Argus Studio handbook

These are interactive docs: concepts are explained in prose, then demonstrated with the same live widgets and constants the tools use — so what you read here and what you see in the app never drift apart.

The same explanations power the info panels next to each control in the tools. Look for the Learn more link inside a parameter's info panel to jump here.

The pipeline

The Argus suite takes you from raw images to an evaluated LoRA:

• Acquire — browse provenance-tracked public-domain images.
• Curate — score, de-duplicate, and select a training set.
• Caption — generate structured, training-optimised captions.
• Forge — turn the export into a ready-to-run training config.
• Proof — evaluate the trained LoRA and feed findings back.

Concepts

The four decisions the Caption tool asks you to make, in the order it asks them:

• Target style — natural-language or booru tag-style, matched to the checkpoint you will train against.
• Target backend — the architecture that consumes the caption, and the token budget it imposes.
• Target category — which caption variant becomes the training caption, and what good input looks like for each.
• Prose enrichment — whether novel prose phrases are appended to the training caption.

More concept and guide pages are on the way (see the interactive-docs roadmap).