Nova SQLite Reader 🗃️

Every serious music workflow eventually ends up with a spreadsheet or a database holding the real metadata: track number, title, ISRC, catalogue number, credits. Retyping that into widgets is how you get a typo on track 7 that nobody notices until the distributor rejects the release.

Nova SQLite Reader 🗃️ pulls a table into the graph as a NOVA_TABLE, which Nova Tag Writer can write onto files and Nova Master Identity can read for release metadata. One source of truth, typed once, in a database.

Inputs

• database_path - full path to an existing .db. Leave it empty and it uses the two fields below to create a new one - which is the path to feed from Nova NamePath Manager's split outputs when existing_database is off.
• new_database_folder / new_database_name - where that new file goes. .db gets appended if you don't give an extension.
• table_name - default albums.
• columns - * for everything, or a comma-separated list.
• where - empty selects all rows. This is raw SQL by design, so AND/OR/LIKE/IN/BETWEEN all work.
• column_set - a preset that picks columns for you: custom, tags (the columns Nova Tag Writer writes) or identity (the columns that map onto Nova Master Identity). A preset only applies while columns is * or empty - type a column list and that wins. Columns a preset asks for that your table doesn't have are skipped, not an error, which is the difference between a preset and a nuisance.
• value_column (optional) - name one column and it also comes out as plain text on the value output: lyrics, a comment, a path. When it's set, the query must return exactly one row, and the column must be among those selected (it always is with *).

Outputs

table (the NOVA_TABLE, for Tag Writer), record_count, column_count, column_headers (a real list output, so Nova Console prints it as one numbered block), identity_json, and value.

identity_json is the quietly clever one: the single selected row mapped onto Nova Master Identity's own field names. Wire it into that node's identity_fields_json input and your catalogue becomes the source of truth for the release - no retyping after a browser reset, and no drift between what the database says and what the graph says.

Read-only, and it means it

The connection is opened read-only, identifiers (table, columns) are validated against the live schema and quoted rather than interpolated, and the where text - raw SQL because you asked for operators - is screened for statement separators and anything that writes. A typo in the WHERE box cannot modify your database. That's a real constraint on what you can do, and it's the correct trade for a node whose job is reading metadata you can't afford to corrupt.

Install

ComfyUI Manager → Nova Audio Player → install → restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/NovaFemme/ComfyUI-NovaAudioPlayer.git

Under ▶️ Nova Audio → 📦 Delivery & Metadata. SQLite is in Python's standard library, so there is nothing to install, no model to download, and no database server to run. If you installed the pack through Manager, the example database at examples/nova_album_example.db is there to poke at before you point it at your own library.

The workflow this exists for

Nova SQLite Reader 🗃️ ── table ──▶ Nova Tag Writer 🏷️ ── files ──▶ Nova Tag Reader 🔖
        │ identity_json
        └──▶ Nova Master Identity 🪪 ── archive_name ──▶ Validator

Read the database, write the tags onto the files, read the tags back to prove it landed - and use the same row to name and fingerprint the archived master. That's the whole delivery chain, and the reader is where the truth enters it. Mark a Length column skip (it's skipped by default) because duration comes from the audio stream and isn't a settable tag; everything else auto-maps to a tag name.