Building Noosa Studio: A Node-Based Workflow for Agentic AI Video Generation with Qwen Cloud
Instead of prompting AI one video at a time, what if creators could build an entire AI production pipeline?
That simple question became the starting point of Noosa Studio.
Noosa Studio is a node-based workflow platform for agentic AI video generation, allowing creators to design their own AI production pipeline instead of relying on a single chatbot prompt.
Our goal isn't simply to generate videos.
Our goal is to help creators build an AI-powered production team.
The Problem
AI video generation has become incredibly accessible.
Today, almost anyone can type a prompt and generate a video within minutes.
However, creating consistent content is still extremely difficult.
Content creators don't just need one video.
They need dozens.
Sometimes hundreds.
Each video needs:
- a good concept
- a complete storyboard
- consistent characters
- quality control
- video generation
- storage
- distribution
Most AI tools treat this as one giant prompt.
We wanted something different.
Introducing Noosa Studio
Instead of one AI doing everything,
Noosa Studio divides the production process into multiple specialized agents.
Each node represents one role inside a virtual film studio.
Users can freely connect, rearrange, replace, remove, or inspect every node.
The workflow becomes visual, transparent, and fully customizable.
Think of it as building your own AI production pipeline instead of repeatedly writing prompts.
Workflow Templates
To make the platform accessible to both beginners and advanced users, Noosa Studio provides several workflow templates.
- Basic — a lightweight workflow for beginners.
- Precise — designed for more structured content generation.
- Advanced — our complete production workflow.
Users can also extend these templates using optional add-ons.
Storytelling
Adds a Text-to-Speech node for narrated videos.
Character-Centric
Adds an Actor Node for character consistency.
This allows creators to choose how simple or advanced they want their workflow to be.
Every Node Has A Job
One of the biggest design decisions we made was splitting the workflow into specialized AI agents.
Instead of relying on a single LLM to solve every problem, each node focuses on one responsibility.
Input Node
Receives prompts directly from the Studio.
Telegram Trigger
Starts the workflow whenever a prompt is received from Telegram.
Actor Node
The Actor Node is responsible for character references.
Users have three options:
- Upload their own character image.
- Describe the desired character using text.
- Leave it empty and let AI generate a brand-new character automatically.
Producer Node
The Producer develops the overall story concept and genre.
Instead of immediately generating a storyboard, it first defines the direction of the content.
Writer Node
The Writer transforms that concept into a complete storyboard and splits it into multiple scenes.
Reviewer Node
The Reviewer is one of our favorite agents.
Unlike other nodes, it doesn't generate content.
Instead, it evaluates outputs produced by previous agents.
If the quality isn't good enough, it automatically asks the responsible node to regenerate until the output passes review.
This creates an automatic quality-control loop inside the workflow.
Video Node
Generates the final video using Wan video models.
Cloud Node
Stores generated assets.
Telegram Node
Delivers finished videos directly to the user's Telegram chat.
Telegram As Both Input And Output
Most AI platforms require users to open a web application every time they want to generate content.
We wanted something more convenient.
Noosa Studio allows Telegram to become both:
- the input interface
- and the output interface
After connecting a Telegram Bot Token and Chat ID, users can choose between three integration modes.
- Disabled
- Output Only
- Full Telegram Mode
In Full Telegram Mode, every prompt sent from Telegram automatically starts the workflow, and the generated video is delivered back into the exact same conversation.
No browser required.
Character Consistency
Maintaining consistent characters across AI-generated videos is still one of the hardest problems today.
To address this, our Character-Centric workflow combines multiple AI models.
The Actor Node first generates a reference character using Qwen Image Plus whenever the user doesn't provide one.
Next, a Vision model extracts important visual attributes such as appearance, clothing, hairstyle, colors, and other defining features.
These extracted attributes are converted into structured text and injected into the storyboard generated by the Writer.
Meanwhile, the original reference image is passed directly to the video generation model.
By combining both textual attributes and image references, we significantly improve character consistency throughout the final video.
Why Multiple AI Agents Instead Of One?
This was one of the biggest architectural decisions behind Noosa Studio.
Rather than asking a single LLM to perform every task, we intentionally split responsibilities across specialized agents.
Each agent only focuses on one job.
This approach makes the workflow:
- easier to control
- easier to debug
- more flexible
- more token-efficient
- easier to customize
- and often produces better output quality
Since every node is independent, users are also free to rearrange their workflows according to their own production needs.
AI Models
Different problems require different AI models.
Instead of forcing one model to do everything, Noosa Studio selects the most suitable model for each responsibility.
Producer, Writer, and Reviewer use Qwen language models such as Qwen3.5 Flash and Qwen3 Max, chosen for their balance between quality and token efficiency.
The Actor Node uses Qwen Image Plus together with a Vision model.
For video generation, we combine Wan 2.7 Text-to-Video and Wan 2.7 Reference-to-Video.
Text-to-Video is used for prompt-based generation, while Reference-to-Video is used whenever character references are available.
This combination helps maintain visual consistency while keeping the workflow flexible.
Alibaba Cloud ECS
Video generation is a long-running task.
Instead of deploying everything on a serverless platform, our AI worker runs on Alibaba Cloud Elastic Compute Service (ECS).
Running the worker on ECS allows long-running video generation jobs to execute reliably while keeping the frontend lightweight.
This architecture also enables one of our most exciting upcoming features.
Upcoming Feature: Fully Automated Content Generation
Imagine configuring a Studio only once.
Set the niche.
Describe your channel.
Build the workflow.
Choose your schedule.
For example:
Every day at 6:00 AM.
Every morning, the worker automatically starts the Studio, generates a brand-new video based on the stored channel niche and description, and delivers the finished result directly to Telegram.
No prompts.
No manual work.
Just wake up and receive fresh content that's ready to upload.
This feature is currently planned as our next major milestone after the hackathon.
Transparency Over Black Boxes
One thing we wanted to avoid was turning AI into a black box.
During execution, every node exposes its own status and output.
Users can inspect:
- generated concepts
- storyboards
- reviews
- extracted character attributes
- generated images
- generated videos
This makes debugging significantly easier while allowing users to understand exactly how every AI agent contributes to the final result.
A Simple Use Case
Imagine waking up in the morning with only your phone nearby.
Instead of opening a laptop, launching a browser, and configuring multiple settings, you simply open Telegram.
Need a 15-second video?
Type a command.
Need Mandarin instead of English?
Update the settings directly from Telegram.
Then send your prompt.
A few moments later, the completed video arrives back in the same conversation.
The entire production workflow happens behind the scenes.
Looking Forward
Noosa Studio is more than an AI video generator.
It is an experiment in making AI production pipelines visual, collaborative, and autonomous.
By combining specialized AI agents, node-based workflows, Telegram integration, Qwen Cloud models, Wan video generation, and Alibaba Cloud ECS, we hope to make video creation feel less like prompting an AI—and more like directing an entire production studio.
We're excited to continue improving Noosa Studio and can't wait to see where this journey leads next.
Thank you for reading!
Top comments (0)