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

Mirelo raises $41M from Index and a16z to solve AI video's silent problem | TechCrunch

AI lets anyone create videos, but many AI video creation tools lack support for audio. Mirelo is building AI that adds soundtracks to match the video's action. Earlier this year, the Berlin-based startup released Mirelo SFX v1.5, an AI model that interprets videos to add synced sound effects (SFX). This attracted attention from VCs gearing up for a generative AI revolution in games. The two-year-old German startup has raised a $41 million seed round led by Index Ventures and Andreessen Horowitz, TechCrunch learned exclusively. This new capital will help Mirelo compete more effectively in its emerging category. While it was still in stealth mode and resource-constrained, large companies such as Sony and Tencent released video-to-SFX models. So did Kuaishou-owned Kling AI, out of China, and ElevenLabs, which is also backed by a16z. While Mirelo already differs from them by its narrower focus, beating these models in the long run requires the startup to make additional hires. Altogether, the startup expects its team of 10 people to "double if not triple" in headcount by the end of next year, Mirelo CEO and co-founder CJ Simon-Gabriel told TechCrunch. These new hires will support Mirelo's R&D, as well as its product and go-to-market strategy. The startup published its models on Fal.ai and Replicate, and expects API usage to drive most of its revenue in the short term, Simon-Gabriel said. But it is also investing in building out its workspace for creators, Mirelo Studio, which could eventually support full professional use. As Mirelo prepares to scale, the startup and its investors are also anticipating concerns around training data that have dogged other generative AI companies. According to Georgia Stevenson, who led Index's investments, Mirelo based its models on public and purchased sound libraries, and is signing revenue-sharing partnerships that respect artists' rights. It's a tension inherent to generative AI tools, but Mirelo isn't displacing musicians and sound designers -- at least not yet. With a freemium model including a recommended plan for creators priced at €20/month (approximately $23.50), the startup is mostly targeting amateurs and prosumers hoping to unmute AI-generated videos. According to Simon-Gabriel, creators can't fully benefit from this new potential without audio. "George Lucas said that sound is 50% of the movie-going experience. It's not an overstatement," he said. "If anything, it's an understatement. You can take exactly the same images, and the sound will shape a completely different ambience, depending on the sound and the music that you put in there." He and his co-founder, Florian Wenzel, are both AI researchers and musicians themselves, and the startup has AI music generation on its roadmap. But Mirelo is seeing more pull for sound effects, in part because there is less research happening than in other AI fields, Simon-Gabriel said. "It's easier to build a real moat here, and then to capitalize on it," he noted. This could pay off for Mirelo. Simon-Gabriel declined to disclose its new valuation, but said it had increased "very significantly" compared to its previously undisclosed pre-seed round. That earlier round was led by Berlin-based firm Atlantic, which also participated in the new funding, bringing Mirelo's total raised to $44 million and helping close its resource gap. The startup is also backed by angels who lend credibility to its technology and could open new doors, including Mistral CEO Arthur Mensch, Hugging Face chief science officer Thomas Wolf, Fal.ai co-founder Burkay Gur, and others. Still, the team is aware that AI-generated videos may not be mute for long. For instance, Gemini's video generator now incorporates soundtracks powered by DeepMind's Veo 3.1 video-to-audio model. But if anything, Simon-Gabriel sounds vindicated. "Now, suddenly, people realize, 'Oh, maybe we should add sound.' But, of course, you should add some. It's a bit like silent movies versus talkies, right? It does make quite a difference!"

