Vango 🎨 - Build with + evaluate Stable Diffusion fast

tl;dr: Vango makes it 10x easier to build and evaluate Stable Diffusion workflows in your app. Developers use Vango to deploy Stable Diffusion in minutes and achieve high-quality images faster than any existing solution.

Hello world — Eshaan and Rajen here!

We’re thrilled to show y’all what we’ve been working on, let’s begin!

🐷 Problem:

We’ve all seen how unique, highly customizable, and beautiful AI art can be — but it’s incredibly tedious to deploy and maintain Stable Diffusion in your app.

Building apps on top of stable diffusion is an unnecessarily tedious and manual process. You have to:

• Select the right models, hyperparameters, and prompts: These have huge impact on the content, quality, and style of outputs. To achieve the best results, developers spend hours testing each model, parameter, or prompt individually.
• Add guardrails: Ensuring that your model does not generate unintended NSFW or discriminatory images takes hours of careful testing.
• Deploy on GPUs: Like everything in AI right now, developers must obtain and configure a juicy high-VRAM GPU to run Stable Diffusion fast.

💜 Solution:

Vango empowers developers to build and evaluate stable diffusion models faster than ever.

🧱 Build workflows & deploy as API

Watch us build a corgi generator and deploy it as a scalable API in seconds.

🤖 Perfect the hyper-parameters

Vango helps you evaluate models quantitatively. Vango will track metrics across thousands of prompts — including NSFW content, crossed eyes, disfigured limbs/fingers, cropped faces, and prompt alignment.

1. Specify your models (either as .safetensors or .ckpt file, Cog container, API endpoint, or Vango workflow)
2. Provide us your prompts or use one of our prompt datasets with thousands of examples.
3. Use our quantitative features to learn how models compare

And for more granular analysis, you can filter all the images based on failure modes. In the video below, you can see us finding the images that have deformed hands.

🐻 About us:

We’re best friends, neighbors back home in Maryland, and college roommates (and literally sharing a triple in SF right now)! We’ve been doing class projects + side projects together for over 3 years now 🙉.

Eshaan has researched multimodal models at Berkeley AI Research (BAIR), low-compute visual detection systems at NASA JPL, and anomaly detection at Citadel Securities. And Rajen has built wide-ranging software at Amazon, Nuro, Two Sigma, and Solana.

And we both graduated from UC Berkeley a year early to work on Vango! We discovered this problem while developing our own diffusion apps after noticing first-hand how AI can help creatives in VFX.

📩 Asks:

If you are (or know anybody that is) building apps with stable diffusion, we’d love to chat and see how we can help! We’re currently onboarding our early partners.

If you’re interested, sign up for our waitlist here or shoot us an email at founders@vango.ai.