QwenChaNa Medias

QwenChaNa Medias turns a rough idea into a narrated film—research, script, storyboard, visuals, voice, and editing handled by one resumable virtual film crew.

Comment

Inspiration

Alibaba Cloud offers a wide range of models, each with different strengths. Instead of forcing one model to do everything, we wanted to combine the best tool for each part of the filmmaking process. That idea became QwenChaNa Medias: a small autonomous film studio built from specialized models.

What it does

QwenChaNa Medias turns a simple creative brief into a personalized narrated video. It coordinates language, vision, voice, video generation, and editing tasks to research an idea, write a script, plan scenes, generate visuals, create narration, and assemble the final cut.

How we built it

We hosted the application on Alibaba Cloud ECS and connected multiple Qwen-powered services:

  • Qwen-Plus for planning, research, scripting, and coordination
  • CosyVoice for narration
  • Wan2.1 T2V Turbo for video generation
  • FFmpeg for final media assembly
  • LangGraph for orchestrating the agent workflow and managing dependencies between tasks The workflow is designed to save progress as it works, so a job can resume after a quota limit or failed generation step instead of starting over.

Challenges we ran into

The biggest challenge was developing with limited model credits. Even after payment succeeded, our Qwen Code token plan did not work as expected. The free model quotas also ran out frequently during development, sometimes halfway through a task. That forced us to make the system more resilient. We added incremental asset persistence, resumable jobs, configurable models, and local fallback media so we could keep testing without repeatedly spending credits.

Accomplishments that we're proud of

We pushed through the quota and tooling limitations to build a complete working demo, not just an isolated model experiment. QwenChaNa Medias can coordinate an end-to-end production pipeline and was successfully hosted on Alibaba Cloud ECS. We are especially proud that the system combines several specialized models through LangGraph into one coherent creative workflow.

What we learned

We learned that the most interesting part of multimodal development is not simply calling a model—it is designing the handoffs between models. LangGraph was new to us, and it helped us understand how complex agent workflows can be represented as connected, stateful steps.

Qwen’s video and voice capabilities were particularly exciting, especially when combined with language planning to create a process that feels more like a real production studio.

What's next for QwenChaNa Medias

Our next focus is more refined editing. We want to support richer visual effects, tighter timing, transitions, improved scene composition, and more creative control over the final video. Ultimately, we want QwenChaNa Medias to feel less like a generator and more like a collaborative film desk.

Built With

Share this project:

Updates