Melodia is still being trained, many of these tips will become unnecessary in the coming weeks.

Tip 0: If Melodia mispronounces a word, try spelling it differently.

Tip 1: Add (or tell the auto-prompt box to) include markers like [Build] and [Drop] for EDM type songs in addition to [Chorus] and [Verse]. Using [Verse 1] causes Melodia to start from the beginning of the song. If you want to "skip to the good part" just start the lyrics without a tag (and maybe without a capital letter for the first word). Make sure to use [square brackets] and not (parentheses)

Tip 2: Use the auto-prompts for descriptions, they're usually good.

Tip 3: Melodia usually doesn’t use the voices of the listed artists, but names in prompts do have an important stylistic influence. It can create similar voices to some artists (like Frank Sinatra, Ella Fitzgerald, Johnny Cash, and Jason Derulo). Ping zaptrem on Discord with others you find!

Tip 4: To make the lyrics sound more like an artist, try adding their name to the verse/chorus tags (e.g., [Chorus: Taylor Swift]) Note this doesn’t always work and may cause the model to say the name out loud instead.

Tip 5: Have a set of lyrics but can’t think of a style? Try advanced + NO description prompt (or just a period)! Melodia will produce some interesting results.

Tip 6: The style and content of lyrics can sometimes have a significant effect on the output song style. Check the lyrics for your target style to see if they tend to phrase things in a certain way. For example, in God’s Plan by Drake:

“I been movin' calm, don't start no trouble with me Tryna keep it peaceful”

Notice the lyrics say “movin’” instead of moving, “Tryna” instead of trying to, and “don’t start no trouble with me” instead of don’t start trouble with me.

Tip 7: Capitalization of lyrics can cause them to be placed in the next verse.

Tip 8: Prompt Boost: For people who are familiar with diffusion models, this cranks up the classifier free guidance scale. It forces the model to pay more attention to the prompt at the expense of song quality. I suggest not using it and am considering removing it.

Examples of good Advanced mode descriptions (try the first one!):

1. A 2022 electronic pop track features vocals from Drew Taggart of The Chainsmokers. Synthesizers create a danceable and lively backbone throughout the song. Drum machines and electronic beats provide the rhythm, maintaining an upbeat tempo. Brief acoustic guitar elements add texture, complementing the electronic sounds. The track mixes electronic production with pop structures for wider appeal.
2. A synthetic beat and electronic sounds open the song. Jason Derulo's vocals dominate, supported by backing harmonies. The bridge features a distinct keyboard melody. Drum machines and synthesizers create a dance-pop rhythm. The composition blends pop with electronic dance elements
3. A synthesizer creates a moody ambiance throughout the track. Drums, with a relaxed tempo, add rhythmic support to the composition. Bass elements provide depth, enriching the overall sound. Drake's vocals are central, showcasing a blend of rapping and melodic singing. The track belongs to the rap genre, typical of Drake's style.
4. A clarinet leads the piece, a signature of Benny Goodman's style. Saxophones and trumpets add depth to the arrangement. The composition combines improvisation with structured sections. The song features a swing rhythm, characteristic of the Jazz genre.
5. It blends pop and rock genres with an alternative twist. A piano and electronic synthesizers create a melodic backdrop. Josh Dun provides energetic drum beats throughout. Tyler Joseph, the lead vocalist, also raps. The song features elements of hip-hop in its composition.