Noisee: Seeing the Unseen
From a Spark to a Story: Why We Built Noisee
I’ve often had this feeling: when a song just clicks, and suddenly, a whole music video starts playing in my head. I can almost see the scenes, the vibe, and the little story unfolding. But then comes the usual roadblock: how do I turn that vision into reality? For most people, the thought of mastering complex video editing software - on top of the challenge of creating footage in the first place - is often where that creative spark fizzles out.
Everyone has their stories to tell, this is something my PM and I talked about a lot when we first started thinking about Noisee, Moonshot AI’s first experimental project. We witnessed these amazing multimodal AI tools that could generate video clips, which was super exciting. But we pushed ourselves to think beyond just the tech:
“What do people really want to create?”
We kept coming back to a simple truth: people care less about tech, or “abilities.” They want to tell stories. It’s fundamental. People tell stories, consume stories, share stories, and eventually connect through stories. Honestly, we’re not looking to just create isolated, silent 4-second video clips. That feels more like an animated GIF than a compelling piece of storytelling.
Noisee: For the Storyteller in Everyone
So, our big idea with Noisee wasn’t just to build another video tool. We wanted to help everyone become a storyteller (or, to quote my PM, ‘become a movie director’). We pictured AI as a creative partner, something that could help you get those stories out of your head and into the world. Through the power of music videos, without needing a film school degree.
Noisee is designed to be that bridge - making it intuitive to combine your chosen song, a visual style, and your narrative ideas into something real. As our slogan says: Seeing the Unseen.
The Gen-AI Advantage: Creation That’s Alive and Iterative
The other thing that got me really excited is how Gen-AI changes the creative process itself. Think about traditional filmmaking - once out of production, it’s pretty much fixed. But AI-native tools are different: they naturally allow for a much more fluid, iterative way of creating.
What you make with Noisee doesn’t have to be the final, perfect version right away. That was a key insight for us. You can easily go back, tweak a prompt for a specific clip, try a different visual idea, or remix someone else’s music video you find fascinating. That continuous back-and-forth, that ability to easily refine and reshape your story, is what makes Noisee so powerful.
Apparently, managing and editing such a timeline isn't for everyone.
Crafting the Vision 1: The Input Experience
My first major design focus, outside crafting a “retro-futuristic” homepage, was the input process. Unlike text-to-image tools where a single prompt box is the primary interface, for Noisee, encouraging users to select a specific section from a music file or a link (from YouTube, SoundCloud, or Suno AI) became a crucial starting point.
Carefully choosing the best portion of the song is vital. This selection not only captures the emotional peak but also enables video clip transitions to synchronize naturally with the music’s rhythm. Simply put, different segments from the same song can dramatically alter the final music video’s impact and feel.
After Noisee processes the uploaded music and analyzes its beats, a timeline selection tool appears. This allows users to define the precise start and end points of their desired audio segment, and play it to ensure this segment is the perfect foundation for their videos.
If music forms the skeleton of the video, the user’s prompt provides the flesh. This prompt is then expanded into a storyline, distributed across the individual clips that make up the short video.
While providing a style reference image is optional, our researchers found that users love this feature, as a reference image drastically improves visual consistency across all clips within the generated video, leading to a more polished and coherent final production.
Lowering Barriers: How Design Iteration Enhanced Initial User Success
As Noisee’s user base grew rapidly, we observed a significant shift in our user profile. Our initial “Noiseers” were often musicians, video makers, and content creators already familiar with concepts like “video assets.” However, an increasing number of new users were simply curious individuals wanting to give Noisee a try.
Analyzing our user funnel, I discovered that some of these new users hesitated at the first step — providing music — while others submitted very simple prompts and often overlooked the style reference option. Unfortunately, this meant some left before even generating their first video, and others received results that didn’t meet their expectations.
To help these new “Noiseers” successfully tell their own stories with music, I made several key iterations to the input experience.
Templates for Instant Inspiration
I designed a template page where users could select from a gallery of pre-made music videos created by our team. Choosing a template provided them with pre-filled music, a developed storyline, and a specific style reference. Users could then hit “Generate” for a similar result or easily modify any of these input to “remix” the template into their own unique creation.
Easier Custom Creation
For users preferring to start from scratch, I enriched the input options to provide more guidance and flexibility:
Music Source: Beyond uploading files or providing links, users could now select music from Noisee’s integrated library.
