InspireFlow

An AI-powered creative workflow assistant for creators.

InspireFlow (升变) is a project built by a team of high school students during AdventureX 2026. It explores how AI can help creators manage the journey from an initial idea to a finished piece of content, while keeping their creative assets organized and under their control.

What are we building?

Creating content often involves scattered ideas, unfinished drafts, collaboration messages, brand requirements, and production plans across different tools.

InspireFlow aims to provide a unified workflow where creators can:

• Capture and organize creative ideas
• Transform rough thoughts into structured content plans
• Assist with outlines, scripts, and production workflows using AI
• Keep track of creative assets and collaboration history
• Create verifiable records for important creative materials

The goal is to make the creative process more organized, transparent, and easier to manage.

Why InspireFlow?

Many creators spend significant time developing ideas, writing drafts, and communicating with collaborators. However, these valuable creative assets are often stored across disconnected platforms, making it difficult to track ownership, changes, and the evolution of an idea.

InspireFlow explores a different approach: creators should have better visibility and control over the creative process behind their work.

How it works

InspireFlow combines several technologies:

• AI assistance — helping creators structure ideas and improve workflows
• On-device processing — designed to keep sensitive creative information local whenever possible
• iOS application — providing a native experience for creators
• Hardware interaction — integrating with devices such as smart rings for easier idea capture
• Blockchain-based records — exploring ways to create verifiable records for creative assets

Current Status

InspireFlow is currently a prototype/demo developed during AdventureX 2026.

The project was built within a short hackathon timeframe, including:

• A complete iOS demo application
• AI-assisted creative workflows
• Hardware integration experiments
• A prototype for creative asset verification

There is still a lot to improve, and we are continuing to explore how AI can better support creators.

Vision

From the first spark of an idea to the final delivery, creators should have a clearer and more connected workflow.

InspireFlow is an experiment toward building tools that help creators focus on creativity while maintaining better control over their creative process.

I’ve been working on the AppleMusic-Downloader project, primarily debugging a Windows-specific issue where the desktop app would hang during downloads. The root cause turned out to be PyInstaller’s --windowed mode setting sys.stdout and sys.stderr to None, which broke async libraries like httpx and asyncio. The fix was calling AllocConsole() on Windows to allocate a visible console window and properly redirect all three standard streams (stdin, stdout, stderr). I also cleaned up unnecessary debug code (log file output, /api/logs endpoint), added step-by-step logging in the download pipeline to help diagnose future issues, fixed a few i18n key mismatches in the frontend, and added a note in both READMEs telling Windows users not to close the console window.

The Dolby format finally downloaded normally. During this stage, copilot helped me debug and test the project.This project is ready to ship

published to PyPi!

workflow currently running!!!

All the rebuild work is complete; it’s time to open the PR.

The interface is ready to use!

I felt this didn’t quite meet my aesthetic standards for the sidebar, so I decided to use Figma to design the UI.

Successful integration of the front-end and back-end.

Remove Windows-specific GUI components and associated files from the project, including CMake configurations, resource files, and source code for the Flutter window implementation. This cleanup streamlines the project structure and eliminates unnecessary dependencies for non-Windows platforms.

I ultimately gave up on using Flutter for frontend development; it was simply too difficult. Instead, I am using Next.js combined with pywebview to build the desktop frontend.

I have completed the API construction; in this phase, I encapsulated all of gamdl’s parameters—aligning them with the definitions in core_downloader.py—into a unified API. Next up is learning Flutter for the frontend.

Rebuild starting now!!! I’m using flutter for fronted FastAPI for backend

All preparatory work for the refactoring has been completed. Thanks to DeepSeek for sorting out this “spaghetti code.”

seems that I can’t fix the atmos download issue : ( but time to rebuild this project, I just made a TODO list

repair downloader_song_legacy.py

This conversation tracked the comprehensive refactoring and bug-fixing of the AppleMusic-Downloader project. The work began by diagnosing why CN-region songs failed with a KeyError on extendedAssetUrls — fixed by switching to .get() with a fallback, and adding a webplayback-based stream path for songs without enhancedHls. We then extracted the entire download orchestration from cli.py (a Click-based CLI) into a new core_downloader.py module — a pure Python API with no terminal dependencies — and rewired both cli.py and download_worker.py (the GUI’s QThread) to call it directly, eliminating a class of Windows GUI crashes caused by Click’s stdin/stdout assumptions. We fixed download_worker.py’s variable name bug (remux_path vs remuxed_path) that silently dropped song files, normalized Path/str type handling, added CENC stream support (ISO-23001-7 with data:;base64, KIDs) by reconstructing Widevine PSSH boxes and passing real KIDs to mp4decrypt, switched ffmpeg remux from -f ipod to -f mp4 and added automatic fragmentation detection with fallback to aac_at/aac re-encoding for playable output, and added backward-compatible config migration for the renamed codec_song enum. We also fixed the Windows manifest name, updated the PyInstaller .spec with missing hidden imports, and corrected several type annotation issues across models.py, downloader.py, and downloader_song.py. The result is that both standard Widevine songs (e.g., ROSÉ’s toxic till the end) and CENC-encrypted songs (e.g., Queen’s Under Pressure via the 28:ctrp256 webplayback asset) now download and play correctly across all platforms.

really thanks deepseek help me to rebuild the interface
I updated the regular expression to account for Apple changing the location of the index.js file on the Apple Music web interface following WWDC2026. After removing redundant code, I used AI to add cross-platform compatibility support, moving closer to a full project refactor once the holidays are over. I also revamped the UI, implementing a card-stacking layout to make the logs easier to read. Huge thanks to AI—it has made fixing baffling bugs far more efficient.