User Growth Nears 3 Million: Domestic AI Music Tool Mureka Upgrades to V7, Recreating the 'Indian Divine Song'}

AI is quietly taking over your playlists.

Recently, while browsing NetEase Cloud Music, I unexpectedly discovered a song that immediately evoked a moonlit, misty atmosphere.

Currently, this song has received 150,000 likes. I wanted to find out who the creator was, but the comments below flooded with: "This is AI-generated!"

Upon closer listening, you can still find clues—such as the muffled sound quality resembling a pixelated image, or the vocals sounding like they’re leaking through teeth. But with continuous evolution, AI music is becoming increasingly indistinguishable from real.

Now, the fire of AI music is burning brighter than ever.

On July 23, Kunlun Wanwei, a major AI model developer, officially released the new generation music model Mureka V7. It is currently the strongest domestically and significantly surpasses overseas AI music platform Suno (V4.5) in key metrics, including average performance score, mixing quality, vocal realism, and overall sound quality.

Moreover, compared to the previous V6 version, Mureka V7 produces higher-quality music, with greatly improved melody motifs, arrangement quality, and enhanced realism of vocals and instruments.

To put it simply, even someone with no musical talent can create ultra-fine, personalized background music with it. For professional musicians, Mureka V7’s generated music is highly innovative and can inspire new ideas.

      ^{Mureka V7 work "Du Fu"}

Now, Mureka V7 is fully available online. Interested users can visit the official website to experience it.

Official website: https://www.mureka.cn/

Next, let’s test whether Mureka V7 still has that "leaky teeth" feeling when creating music.

Hands-on Test: Can I Imitate Faye Wong and Generate a 'Tacky' MV?

Mureka V7 is not just about "AI writing a song"—it now features a new function: Custom Singer.

You can upload audio or simply provide a video link, and the AI will automatically imitate the tone and sing a newly created song.

Take Faye Wong, for example. Known as Teresa Teng’s loyal fan, she performed a duet of "Qingping Tune" at the 2013 "Teresa Teng 60th Anniversary Concert."

We let Mureka imitate Faye Wong’s voice, then recompose and sing. The generated voice captures her ethereal, transparent quality, with a slight breathy effect characteristic of her singing style. The pronunciation is natural, especially in the tail notes, with a lazy Faye-like vibe.

We also tried its Music Reference feature.

This feature analyzes uploaded music to accurately identify style, rhythm, instrumentation, and mood, then generates original works with similar style.

Recently, Chinese internet celebrity "Hao Gege" remixed the Indian divine song "Tunak Tunak Tun" (famous for the "I’m in Northeast playing mud" meme), creating a catchy parody called "My New Plane Got Hit."

Video from blogger "Hao Gege - Catchy Remix"

This song went viral on social media worldwide, even prompting Indian netizens to petition the United Nations.

We also made a version with Mureka, featuring a super catchy tune, with a slightly more Indian accent and a tacky MV with abstract visuals and yellow outlines—both cheesy and fun.

Additionally, Mureka has upgraded features like Song Description and Pure Music Generation.

For example, inputting Li Bai’s "Bring in the Wine" and selecting styles like "Rap Metal, Alternative Metal, Rap Rock, Male Voice," Mureka instantly transforms into a rock veteran, passionately singing.

Or generate copyright-free BGM directly from text prompts:

Prompt example: "A warm piano melody reminiscent of childhood"

You can also upload reference audio for the model to create similar style instrumental pieces.

If unsatisfied with the generated music, Mureka V7 offers Audio Editing features, allowing partial edits, extending songs, separating instruments, and cropping audio, supporting AI music creation in 10 languages.

Self-developed Music Thinking Chain "MusiCoT" Evolves Again

Within just four months, Mureka V7’s performance has significantly improved over Mureka V6, thanks to continuous optimization of the proprietary music generation thinking chain—MusiCoT.