Investing in Mirelo | Andreessen Horowitz

In a 1992 New York Times interview, George Lucas said, "Sound is half the experience in seeing a film." For anyone who's enjoyed the satisfying swish of a lightsaber, the beep-boops of R2-D2, or the zoom of a Podracer (or the millions of other sounds characterizing films beyond the Star Wars universe), that remark makes complete sense. Over three decades later, that insight still holds true -- even as the medium itself has evolved dramatically. Cinema has given way to social media as the dominant cultural stage, and generative AI can now produce increasingly long and complex clips. But one simple fact remains: sound doesn't just accompany what we see -- it defines what we feel. That's why we're excited to be co-leading Mirelo's Seed round. Mirelo is building a state of the art foundation model focused on the sound layer of video. When testing the frontier of AI-generated sound, we evaluated a wide range of models and Mirelo stood out as a clear leader. It's rare to find a team that combines in-depth AI research experience from some of the best labs in the world with a background as former musicians with a passion for building creative tools for audio. We are also excited about the market. Looking ahead, every video model will generate audio. But as we have seen in other model categories, sophisticated creators don't just want a default output, they want control. They'll want the ability to audition different versions of a scene's audio, exploring alternative soundscapes to find the one that best fits their creative intent. They'll need tools to remix and rebalance stems, adjusting the relative emphasis of dialogue, effects, and ambience just as skilled sound engineers do today. Just as image models allow precise edits -- removing objects, refining lighting, or adjusting composition -- audio tools will evolve to let creators edit sound with comparable precision. And to maintain continuity and consistency across a project or an entire body of work, they'll rely on reference sounds, much like color reference frames or style guides in visual production. Even in traditional film, sound design is far more active than most viewers realize. Much of what you hear -- from wildlife to weather -- is re-created or enhanced in post production. Sound adds emotional depth, texture, and realism. It's the half of the experience that Lucas was talking about. We're excited to partner Carl-Johann Simon-Gabriel, Florian Wenzel, and the amazing Mirelo team and can't wait to hear what the future holds.

Audio generation startup Mirelo nabs $41M investment - SiliconANGLE

Mirelo AI GmbH, a Germany-based developer of audio generation models, today announced that it has raised $41 million in funding. Index Ventures and Andreessen Horowitz led the seed investment. They were joined by Atlantic.vc and TriplePoint Capital. Mirelo's flagship AI model, Mirelo SFX 1.5, is designed to generate audio for silent videos. In a demo published today, the model analyzed a clip of a drummer and generated a drum solo synced to the drummer's movements. Mirelo says that SFX 1.5 bested two popular alternatives by a wide margin in an evaluation carried out by a group of independent reviewers. According to the company, one of the model's advantages over the competition is that it generates fewer unwanted sounds. Additionally, SFX 1.5 is capable of processing fast-paced videos. Mirelo says that the model can sync the audio it produces to the events depicted in a video even when there's rapid motion involved. SFX 1.5 is available through an application programming interface and a non-technical app called Mirelo Studio. According to the company, the app enables users to create audio files with text prompts. They can have Mirelo Studio generate multiple versions of the audio, save the version one that best matches a given clip and then manually refine it if necessary. The company today revealed that it's developing a new AI model with more advanced audio capabilities than SFX 1.5. According to Mirelo, the algorithm will be better at maintaining consistency between different audio files. The company's longer-term development roadmap places an emphasis on extending its models to additional use cases, notably generating audio for films and video games. A job posting reveals that Mirelo is using a cluster of Nvidia Corp.'s previous-generation H100 and H200 chips to train its AI models. Another listing suggests that the cluster is powered by Slurm. Released in 2002, Slurm is an open-source tool for distributing workloads across a large number of chips. Mirelo's job postings indicate that it's using PyTorch to power its AI research efforts. PyTorch is a popular model development framework created by Meta Platforms Inc. One of the Mirelo job postings mentions the framework's FSDP feature, which makes it possible to split large models across the memory of multiple servers to speed up training. Mirelo is seeking to recruit a researcher knowledgeable in the diffusion and autoregressive model architectures. Those are the two most widely used approaches to designing audio generators. Diffusion is the go-to architecture for media generation models. Autoregressive AI, in turn, is a fairly broad technical term that encompasses, among others, transformers. The transformer architecture is most closely associated with large language models, but also lends itself to audio generation. Nvidia released a transformer-based audio generator called Fugatto last year. Mirelo will use the proceeds from its seed round to grow its research, product and go-to-market teams.