Wavemaker Blog
Page 5 of 12
RSS feed-
How to Pick the Right LoRA Epoch (Not the Last One)
Wavemaker saves every training epoch with comparable sample grids—here is how to choose a checkpoint that generalizes instead of overfitting your dataset.
-
ComfyUI JSON Formats Explained: API vs UI Save (and What Imports)
ComfyUI API JSON vs UI save JSON — structure, widgets, links, and what Wavemaker's importer reads. PNG metadata is a third path. Transpile-only; custom nodes are not executed.
-
Memoized Workflow Runs: Why Changing the Last Step Should Not Re-Bill the First
Wavemaker content-addresses block outputs per org — identical inputs hit the memo cache, while retryUntil and bestOfN use instance salt. How memoization makes iterative workflow runs affordable.
-
LoRA Dataset Preparation: Curation, Captions, and Safety Gates
How to prepare image, video, and voice datasets for Wavemaker LoRA training—counts, captions, moderation, and promote-from-runs workflows that beat raw folder dumps.