Hotshot was an iOS app & website that let you make personalized generative AI images & videos.
I was a Full-Stack Software Engineer for Hotshot at Natural Synthetics from November 2022 to November 2024.
See my full resume here.
VentureBeat: Hotshot launches new text-to-video AI generator
Tom’s Guide: Hotshot is a new free text-to-video platform — and it’s very good indeed
TechRadar: Hotshot is heating up AI video-making – here’s how you can try it for free
MosaicML: Natural Synthetics: Customer Spotlight
Youtube: Hotshot - Is This the Most Realistic AI Video?
Hotshot had 4 main product “eras”:
Hotshot was a free iOS app that let you generate pictures of you and your friends.
Using fine-tuned Stable Diffusion models, Hotshot would generate personalized images at 512x512, and later did upscaling to 1024x1024.
On sign up you would scan your face or upload a few photos and Hotshot would learn what you looked like in a matter of minutes. Using text-to-image prompts and a selection of pre-made styles, Hotshot could generate pictures of you and your friends doing anything you can imagine!
GitHub | HuggingFace |
Hotshot’s first text-to-video model: open source and built on SDXL.
Hotshot-XL generates 1s, 8 fps GIFs at 512x512 resolution.
Trained to generate 1 second, 8 fps videos at 512x512 resolution.
Paper | Video |
Follow up to Hotshot-XL, but scaled up. Features higher resolution, improved coherence, and better pop culture knowledge.
Hotshot ACT 1 generated 3s, 8 fps videos at 768 resolution.
Paired with an updated website featuring an all-new design and streamlined user experience.
App | Website | Technical Report | Model Evals |
A state-of-the-art, large-scale text-to-video diffusion transformer model. Built from the ground up by a team of 3 over 4 months.
Hotshot ACT 2 generates up to 10s, 24 fps videos at 1024 resolution.
It excels in prompt alignment, consistency, and motion, while being highly extensible to longer durations, higher resolutions, and additional modalities.