Creating a Music Video with Seedance 2.5 and Suno

The music video for the song "Forest" was created using Seedance 2.5 and Suno models, where the entire visual sequence is generated by a single neural network to ensure continuity.

The project creation took three days and cost 8,000 rubles. The author used an audio reference of their own recording to convey tempo, manner, and vocal timbre.

Suno allows you to create a dense arrangement that accurately replicates the original vocal delivery, eliminating the effect of music simply accompanying the text.

For Suno configuration, it is recommended to use an audio reference strength of 95, style strength of 35–45, and set the weirdness parameter to 30.

It is important not to specify the tempo (BPM) in the style prompt so that the model takes the rhythm from the rap, avoiding divergence from the source.

Using only the Seedance 2.5 model guarantees seamless continuity: identical faces, color logic, and overall style across all scenes of the music video.

During lip-syncing, the voice is transmitted via an audio reference; it is crucial to explicitly specify in the prompt which parameters are taken from the sound: phrasing, breathing, and line order.

Video beats are calculated based on verse lines rather than seconds, ensuring precise synchronization of camera movement with the song lyrics.

Cutting the video within a sung line disrupts sync, so editing should be performed strictly between lines.

Copyright filters may block generation if brand elements or currency symbols are present, even if the prompt passed initial checks.

To avoid rejection, do not name car brands or license plates in the text, leaving them empty or using hashes.

A photorealistic reference can turn a frame into a static image, so it is better to use it as the first frame and define movement via the prompt.

The description of a night scene should include only a mention of darkness; any lighting details might cause the model to create a daytime background.

Final editing and color correction were done in Adobe Premiere, while texts and prompts were created using Claude.