Research and fine-tuning
Self-host a repo when you need to change the model.
AI Video Generator GitHub Repos, Side by Side
Well-known AI video generators on GitHub with public weights include Wan2.1, CogVideoX, Open-Sora 2.0 and Mochi 1, and each README lists its license and the graphics memory it needs. Wan2.1's 1.3B text-to-video model needs 8.19 GB of VRAM and makes 5 seconds of 480P video in about 4 minutes on an RTX 4090. Reviral runs hosted models in the browser instead, and a 5-second Wan 2.6 Flash clip costs 21 credits (~$0.26).
No credit card required to try — sign in with email or Google.
- License, output size and GPU memory per repo
- Numbers read from each README on 21 October 2026
- Hosted Wan models in the browser if you have no GPU
- Price shown before every render
Four GitHub video models, from their READMEs
| Repository | License | Output | GPU memory |
|---|---|---|---|
| Wan2.1 (weights 25 Feb 2025) | Apache 2.0 | Text and image to video, 480P and 720P | 8.19 GB for T2V-1.3B |
| CogVideoX (2B, 5B, 1.5-5B) | 2B: Apache 2.0 · 5B: CogVideoX License | 720 x 480; 1.5-5B: 1360 x 768, up to 10 s | From 4 GB (2B) to 10 GB (1.5-5B) with optimizations |
| Open-Sora 2.0 (11B, 12 Mar 2025) | Apache-2.0 | 256px and 768px | About 60 GB at 768 x 768 on one GPU |
| Mochi 1 (10B) | Apache 2.0 | 480p | About 60 GB on one GPU; one H100 recommended |
Read from each site's own page on 21 October 2026: Wan2.1 on GitHub · CogVideo on GitHub · Open-Sora on GitHub · Mochi on GitHub
What running one yourself takes
You need a graphics card with the memory in the table, a Python setup, and room for the model weights. The smaller models fit a home GPU; the 10B and 11B models expect data-centre cards.
Open-Sora's README says its 2.0 model was trained for $200K, and Wan2.1's lists text-to-video, image-to-video, video editing, text-to-image and video-to-audio among its tasks.
Read from each site's own page on 21 October 2026: Open-Sora on GitHub · Wan2.1 on GitHub
Hosted instead of self-hosted
With no GPU, Reviral's Cinema studio runs hosted models in the browser. Wan 2.6 Flash makes 5, 10 or 15 seconds from one photo for 21 credits (~$0.26) per 5 seconds; Wan 3.0 runs 2 to 30 seconds for 99 credits (~$1.24) per 5 seconds.
Ways to use this tool
Offline work
Run locally when footage cannot leave your machine.
Quick results
Use a hosted model when you just need the clip.
Comparing models
Try several hosted models on one prompt in Cinema.
How it works
Step 1
Read the README
Check license, GPU memory and supported sizes.
Step 2
Match your hardware
Pick a model your graphics card can hold.
Step 3
Or open Cinema
The Try button opens Wan 2.6 Flash in the browser.
Step 4
Render and compare
See the charge before each hosted render.
What it costs
A 5-second Wan 2.6 Flash clip at 1080p costs 21 credits (~$0.26).
Changing the video's length, media type, or resolution can change the final price.
Dollar figures use the $4.99 Mini Pack rate. Cached for up to 1 hour.
Frequently asked questions
- Is there a free AI video generator on GitHub?
- Yes. Wan2.1, CogVideoX-2B, Open-Sora 2.0 and Mochi 1 are published under Apache 2.0, and CogVideoX-5B under its own CogVideoX License, so the code and weights cost nothing to download. You pay for the hardware and electricity to run them.
- Which GitHub video model runs on a home PC?
- The small ones: Wan2.1's 1.3B model needs 8.19 GB of VRAM, and CogVideoX-2B starts at 4 GB with its memory optimizations turned on. Mochi 1 and Open-Sora 2.0 need about 60 GB.
- Can I use these models without a GPU?
- Yes, with a hosted studio that runs them for you in the browser: on Reviral, a 5-second Wan 2.6 Flash clip costs 21 credits (~$0.26).
More in AI Video Models
Compare closely related tools, or browse the complete AI Video Models topic hub.
AI Video Generator From Audio: Song or Voice InUpload an MP3 or WAV and get moving pictures that follow it, with or without a photo.Cheap Image to Video AI: Live Prices Per ClipThree low-priced photo-to-video models, priced live per clip, with no subscription.AI Based Video Generator: How the Models WorkText, a picture or a reference clip in; a video model renders the motion; you pay per second or per clip.Image to Video AI with Audio: Photo to Sound ClipUpload a photo, describe the motion and the sound, and get a clip with audio from a model that makes sound.Lyria API: Google's Music Model, Prices and LimitsWhat Lyria is, how Google offers it through the Gemini API and Google Cloud, what each call costs, and how to put music under video.How does AI video generation work?Cheap image to video AIAutomatic AI video generator