Reimagine Reality: An AI-Powered SaaS Platform for Video Transformation
Vidia is a Software-as-a-Service (SaaS) web application designed for AI-powered video transformation, emerging from Federico Arboleda's prior "ExMachina" research. The vision was to make complex generative video techniques, primarily leveraging the ComfyUI engine, accessible to filmmakers and content creators through an intuitive web interface. Recognizing that ComfyUI's power came with significant intricacy, Federico embarked on building a complete product, leading core product strategy, AI workflow design, frontend architecture, custom ComfyUI node development, Docker containerization, and critical backend/infrastructure.
The project aimed to bridge the gap between cutting-edge AI capabilities and practical usability, transforming a complex technology stack into a polished, user-friendly tool. Vidia allows users to harness the power of generative AI for video without needing to understand the complex underlying systems. The platform is accessible at vidia.tools.
Democratize complex AI video transformation by creating an intuitive SaaS platform powered by ComfyUI.
Approx. 10 months (June 2024 - April 2025)
Federico Arboleda: Lead Product Strategist, AI Workflow Architect, Lead Frontend Developer, Core Backend & Infrastructure Engineer. Collaborator: Andrés Daza (specific infrastructure tasks).
Vanilla JavaScript, HTML, CSS, Cloudflare Workers, Cloudflare R2, RunPod (Docker, ComfyUI), Magic Link, Resend, Git, Custom ComfyUI Nodes.
Developing Vidia involved navigating a complex landscape of emerging AI technologies, intricate system integrations, and the inherent difficulties of building a full-stack application largely single-handedly. Key challenges included:
Federico's approach was characterized by deep technical engagement, persistent problem-solving, and pragmatic decision-making, leading to innovative solutions:
Visualizing Vidia's Architecture: Mapping the communication flow between frontend, backend, and AI services.
The development of Vidia was a phased journey, marked by iterative refinement, deep technical dives, and strategic adaptation to challenges. Federico led this process, driving the project from initial concept to a functional application.
The initial phase focused on building the Vidia frontend UI and establishing reliable communication with a ComfyUI backend running locally on Federico's Mac. This allowed for rapid prototyping and AI workflow development. After designing the UI, Federico implemented HTTP requests and WebSocket connections for task initiation and real-time progress monitoring. A custom ComfyUI node was created to serve Vidia's frontend, enabling a self-contained local development environment.
Early concept: Original hand-drawn wireframe for Vidia's interface.
This phase centered on implementing Vidia's core video generation modes (Trace, Evolve, Forge) and advanced feature toggles (LoRA, Detailer, etc.). This required dynamic modification of ComfyUI workflows. Federico designed the frontend logic to load a base workflow and modify node parameters. Crucially, he abandoned brittle manual JSON manipulation for feature toggling, instead leveraging ComfyUI's native conditional nodes. The frontend toggled inputs to these nodes, allowing ComfyUI to reliably manage execution paths. UI feedback systems were also established to address early UX concerns about clarity and wait times.
Detailed planning: Chart listing and describing every planned feature for Vidia.
The focus shifted to containerizing the backend with Docker and deploying to RunPod cloud GPUs, supporting both Pod and Serverless modes. Federico led the demanding Dockerization and RunPod integration. Andrés Daza collaborated on cloud storage uploads (initially S3, then Cloudflare R2) and secure credential handling. A key event was the Google Makeathon (Oct 2024), where Federico demoed Vidia, received positive validation from Google designers, and networked successfully. The most significant blocker was resolving RunPod Serverless path/volume issues, where ComfyUI couldn't find models/nodes. After extensive debugging, Federico implemented a solution (Feb 2025) by dynamically configuring ComfyUI to recognize multiple volume paths, a massive breakthrough.
Federico presenting Vidia at the ComfyUI SF Meetup hosted at GitHub HQ (Jan 2025).
With cloud deployment achieved, the next challenge was ensuring reliable execution of Vidia's complex AI workflows via the RunPod serverless endpoint. Federico focused on debugging the RunPod handler and the custom `VidiaVideoSaver` ComfyUI node. This involved fixing an infinite loop in the handler caused by mismatched WebSocket signals, addressing API payload discrepancies between the frontend and the RunPod wrapper, and resolving a bug where the frontend sent empty inputs for disabled features, causing workflow failures. The `VidiaVideoSaver` node was refactored to integrate direct R2 uploads, watermarking, and audio muxing.
The final pre-launch phase involved implementing remaining features, refining the frontend, establishing user authentication, and polishing the UI. Federico managed the complexity of numerous features through rigorous modular design. For user authentication, he independently implemented a Magic Link system for frictionless email verification and login, integrating Resend for email delivery and Turnstile for bot protection. Mobile optimization was initiated, separating CSS and JS for mobile-specific enhancements. Vidia soft-launched at `app.vidia.tools` on April 13th, 2025, followed by a period of stabilization and debugging based on initial user feedback.
Federico's approach throughout the Vidia project was characterized by visionary leadership, deep technical immersion, persistent problem-solving, and pragmatic decision-making. This led to several key breakthroughs:
Throughout Vidia's development, Federico actively engaged with the tech and AI communities, seeking feedback, sharing progress, and building connections. These interactions provided valuable validation and motivation.
Federico connected with key figures in the AI space, including ComfyAnon (creator of ComfyUI) and Vojta Jína (ML engineer at Apple), at the ComfyUI SF Meetup in January 2025. These interactions provided insights and fostered a sense of community.
At the USC Google Makeathon in October 2024, Federico demoed Vidia to Google engineers and product managers. The positive feedback on Vidia's intuitive UI and impressive capabilities was a significant validation point. Read more on the Vidia Blog: Vidia and Google.
Vidia was publicly announced and demonstrated at the ComfyUI SF Meetup hosted at GitHub HQ (January 2025) and later at the LA Meetup (March 2025), generating interest and contacts from companies like Netflix. Read about the announcement on the Vidia Blog: ComfyUI Meetup Announcement.
Vidia's development culminated in a functional SaaS application, successfully translating complex AI research into a user-facing product. The soft launch marked a significant milestone, demonstrating the viability of the platform and its potential to democratize advanced video AI tools.
Beyond the numbers, Vidia achieved several important qualitative benefits:
"The Makeathon was an incredible opportunity... Having Google engineers experience Vidia firsthand and provide their insights has already sparked new ideas for enhancing the user experience."
— Federico Arboleda, reflecting on the USC Google Makeathon
The Vidia project was an immense learning experience, pushing the boundaries of what was thought possible with emerging AI technologies and lean development. It underscored the power of persistent problem-solving and the importance of a clear product vision.
Vidia's journey from a research-inspired concept to a functional SaaS platform was a testament to the power of focused dedication and deep technical engagement. It reinforced my belief that complex, cutting-edge technologies can be productized and made accessible through thoughtful design and relentless problem-solving. This project solidified my passion for building tools that empower creators and demonstrated the immense potential at the intersection of AI, cloud computing, and user-centric product development. The ability to navigate and overcome significant undocumented technical challenges largely independently was a profound growth experience.
