-
-
Submission form
-
Finished reels :p
Inspiration
Short-form video has become the dominant format of the internet. The presence of this content has increased by 183% the last two years, and global short-form video ad spending has surpassed $111 billion in 2025!
However, most people still lack the technical skills, editing experience, and tools required to produce these videos.
I’ve had countless moments, especially as a busy college student, where I had a fun or creative idea but didn’t have the time to turn it into a polished video. I’ve also seen friends around me struggle with the same thing: great ideas stuck in notes apps and their brains because production feels overwhelming.
Seeing that gap between ideas and execution, I wanted to build something that removes that friction, and helps storytellers create entertaining content and bring their stories to life.
What it does
Syntale transforms written story ideas into fully structured short-form videos, complete with images, a storyline and script, voiceovers and subtitles.
Users simply input a story idea and customize elements such as creativity level, the number of scenes, the voiceover, background music, language, and more. Upon submission, Syntale breaks the story into structured scenes and crafts a story for it, generates corresponding videos for each scene, adds voiceover narration, creates synced subtitles, and formats everything vertically into a video ready to be posted on any social media platform.
Within 2 minutes, the result is a downloadable, platform-ready short video with a full fleshed out story.
How we built it
Frontend:
- React
- Tailwind <3
"Backend":
- Uses a webhook to trigger a workflow on Make.com
- A super complex Google Sheet database for lightweight state-management
- OpenAI ChatGPT for script and prompt generation,
- Freepik classic for image generation
- Azure for voice generation
- Third party API for combining all of it into a video.
Challenges we ran into
The biggest challenge was orchestrating multiple asynchronous AI services into a single deterministic pipeline.
With only realistically 3-4 hours to make this project (I slept through the morning then screwed around during lunch lol), I decided to make some very informed engineering decisions. I didn't want to go through the struggle of creating and deploying a whole backend, so I spun up a Google Sheet for my database and then found this No-Code Workflow service online called Make.com which allows me to connect a bunch of different APIs together. Learning and configuring the workflow automation under time constraints was definitely non-trivial. Furthermore, the learning curve with learning the platform and setting up everything on the site was difficult and involved more clicking on my screen than I expected.
Another challenge was trying to not burn through a bunch of money on the tokens. I had to do a good amount of searching online for model options that offered free tokens, and came a third party API which uses Azure and Freepik very sparingly so I can save some cash. I explicitly decided to not host it online so someone can't screw with it after I publish this project and spend all my tokens, so if you want to test it you have to run it locally (pls don't burn through my tokens still though).
Accomplishments that we're proud of
I was able to generate this beautiful video story with my own project revolving around one of the ACM Hack Interns going to get Acai (he is allergic to Salpicon Acai): https://youtube.com/shorts/of_eaoerF4Y
I also now have my own little custom fully automated end-to-end brainrot AI content pipeline. The system successfully generates vertical social media ready content in under 2 minutes. I can directly customize it to my liking to actually create a video I want (unlike where other platforms restrict you to their schema). It also uses significantly less tokens than any other platforms online that offer free trials. My system is token-efficient, allowing dozens of full video generations within free-tier limits.
What we learned
Generative AI is very powerful, but orchestration is pretty tedious. There's a lot of friction when orchestrating many different components... Many of these models are impressive on their own, but the real value comes from being able to combine them to create a whole different product.
I also learned that making something that makes brainrot content makes me want to make my brainrot :p
What's next for Syntale
- Use a proper database so I can have user accounts and store the reels individually to users
- More customization presets
- Direct posting to social media platforms
- Different choices of models
- Prompt engineering experimentation to see what results in the best outputs
Log in or sign up for Devpost to join the conversation.