How to make an AI music video from your own song

01 / Music Video workflow

Start with the right workflow

Music Video keeps your uploaded or imported audio as the soundtrack. It does not compose a replacement song. Prepare an audio file up to 6 minutes and 50 MB, and make sure you have permission to use the recording, composition and lyrics. A Donatello song can be imported after its generation finishes.

Choose 16:9 for a landscape music video or 9:16 for a vertical release before generating scene images. These are different compositions: a wide two-person shot can become cramped when repurposed vertically. Check the framing again rather than assuming a crop will preserve the story.

02 / Music Video workflow

Review the words before the pictures

Run the audio analysis, select lyrics or instrumental mode as appropriate, and compare the recognized text with the actual recording. Singing stretches vowels and repeats phrases; a plausible-looking transcript can still omit a word or place a line too early.

In Lanterns at Dawn, the first caption initially began with “lanterns.” The sung word “Paper” was missing. We added it and corrected its onset to 9.30 seconds. That is an editorial correction in this demo, not evidence that every song requires the same offset.

• Listen to the first and last sung word of each passage.
• Check every repeated chorus separately; timing can differ.
• Leave instrumental gaps free of unrelated lyric text.
• Preview karaoke highlights and individual words, not only the plain SRT text.

03 / Music Video workflow

Give each shot one clear visual job

Use a small cast with stable descriptions: clothing, colours, silhouette and recognizable features. Review the character references before generating the scene board. Describe a concrete subject, setting and composition for each shot; reserve camera movement for the animation direction.

Our demo uses Clara and Elias, an ivory outfit and a tan jacket, a rooftop at dawn and warm paper lanterns. Five scene-image revisions were made during review, including clothing and prop continuity. A curated example is a record of iteration, not a one-click quality promise.

• Keep a recurring prop the same size, material and position between connected shots.
• Check hands and object contacts at full image size.
• Avoid asking one short shot to introduce a character, change location and resolve the story.
• Regenerate only the weak image; preserve successful work.

04 / Music Video workflow

Animate only the images that already work

Quality animation in Music Video uses LTX scene clips of 1–5 generated seconds. The export adapts these clips to the scene timing. A 45-second music video is therefore an edited sequence of short source clips, not a single continuous 45-second generation.

Watch the beginning, middle and end of every clip. In this demo, three animation attempts were repeated after review. Reject disappearing props, unstable faces and implausible action even when the first frame looks attractive. There is no guarantee of lip-sync or exact beat synchronization.

05 / Music Video workflow

Compare three lyric modes on the same passage

The three clips below use the same original song, scene sequence and corrected word timing. Only the subtitle mode changes: full line, karaoke or one word at a time. They are cropped excerpts rendered with Donatello’s production typography code, not three separately generated songs.

Full lines give readers more context. Karaoke keeps the line while highlighting the current word. Word-by-word captions emphasize short phrases but demand especially careful timing. None of these modes fixes a transcription error automatically.

06 / Music Video workflow

Check the final MP4, not only the editor preview

Use a readable font, strong contrast and a position that keeps lyrics away from faces and platform controls. Preview the busiest passage at the size people will actually watch. Export at 1080p and 30 fps, then play the downloaded file with sound from start to finish. Export resolution does not increase the detail of a lower-resolution source clip.

Download the MP4 and SRT before the 72-hour retention window expires. Active queued or processing work and its source media are retained until it finishes. The working library holds up to 30 retained Music Video projects; it is not a lifetime generation quota.

• Check for missing first words, clipped tails and captions outside the frame.
• Listen for audio dropouts and verify the ending does not cut a sung phrase.
• Check both a phone-sized view and full-screen playback.
• Review a new vertical composition independently from a landscape version.

07 / Music Video workflow

Free processing and commercial use: what the demo proves

Free to use with welcome or earned credits. When credits are insufficient for an operation and no surveys are available, eligible automatic free processing uses the same lower-priority queue, configured wait and free-capacity safeguards as Full Video Creator. No subscription is required. Optional paid credits provide faster processing.

Unlimited means ongoing eligible free access, not instant processing, unlimited concurrency or uninterrupted computing capacity. Duration, storage and fair-use limits apply. Longer projects can take days in the free queue.

Commercial use of generated videos is permitted under Donatello’s Terms, including eligible free outputs, subject to rights in your music, lyrics, scripts, references, voices and other third-party material. Copyright, exclusivity and approval for monetization on another platform are not guaranteed.

Lanterns at Dawn was made using credits through the production workflow. Its published record includes discarded attempts. It demonstrates an export, not the waiting time or completion of a free-queue Music Video job. We do not label a credit-funded run as a free-run test.