Seed Music Pro v2.0: Community, Social, and a New AI Music Agent

Two weeks ago, we shipped a model upgrade. This week, we're shipping something much bigger.

Seed Music Pro v2.0 introduces a community layer and an AI music agent — two features that change how you create, discover, and share music on the platform.

Community & Social (v2.0.0)

Until now, Seed Music Pro has been a solo experience. You generate a song, download it, and that's it. v2.0 changes that with a full community layer.

Explore

The new Explore page is a feed of music created by the community. Browse fresh songs, discover playlists curated by other users, and follow creators whose work you enjoy. You can search by song title or artist name.

Playlists

Create playlists to organize songs around a mood, a project, or a theme. Open any playlist and it plays through continuously — no manual skipping needed.

My Works

Your own songs, liked songs, and bookmarked songs are now in one place. No more digging through different pages to find something you saved last week.

Notifications

When someone likes your music, bookmarks it, or starts following you, you'll get a notification. It's a small thing, but it makes the platform feel alive.

Music Agent (v2.0.1)

This is the feature we're most excited about.

The Problem We Saw

We noticed something in our backend data: some users were generating the same song 20, 30, even 50 times with the exact same prompt. They were hoping that if they kept trying, eventually the model would produce something they liked.

The issue wasn't luck. The prompt itself wasn't communicating what they actually wanted.

We tested this manually — took a user's prompt, improved it, and regenerated. One try. They got exactly what they were looking for.

That's when it clicked: if we could give users a way to talk through what they wanted, the AI could handle the prompt engineering for them. No more guesswork, no more "gacha" rolls.

How It Works

The Music Agent is a conversational interface. You describe what you want — even vaguely — and the agent turns it into a finished song. If the result isn't right, you tell it what to change. It adjusts and tries again. The whole loop stays in one chat thread.

Start rough. Say something like "a chill lo-fi beat with a female vocal humming" and the agent generates it. You don't need to know model parameters or technical terms.

Iterate naturally. If the tempo feels wrong, say "make it slower." If the vocal style isn't right, say "try a deeper voice." The agent understands context from the conversation, so each revision builds on the last one.

Reference existing songs. Use @ to bring back a song you generated previously and continue working on it. Pick up exactly where you left off.

Compare versions. Generate multiple takes on the same idea, listen side by side, and keep the one that feels right.

Why This Matters

Most AI music tools treat generation as a one-shot event: you write a prompt, hit generate, and hope for the best. If the output doesn't match your vision, you start over from scratch.

The agent approach is different. It treats music creation as a conversation — because that's how most people actually think about music. You know the feeling you want, even if you can't describe it in technical terms. The agent bridges that gap.

What's Next

These two updates are just the beginning of v2. We have more planned for the coming weeks, including improvements to the agent's understanding of genre and arrangement, and new community features.

Try the agent at /app/chat, or explore the community at /explore.