea2_gui.py — 27 passed, 1 skipped (the optional private dataset fixture is not included in the public source package).