Story Development: With a new “Extended” feature, users could input a simple idea or a few keywords, and our LLM would expand it into a richer, more detailed storyline. This provided a more developed narrative foundation, helping set clearer expectations for the final video.
Style Selection: In addition to uploading local images, users could now choose from a curated set of 11 preset style references, ranging from “Cinematic” and “3D Anime” to “Dreamscape.”
My aim with these additions was to offer solid starting points and sources of inspiration for those who wanted to experience Noisee but perhaps lacked a clear initial concept. Ultimately, these features aimed to level the playing field between ordinary users and seasoned content creators because while Noisee provides the tools, the power of the story is key.
The Impact of Iteration
The positive effects of these changes were quickly evident. Just one week after deployment, I observed an 11% decrease in users abandoning the process before hitting the 'Generate' button, and an 8% increase in users downloading their videos after generation. These metrics clearly indicated that more users were successfully navigating the initial creation steps and were significantly more satisfied with their music videos, feeling confident enough to save or share them.
Crafting the Vision 2: The Editing Experience
Three major design goals guided Noisee’s editing experience:
Clearly communicate music video structure:
A music video is a sequence of video clips set to a music segment. The design must allow users to easily view all clips, their visuals, and timestamps in a clear, ordered timeline. Users should also be able to preview the entire music video, jump to the start of any clip, or navigate to a specific second by dragging the player control.
Define editor capabilities and limitations:
Noisee expands the initial user prompt into a complete story, breaking it into distinct video clips. Consequently, the original overarching story isn't editable after the initial generation. However, users can iteratively edit the prompt for each individual clip. This allows them to refine storytelling, modify visuals, or even significantly alter the narrative of specific segments.
Convey multimodal LLM status and errors:
Regenerating a clip takes time. While precise time estimation is difficult, a pseudo progress indicator on the clip thumbnail visually communicates the regeneration progress. The editor also provides clear error messages if an iterated prompt contains sensitive content or if the multimodal LLM fails to respond, preventing the intended redraw.
Upon release, the editor received enthusiastic feedback from our Discord community. One music artist posted this opinion in an X post:
"Dear AI video tools, all of you, this is the ‘timeline’ editor I want, and @NoiseeAI just made it. I believe the old school timeline will always be useful, AND there are other more powerful and easier approaches that co-exist."
This positive reception directly reflects my commitment to an iterative design process. It was achieved through rigorous exploration and prototyping, all guided by the three core design goals established at the outset.
Beyond the Interface: Noisee's Vibrant Community
A story truly comes alive when it’s shared - a core principle in my design thinking for Noisee. The team believed our users’ music video creations should be compelling enough to share widely, sparking further creativity.
The vision rapidly found its home on Discord, which quickly became the vibrant hub of the Noisee community. Amazingly, we welcomed 10,000 “Noiseers” just two months after the initial release! Our Discord channel is filled with members showcasing their fascinating works and providing invaluable feedback directly to our team.
To help our community share these creations even more broadly beyond Discord, I designed a set of covers for videos with different aspect ratios. These act as polished, instantly shareable thumbnails, automatically generated to be the first frame of the video.
I even created a special edition cover to celebrate our partnership with Suno AI. Recognizing the fantastic workflow of users generating songs on Suno and then bringing them to life in Noisee, this cover gave a unique visual flair to these collaborative creations.
A Fond Farewell: Reflecting on Noisee's Journey
On January 15th, 2025, the Noisee server and website were officially taken offline. This difficult decision was made as Moonshot AI chose to strategically focus all computing resources on its core product, Kimi, and the intensive graphical demands of video generation were a significant factor.
However, the sunset of Noisee doesn’t signify a failure; far from it. As Moonshot’s first major experimental project in this domain, Noisee was a resounding success in many ways. It was a platform where more than 10,000 users brought their creativity to life, generating over 5,000 unique and amazing music videos every single day at its peak. More than just numbers, Noisee vividly showcased the incredible power of community, the boundless joy of creation, and the universal human need to tell stories.
The true impact of Noisee is best expressed through the voices of our community, and here are just a few that capture the spirit of what we built together:
I am immensely grateful for the dedication of the team who poured their hearts into this project, and for our wonderful, vibrant community of “Noiseers” who made the platform what it was. This one-year journey with Noisee was a profound reminder that when you provide intuitive tools designed to help people tell their stories, the desire to create and connect is truly limitless.