Osmanthus wine of longing

I used the Refly.ai tool to integrate the results of multiple AI large models and completed the creation of a song lyric after manual adjustments. At the same time, I tested the capabilities of each model in advanced text creation.

Initially, my goal was to explore ways to improve the quality of AI-generated lyrics. Therefore, I first sorted out the existing issues with AI lyric generation, then had AI provide revision suggestions based on the prompts. Subsequently, I directly adopted AI's suggestions to generate a version of the lyrics and felt that the quality was improved to some extent compared to the often blindly generated results of AI.

Currently, AI has the following problems in Chinese lyric creation:

• Clichéd wording: The words used are too common. In fact, when writing lyrics, sometimes colloquial expressions are needed to appear humorous, while at other times, fresh and beautiful classical vocabulary should be employed to enhance elegance.

• Lack of creativity: AI often directs themes and intentions towards positive aspects, making it difficult to create alternative songs with self-deprecating or melancholic styles that satisfy the desired content.

• Rigid format: It presents a structured format similar to "eight-legged essays". Ideally, lyrics should incorporate various forms like Bridge, short rap segments, Rabato, etc., with flexible and varying sentence lengths to enhance the richness of the song.

• Stiff rhymes: The lyrics generated by AI either have poor rhyming effects or add irrelevant or atmosphere-destroying words for the sake of rhyme.

• Lack of allusions: In Chinese songs, especially ancient style songs, new words are often created through allusions, borrowing from ancient poetry, or coining phrases, yet the lyrics automatically generated by AI are basically composed of common words.

• Limited English expression: Some common songwriting AIs, like Suno, only accept English prompts or have an inaccurate understanding of Chinese, with a word limit of around 200 words. This makes it quite difficult for non-native English speakers and those not very familiar with musical terminology to accurately express their ideas under such constraints.

During the creation process, I tried various large models and tested the text generation abilities of these models. However, since the final results still did not meet expectations, I shifted my focus to revising and adjusting this specific song.

The final conclusion is: AI is a powerful creative tool, but it ultimately cannot fully understand our thoughts. To achieve writing outcomes that perfectly meet requirements, especially in creations like poetry and lyrics, manual intervention is always necessary at present (and even in the future). However, large models can quickly provide a wealth of reference content, which is still significantly helpful in the creative process. Especially in tools like Refly that integrate multiple models, it allows for comparison of different AI outputs, and adjusts AI output through progressive input or timely modification of reference content, making it very convenient to use.

Canvas Preview

Canvas link: https://refly.ai/share/canvas/can-v2489sx2bqat6wc9ykig4xli

E.T. Music https://etmusic.framer.website