Reform AI

Transform content in seconds.

Comment

Inspiration:

From my 2 years of content creation experience, I learned one thing. Time is your most valuable resource. Making a video isn't just exhausting. it's extremely time consuming. In content creation, many creators often "copy" or take "inspiration" from an existing video that went viral, and switch it up minimally to gain the same results of virality. To find the source clip, make your own script, generate your own ai audio, add your own captions manually is not only really expensive and complicated, but it is extremely time consuming. As someone who has done this, my primary goal was to construct something that could solve this. I've heard of websites such as Crayo.ai before, which is a social media video generator, but I've never heard of a full, multi- repurposing tool. This tool's primary goal is to help content creators, clippers, be easy to use, save money, and most importantly, save time.

What it does

Reform AI has three main functions, an Audio Reformer, Video Reformer, and a Source Finder. The user uploads an audio file of the video they want to 'reform', and they click reform. The AI will transcribe the audio given, and the user has the ability to edit it if there are any mistakes. Then, it will give a transformed, reformed script which will ideally be a social media optimized paraphrased version of the original script. The user can then chose an AI voice to speak the script out and download that mp3 file. After this, the user can upload a screenshot of the clip from the original video, to find the raw source clip used. The AI will scan the internet to find potential links with the same clip, often from TikTok or YouTube. Then, it'll open up cobalt.tools which is a social media link to mp4 downloader. This part couldn't directly be implemented into the app because of YouTube and TikTok's privacy guidelines. With the audio and video, the user can upload both in the video editor, and it'll add animated auto captions which the user can then export.

How we built it

To build this, I used a mix of vibe coding + manual coding + APIs. For the full stack development of this app I used bolt.new. As for the API keys, I used Assembly's API key for the audio to text transcription, Open Router's Claude Sonnet model for the rephrase AI (which uses a specific social media retention based prompt to get the write reform), Open AI Whisper for the transcription + auto captions, ffpmeg for the exporting. I used figma to make my own UI designs, then I used builder to export it as code and used several methods of prompt engineering + tokenization to get the right prompts into bolt.new. This helped my project construction as I could get the desired output in one go while saving tokens.

Challenges we ran into

Some challenges I ran into were token usage, database issues, and prompting. To fix token usage, I used a combination of chatgpt + gemini + claude's chatbots to debug issues and prompt it in a way that it wouldn't cause any errors, and I also made an error handling system which wouldn't break the flow of the app even if there were an error. With this effective prompting, I watched several prompt engineering videos and google courses and learned how to prompt effectively which solved 2/3 challenges. To overcome the database issues, i tried a variety such as GCP, AWS, and finally Supabase. Supabase was the final one I decided to use as it's extremely cheap, simple to use, and can handle the perfect amount of data of users for the website.

Accomplishments that we're proud of

Some accomplishments that I'm proud of include creatively brainstorming this idea, getting it approved my the content creation market (reached out to big creators and clippers), and the fact that I developed this all on my own by spending less than $10 on apps.

What we learned

I learned several skills such as marketing, coding + vibe coding, designing, API keys, software development, agent creating, and many more. These skills will truly help me in the future.

What's next for Reform AI

There is a lot of things to add in the future, and fine tuning will surely take some time. Some of these features are better audio voices, better video editing, and more software. Originally, this app was made to take payments from stripe for member ships but as this is a Hackathon it is not included.

Built With

  • api
  • assembly
  • bolt.new
  • claude
  • elevenlabs
  • ffpmeg
  • node.js
  • openai
  • openrouter
  • react
  • stripe
  • supabase
  • whisper
Share this project:

Updates