Quinn

1-shot Text to Video, Now 1-shot Text to Interactive Video as well. Fully composable, use Alibaba Cloud models for QWEN, Happy Horse and WAN. Use SoTA public adapters for other parts of the pipeline.

Comment

Inspiration

The inspiration began over three years ago with a simple image-slideshow video tool. Since then, Samsar.One has evolved into a state-of-the-art, model-agnostic text-to-video agent capable of producing videos up to three minutes long.

What it does

Samsar.One creates coherent, one-shot text-to-video experiences with lip sync, consistent characters, and unified themes. Its composable pipeline integrates Alibaba Cloud models, including Qwen 3.7/3.8, Wan 2.7 Pro, and Happy Horse 1.1.

Quinn extends Samsar.One through an A2A connector, allowing other agents to create accounts, purchase credits, submit video requests, monitor renders, and retrieve completed videos.

How we built it

I have been building toward Samsar.One for more than three years, creating and sharing demos at every stage of its development. The entire journey is chronicled on my YouTube channel.

This hackathon gave me a unique opportunity to bring the latest pieces together: native Alibaba Cloud model integration, one-shot interactive video, and Quinn’s A2A interface for agent-driven video creation.

Challenges we ran into

Qwen’s capabilities differ across model variants. Text-only and multimodal requests must be routed to compatible models, which adds complexity to the orchestration layer.

We also encountered inconsistencies in Qwen response JSON, especially when requests passed through third-party adapters. With Happy Horse 1.1 image-to-video, images with a strong blue tint sometimes caused faces and teeth to appear blue during speaking scenes.

Accomplishments that we're proud of

  • A one-shot text-to-video agent that produces videos up to three minutes long with lip sync, character consistency, and theme coherence.
  • Native Alibaba Cloud integration across inference, image generation, and video generation.
  • Quinn, an A2A connector that makes Samsar.One accessible to other agents.
  • One-shot interactive video, the latest feature completed during this sprint.

What we learned

Qwen inference performance varies by provider. In our testing, native Alibaba Cloud inference was faster than inference routed through third-party providers.

We also learned that reliable multimodal generation depends on more than model quality: provider routing, response normalization, and careful handling of visual artifacts are equally important.

What's next for Quinn

Quinn is the A2A gateway for Samsar.One. Next, we plan to deepen its A2A interoperability and expose more Samsar.One workflows so agents can create, fund, monitor, and retrieve generative-media projects through a consistent interface.

Share this project:

Updates