FL Song Gen Style Transfer:
The FL_SongGen_StyleTransfer node is designed to generate music by leveraging a reference audio clip to condition the style of the output. This node is particularly useful for artists and creators who wish to produce new music that aligns with the style of an existing piece. By providing a reference audio, typically a chorus section, the node can synthesize new music that matches the style of the reference while incorporating user-provided lyrics. This capability allows for creative exploration and experimentation, enabling users to craft unique musical pieces that maintain a desired stylistic consistency. The node's functionality is enhanced by its ability to handle various output types, such as mixed tracks, vocals only, or background music only, providing flexibility in the creative process.
FL Song Gen Style Transfer Input Parameters:
model
This parameter requires a loaded SongGeneration model, which is essential for the node to function. The model acts as the core engine that processes the input data and generates the music output. It is crucial to ensure that the model is properly loaded to avoid any execution issues.
lyrics
The lyrics parameter allows you to input formatted lyrics with section tags, such as [intro-short], [verse], [chorus], and [outro-short]. This structured input helps the node understand the lyrical content and its intended placement within the generated music. The default value is a sample format: [intro-short] ; [verse] Hello world.This is a test ; [chorus] Singing along.Making music ; [outro-short].
reference_audio
This parameter accepts an audio file that serves as the style reference for the music generation process. The node uses a maximum of 10 seconds from this audio to condition the style of the output. It is recommended to use a chorus section for optimal results, as it typically encapsulates the essence of the song's style.
description
An optional parameter that allows you to provide a style description to be combined with the reference audio. This can include descriptors like female, pop, emotional, which help refine the style conditioning process. The default value is an empty string.
duration
This parameter specifies the target duration of the generated music in seconds. It ranges from 30.0 to 270.0 seconds, with a default value of 60.0 seconds. Adjusting this parameter allows you to control the length of the output track.
temperature
The temperature parameter influences the randomness of the sampling process during music generation. A higher temperature value results in more random and diverse outputs, while a lower value produces more focused and consistent results. The range is from 0.1 to 2.0, with a default of 0.9.
cfg_coef
This parameter stands for classifier-free guidance strength, which affects how strongly the model adheres to the provided style and lyrical content. The range is from 0.5 to 5.0, with a default value of 1.5. A higher value increases adherence to the input conditions.
top_k
Top-k sampling is a technique that limits the number of potential next steps in the generation process to the top k most likely options. This parameter ranges from 1 to 500, with a default of 50. Lower values result in more focused outputs, while higher values allow for more variation.
gen_type
This parameter determines the type of output you wish to generate. Options include mixed, separate, vocal, and bgm. The default is mixed, which provides a complete track with both vocals and background music. The other options allow for more specific outputs, such as vocals only or background music only.
seed
The seed parameter sets the random seed for the generation process, allowing for reproducibility of results. A value of -1 indicates a random seed, while any other integer value will produce consistent outputs across runs. The range is from -1 to 2147483647, with a default of -1.
FL Song Gen Style Transfer Output Parameters:
mixed_audio
This output provides the complete audio track, combining both vocals and background music. It is the result of the style transfer process, incorporating the reference audio's style and the provided lyrics.
vocal_audio
The vocal_audio output contains only the vocal component of the generated music. This is useful if you wish to isolate the vocal track for further processing or mixing.
bgm_audio
This output delivers the background music component of the generated track, excluding the vocals. It is ideal for users who want to focus on the instrumental aspect of the music.
FL Song Gen Style Transfer Usage Tips:
- To achieve the best stylistic match, use a clear and representative chorus section as your reference audio.
- Experiment with the temperature parameter to find the right balance between creativity and consistency in your music generation.
- Utilize the gen_type parameter to tailor the output to your specific needs, whether you require a full track or isolated components.
FL Song Gen Style Transfer Common Errors and Solutions:
Model not loaded
- Explanation: The node requires a loaded SongGeneration model to function, and it appears that no model is currently loaded.
- Solution: Ensure that you have properly loaded a SongGeneration model before executing the node.
Reference audio too long
- Explanation: The reference audio provided exceeds the maximum length of 10 seconds.
- Solution: Trim your reference audio to a maximum of 10 seconds, ideally focusing on a representative section like the chorus.
Invalid seed value
- Explanation: The seed value provided is outside the acceptable range.
- Solution: Ensure that the seed value is between -1 and 2147483647. Use -1 for a random seed or any other integer for reproducibility.