We know that large language models generate content by predicting the next token, which is different from music creation. To address this, Kunlun Wanwei introduced the popular CoT (Chain of Thought) prompting method into Mureka, first showcased in V6.

This time, Mureka V7 further refines MusiCoT (Analyzable Chain-of-Musical-Thought Prompting), greatly enhancing the coherence and expressiveness of generated music, with three key innovations:

• Think structure first, then generate, mimicking human creativity.
• Before producing audio tokens, the model generates a global plan of the music structure, including sections, mood, and arrangement, ensuring clear composition.
• Interpretable and controllable generation.
• Using CLAP (Contrastive Language-Audio Pretraining), MusiCoT’s explicit reasoning chain makes the music more readable and controllable. Users can input style references of any length.
• Subjective and objective validation achieves industry-leading results.
• Extensive experiments show MusiCoT outperforms traditional methods in structure, melody, and overall musicality, with top-tier evaluation scores.

With Mureka V7, MusiCoT not only aligns with music creation thinking at the structural level but also improves controllability and scalability through data expansion and finer information embedding.

Thanks to MusiCoT’s upgrade, some generated works already inspire musicians and accelerate the journey from inspiration to finished product.

Better Domestic Voice Synthesis Options

In addition to more natural and human-like music generation, Kunlun Wanwei launched a new Audio Model — Mureka TTS V1, now available on the official website.

Unlike music, which emphasizes melody, harmony, rhythm, and style, the TTS model focuses on universal representation and understanding of all sounds, including speech, vocals, environmental sounds, and effects. The highlight is the Voice Design feature, allowing users to define voice characteristics via text.

This means that real or virtual characters, or voice-over roles, can be controlled through text commands, moving beyond preset voice libraries. Users can freely specify gender, age, emotion, tone, and pace, enabling personalized and scene-specific voice synthesis.

In tests against ElevenLabs TTS V2, Mureka TTS V1 outperformed in voice quality, phoneme accuracy, and overall listening experience, with only slightly lower pronunciation accuracy. This gives Kunlun Wanwei a leading edge in voice synthesis, capable of supporting advanced voice creation and interaction scenarios.

Ultimately, the real test is the effect. Let’s listen to a sample: "Child female voice, around 12 years old, bright and cheerful, slightly fast but not rushed."

And another: "Male news anchor, clear and steady voice, calm tone, moderate speed, neutral and objective, with a sense of depth, conveying professionalism and credibility."

This full-process voice generation from description to output greatly enhances efficiency and creativity, expanding the application scope of voice synthesis and opening new possibilities for content creation and interaction. In the future, it can be widely used in film, gaming, advertising, and other voice-over scenarios.

Final Words

Recently, as scaling laws slow down and the marginal gains from increasing model size diminish, many companies are slowing their efforts on large foundational models. Instead, the value realization and commercialization of large models are accelerating.

Vertical large models, such as video and music models in the AIGC field, are becoming new focal points in the tech race and product competition. Companies are racing to scale up, aiming to turn technology into products and dominate high-frequency content creation, marketing, and entertainment scenarios.

This aligns naturally with Kunlun Wanwei’s long-term strategy. Driven by the mission to "achieve general artificial intelligence and enable everyone to better express themselves," the company has built a full industry chain: AI frontier research — foundational models — AI matrix products and applications. It continues to push forward in AIGC creation, launching innovative products across video, music, and agent domains.

Since its debut in April 2024, Mureka has gained increasing recognition as a "thinking" music model. It has undergone two major updates, with the last version, V6, still highly praised by international users.

In just a few months, nearly 3 million new users have joined Mureka since late March. Clearly, Kunlun Wanwei’s music large model has gained high user approval and is leading a revolution in music creation.

As the model’s capabilities continue to grow and creation barriers lower, AI is expected to become the core driver of music production. Meanwhile, music creation will further democratize, breaking professional barriers and enabling everyone to express themselves freely.