Skip to content
AI Music Video Generator

Settings

The Settings tab is a grouped panel where you manage your brand and app-wide options. Changes auto-save as you edit them, and every section is collapsible so the page opens short — expand only the thing you want to edit.

The Settings tab

The left rail has two tabs:

  • Brand — everything tied to the active brand: identity, default models, typography, lyric mode, equalizer style, voice, showcase, and platform connections.
  • App — output formats, the AI Lyric Generator, the Prompt Builder, Suno, and full-machine backup.

Brand

The Brand tab edits whichever brand is currently active. The header shows the active brand’s name. Below the panel, Create another brand and Import brand buttons let you manage additional brands inline — there is no separate Brands tab.

See Brands for how the rest of the UI changes once you have more than one brand.

Identity

  • Artist / Channel name — shown on the title card and bottom-left branding strip.
  • Logo — file picker for the PNG/JPG overlaid on every render.
  • Logo Height in pixels (default 200).

Identity values are baked into renders, so changes only affect future videos. See Set up your brand for the starter guide, or Fine-tune your brand for a field-by-field deep dive.

Logo overlay

  • Embed logo on the rendered video toggle — turn off to render without any logo overlay.
  • Position — pick which corner the logo sits in. Top-left, top-right, bottom-left, or bottom-right. Bottom-left sits above the song name in the bottom corner. Defaults to top-right.

Typography

  • Lyric font — font used for the lyric text overlaid on the video (the running line/karaoke captions). Defaults to Arial.
  • Title-card font — font used for the song title shown on the opening title card and the bottom-corner song name. Defaults to Arial.
  • Title-card font size — pixel size of the title-card title. Defaults to 72.

Lyric mode

  • Line — lyrics appear scattered across the screen in varied sizes and positions, one line at a time. (Default.)
  • Karaoke — lyrics appear bottom-center and highlight word-by-word as the song plays.
  • None — no lyrics displayed. Background visuals and music only.

Every render also accepts a per-song override on the Create page.

Equalizer

  • None — no equalizer bars. Cleaner background-only look. (Default.)
  • Glass Mirror — sharp bars on top with a dimmed reflection below — a glossy, reflective surface look.

Typography, lyric mode, and equalizer are all part of the brand’s visual identity, so different brands can run different fonts and lyric styles without overwriting each other. See Style & fonts for examples.

Voice

Short description of your voice and attitude as an artist, plus genre, vocal type, explicit-content policy, themes, and references. Drives the narrator the AI Lyric Generator writes to (always available).

Showcase

Opt-in community feature that lets other users discover your channel. Enter your YouTube channel URL and/or TikTok handle, toggle Show my accounts in Showcase on, and your card appears in the shared pool. See Showcase.

Platforms

Connections for each format you have enabled under App → Output formats. Each brand stores its own OAuth tokens — connecting YouTube under one brand does not connect any other brand.

  • YouTube — OAuth sign-in, Tags, Description Template (supports {title} and {channel}); once connected, Privacy (private / unlisted / public) and Category (YouTube category ID; 10 = Music). See Connecting YouTube.
  • TikTok — OAuth sign-in, Clip Duration (default 30s), End Card Text; once connected, Privacy, Post Mode (inbox = review in TikTok app, direct = goes live immediately; direct requires additional TikTok approval), plus Disable Duet, Disable Stitch, and Disable Comments toggles. See Connecting TikTok.
  • Instagram — OAuth sign-in via Meta, Clip Duration (default 60s; Meta’s current Reel limit is 90s), End Card Text, Caption Template (supports {title} and {channel}), and Share to Feed toggle. See Connecting Instagram.

Danger zone

  • Export brand — download a portable .mvg-brand zip containing identity, voice, showcase, and logo. OAuth tokens are excluded.
  • Delete this brand — visible only when you have 2+ brands. Deletes the active brand along with its renders, songs, and logo after a confirm dialog that tells you how many renders and songs will be cascaded. Can’t be undone.

App

Output formats

Choose any combination of YouTube (1920×1080 horizontal), TikTok (1080×1920 vertical clip), and Instagram (1080×1920 vertical Reel). Turning a format off hides its panel under Brand → Platforms.

Downloads & API keys

The HuggingFace access token lives here — needed only for downloading gated models like Cydonia 22B. It’s a global setting that spans every brand. The AI Lyric Generator itself is always available as the first step on the Create flow; pick which local LLM each brand uses under Brand → Default models. See AI Lyric Generator for the full walkthrough.

Prompt Builder

The Prompt Builder is the local LLM that turns lyrics into a per-act SCENE prompt for the background generator. Unlike the AI Lyric Generator (which is opt-in), the Prompt Builder runs on every render — pick a model in this panel before your first render so the app has something to write with.

Four tiers are offered, all running entirely on-device:

  • Qwen 2.5 3B Instruct — Fast. ~1.9 GB on disk, 4 GB RAM minimum (8 GB recommended). Lightweight default; builds usable per-act prompts in a few seconds.
  • Qwen 2.5 7B Instruct — Balanced. ~4.7 GB on disk, 8 GB RAM minimum (12 GB recommended). Noticeably better instruction-following — picks up lyric nuance the 3B model glosses over. The auto-suggested default whenever your machine has the headroom.
  • Qwen 2.5 14B Instruct — Best. ~9 GB on disk, 16 GB RAM minimum (24 GB recommended). Strongest Qwen tier; coherent multi-act arcs at ~3–6 s per act.
  • Cydonia 22B v2q — Uncensored. ~13 GB on disk, 24 GB RAM minimum (32 GB recommended). Creative-writing fine-tune that handles intimate, dark, or explicit lyric beats without softening them — useful when Qwen smooths content into chaste descriptions.

Each card shows a compatibility chip — Fits your machine / Tight fit / Not recommended — based on installed RAM, plus a download button, a progress bar mirrored at the bottom of the app, and Use this / Delete controls. Weights live in <app data>/prompt_models/<model_id>/.

The 14B and Cydonia weights are shared with the AI Lyric Generator’s 14B and Cydonia options — installing either model once serves both pickers.

You can override the active prompt model for a single song from the Create page without changing this default; see Per-song overrides.

Suno

Placeholder for a future Suno integration that will send your style prompt and generated lyrics directly into a Suno session and pull the finished audio back into the Create flow.

Backup

  • Export Data — writes a mvg-backup-YYYY-MM-DD.mvg file containing your settings and library metadata. Useful for migrating to another machine or keeping a snapshot before a big change.
  • Import Data — restores settings and library metadata from a previously exported .mvg file.

Backup covers app-wide state. To move a single brand between installs, use the Export brand / Import brand buttons on the Brand tab instead.

App-menu Clear actions

Two bulk actions live in the app menu (macOS app menu / Windows File menu), alongside the per-render delete button on each Library card:

  • Clear Renders… — deletes every render in the Library along with the renders/ and tmp/ folders on disk. Your source MP3s, downloaded LLM models, config, and saved auth tokens are kept.
  • Clear All Data… — nukes the entire app data directory: source MP3s, models, settings, tokens, everything. The app relaunches afterward as if freshly installed. Use with care.

Both actions prompt for confirmation before running.