Show HN: Create short form videos on any topic using Generative AI
topic2clip.streamlit.appHey HN,
Long time lurker first time posting. This is a streamlit app I wrote as a side project. It's a Generative AI webapp for creating short 30 second informative videos on any topic or subject under the sun. It's not perfect by any means, but a cool demo.
It takes around a minute to generate so please be patient. Email is optional, only if you want link to watermark removed completed video clip.
Any feedback/suggestions would be appreciated.
Thanks! JD
P.S. I am actively looking for Senior Python/DevOps/MLOps roles in California. Contact info in my profile. Thanks.
The tool shows promise and it's really fast. I generated a video transcript and the narration was good. However, some images were contextually incorrect or seemed like half-baked AI-generated images. This will be much better if the following can be done - Issues with image generation - Add some transition - option to add an avatar like https://www.synthesia.io - add a custom intro and outro
Thanks for the feedback. Added those to the backlog. Unfortunately this app has pushed Streamlit to its limits. It works as intended as a MVP, but the next version architecture will be more scalable.
Whoops, crashed for a bit. Now up again. Please comment if it goes down. Also, for those asking for the tech stack:
- OpenAI Instruct API for generating script
- SDXL model for generating images
- ElevenLabs API for narration
- Streamlit for hosting and UI
Not working for me at the moment
Fixed! Got hit with spammers. Email is required for now.
Incredible did one about beef tallow and honestly might be cool to use this to learn concepts depending on how complex can break down topics
Hey try again with a similar topic, made some changes for better quality text and image generation.
Example “Some stunning facts about the universe” https://topic2clip.streamlit.app/?vid=7549e327-89f3-478d-9dc...
Thanks for trying it out! Yeah its very basic as is, but in future I would want to create a multimodal model that can handle complex topics.
Quite cool, but won't allow you to use iCloud Hide your Email, which is a bit annoying.
Weird. Can you check if its getting caught in email filter by going to https://regex101.com/r/y8xh8k/1 and typing the email under TEST STRING?
Ah, got it.
In [.-_] the hyphen makes it so it matches chars between . and _. It needs to be rewritten with the hyphen at the end so it doesn't treat it as that: [._-]
Thanks! Fixed. Btw the app is now hosted on https://topic2video.streamlit.app
Cool stuff!! I just tried creating a info video for my app and came out better than I expected.
Yeah, thats an interesting use-case. For now its just merging the SDXL images using ffmpeg and adding narration. For intro videos it would need cooler transitions. Maybe in the next version.
Great work, if you open source it I'd love to help contribute if I can!