Mirelo Raises $41M from Index and a16z to Solve AI Video's Silent Problem with Audio Generation

Mirelo Secures Major Seed Funding to Address Critical Gap in AI Video

Berlin-based startup Mirelo has raised $41 million in seed funding led by Index Ventures and Andreessen Horowitz, with participation from Atlantic.vc and TriplePoint Capital 1

1

3

3

. The investment addresses what many consider AI video's silent problem: while AI audio generation tools can create stunning visuals, most video creation tools lack robust support for audio. This funding brings Mirelo's total capital raised to $44 million and positions the two-year-old company to compete against recent entries from Sony, Tencent, Kuaishou-owned Kling AI, and ElevenLabs 1

1

.

Building an AI Foundation Model for Sound Layer of Video

Mirelo's flagship product, Mirelo SFX 1.5, functions as an AI foundation model for sound that interprets video content and generates synced audio for silent videos 1

1

3

3

. In demonstrations, the model analyzed a clip of a drummer and produced synchronized sound effects perfectly timed to the drummer's movements. Independent reviewers found that SFX 1.5 outperformed two popular alternatives by a wide margin, particularly in generating fewer unwanted sounds and processing fast-paced videos with rapid motion 3

3

. The technology delivers what Andreessen Horowitz describes as advanced AI tools for audio that define what viewers feel, not just what they see 2

2

.

Technical Infrastructure and Development Approach

Mirelo trains its models using a cluster of Nvidia H100 and H200 chips powered by Slurm, an open-source workload distribution tool 3

3

. The company leverages PyTorch, Meta's model development framework, and is exploring both diffusion models and autoregressive models in its research 3

3

. This dual-architecture approach positions Mirelo to capitalize on the strengths of both techniques: diffusion for media generation and transformer architecture for sophisticated audio creation. The company's next-generation model promises better consistency between different audio files and will extend capabilities to film and video game audio applications 3

3

.

Monetization Strategy Through Mirelo Studio and API Access

The startup has published its models on Fal.ai and Replicate, with API usage expected to drive most revenue in the short term 1

1

. Mirelo also offers a freemium model through Mirelo Studio, its workspace for creators, with a recommended plan priced at €20 per month (approximately $23.50) targeting amateurs and prosumers 1

1

. Mirelo Studio enables users to generate audio files with text prompts, create multiple versions, and manually refine outputs 3

3

. The platform could eventually support full professional use as sophisticated creators demand more control over soundscapes, similar to how they edit lighting and composition in visual production 2

2

.

Ethical Data Sourcing and Artist Rights

Anticipating concerns that have plagued other generative AI companies, Mirelo has prioritized ethical data sourcing in its training approach. According to Georgia Stevenson, who led Index Ventures' investment, Mirelo based its models on public and purchased sound libraries and is signing revenue-sharing partnerships that respect artists' rights 1

1

. This positions the company to avoid the legal and ethical controversies that have surrounded other AI firms. Co-founders Carl-Johann Simon-Gabriel and Florian Wenzel, both AI researchers and musicians themselves, bring credibility to this approach 1

1

.

Competitive Positioning and Growth Plans

Mirelo expects to double or triple its current team of 10 people by the end of next year to support R&D, product development, and go-to-market strategy 1

1

. The startup's narrower focus on synchronized sound effects differentiates it from competitors, and CEO Simon-Gabriel believes there's an opportunity to build a real moat in this space where less research is happening compared to other AI fields 1

1

. The company is backed by notable angels including Mistral CEO Arthur Mensch, Hugging Face chief science officer Thomas Wolf, and Fal.ai co-founder Burkay Gur 1

1

. While acknowledging that competitors like Gemini now incorporate soundtracks to AI-generated videos powered by DeepMind's Veo 3.1, Simon-Gabriel sees validation in the market's recognition that sound matters, comparing the shift to the transition from silent movies to talkies 1

1

.