Legend.video is your open source AI director. Describe your concept, it will make a video with Gen AI. You can try it at Legend.video.
From a simple prompt (e.g. "San Franciso romcom"), Legend.video will first generate a storyboard with scenes consisting of individual shots. You can then edit the storyboard, generate still images for each shot, and then generate a video from the still images. The output is a video that tells a story.
Current video models generate 2-4 second clips which fail to tell a story. Legend.video helps you build a story from these short 2-4 second shots.
- Next.js App Router, Tailwind CSS, shadcn/ui
- Supabase for OAuth (via Github and Google) and Postgres DB
- gpt-4o for generating storyboards (but this can be replaced with gpt-3.5-turbo and any model that supports function calling)
- text-to-image and image-to-video models on Replicate for generating still images and videos
- We're hosting https://legend.video/ on Vercel, but you could easily run it elsewhere too.
- We're using Stripe to charge users for additional generation credits after the initial 200 free credits every user receives.
Copy the .env.example into a local .env.local file to set up the various environment variables and secrets: We're using Supabase for storage, Vercel for blob storage, Github and Google for login, and Replicate for AI generation.
$ cp .env.example .env.local
DB storage: You'll need to start a Supabase project and preload the DB schema from docs/db_schema.sql.
Vercel Blob storage: We're using Vercel blobstore. You can create a blob store on their free plan and fill in the BLOB_READ_WRITE_TOKEN variable.
Login: You can create a GitHub OAuth token in your settings. For Google, you'll need to start a new project and create OAuth keys.
Replicate: We're using Replicate for their hosted AI models. Sign up for a key here, and copy your token into REPLICATE_API_TOKEN.
Then run these commands to install and dependencies and start the server:
$ pnpm install
$ pnpm run dev
You should now be able to see the app running at localhost:3000.
If you'd like help with any of the following, get in touch:
- Keeping styling consistent between shots - we've made some progress with this, but it's not perfect
- Keeping characters consistent betweent shots - we've experimentd with prompting methods for this, but we need a more sophisticated solution
- Sound - we'd love help with adding sound to the videos
- Edit by prompting - Allow users to edit their stories ChatGPT style.
- UI Refinements - We'd love help with making the UI more intuitive and user-friendly
- New models - If you have access to newer and better video models that you'd like to make accessible throgh Legend.video, let us know. We'd love to give it a try.
Please report issues at the Github Issues page.
If you'd like to contribute bugfixes or new features, please fork the repo, make your changes in a branch, and then submit a pull request. We'll review it as soon as possible and merge it in if it's good.