Radio Venice Admin

View Radio

Upload Song

Upload an existing MP3 file with its metadata.

MP3 File *

Title *

Artist

Genre

Generate Song

Create a new song using Venice.ai's music generation API.

Venice API Key

Stored in your browser only. Never sent to our servers.

Model

Duration — 120s (~$1.50)

3s2min4min6min8min10min

Prompt *

Lyrics

Song Metadata

Title

Artist

Genre

MiniMax v2.5 / v2.6 Test Matrix

Runs 5 empirical tests to verify how force_instrumental and lyrics_optimizer behave on the selected MiniMax variant. Each test queues a real generation (~$0.04 each, ~$0.20 total). Results appear below. Use this to resolve the UNKNOWN entries in RADIO_VENICE_MUSIC_MODEL_REFERENCE.md.

Test against:

Pending Review

Review and approve songs before they go live on the radio.

Song Library

Manage all songs across all channels.

Stations

Manage the stations that appear on the radio dial. Each station can have songs assigned to it via the Library → Edit Song modal. The dial shows the active 10 stations across 2 rows.

Add Station

Genre Options

Manage the genre options available in song edit/upload forms. Removing an option does not change songs that already use it.

Category Options

Manage the category (mood/use-case) options.

Style Options

Manage the style suggestion list shown when editing songs. Styles are free-text — this list provides curated suggestions.

Section Marker Templates

Custom section-marker templates that appear in the admin-only "Markers ▾" dropdown next to every Lyrics field. Built-in templates are defined in src/section-markers.js and cannot be deleted here; only custom templates added below can be edited or removed. Changes take effect on all devices within a page refresh.

AI Prompts

Edit the system prompts sent to the AI for each generation feature. Changes are saved to the database and take effect on all devices immediately. A full version history is kept — use Restore on any previous version to roll it back.

Generation Errors

A log of failed AI generation attempts. Records include the model used, number of retry attempts, error type, the prompt that triggered the failure, and the raw response received. No user identity is stored.