Changelog
repo: SmartLLM Nova
2026-10-16
Version: 1.0.0 — Initial release
Five compact panel nodes for language/vision models, image tagging, detection, mask/bbox conversion and structured song prompts, with native and Docker backends.
A shared panel design inspired by DaSiWa, adapted for classic ComfyUI and Nodes 2.0 with searchable model/task selectors, collapsible sections, pinned sections and resizable text fields.
Controls update through editing and lifecycle events, reducing canvas work without native-widget visibility or recurring graph scans.
Independent Nova node IDs, menu, routes, settings, configuration and container ownership allow the original SmartLLM pack to remain installed. Saved server settings hydrate the UI, and native/Docker backends share the current model folder.
Connected prompt, detection and seed overrides, chained tasks, independent Standard/NSFW training selection per loader, exact 64-bit seed controls, and preserved image-batch and prompt-list processing.
Registry Manager retains model selections and backend quantization across reopening and follows the selected theme, with grouped model/source/integrity sections, matching mode bars and controls, and Docker cards for setup diagnostics, immutable image references and runtime actions.
Neutral as the default node design, plus Jade, Ocean, Amethyst, Amber and custom JSON themes. Theme selection persists in Nova’s config; bundled palettes are extracted once and preserve user edits and deletions.
Installable dependency extras and the bundled Florence implementation with declared vision dependencies and actionable dependency messages.
Guides and an illustrated visual tour covering the Nova panels, model selection, detection and converters, plus detailed model registration, file verification and Docker/Ollama runtime walkthroughs.