June Recap: What’s New at OctoML
Product launch, Stable Diffusion Fine Tuning, ⚡️Fast Whsiper and a Sneak Peek at an Upcoming Collab 👀
June was a big month for us at OctoML 🎉
We launched our brand new compute service, OctoAI, which helps developers run, tune, and scale models so they can focus on building great AI applications. We want you to have easy access to efficient, scalable AI compute, with the freedom to build how you want.
OctoAI Compute Service Launch
We are so grateful for the support of early access customers, who provided valuable feedback that shaped the development of OctoAI, and our launch partner, AWS. In case you missed it, here’s a replay of our live virtual launch event, and a fun video from our IRL launch party at the AWS Loft in San Francisco.
Since launch, we have event more exciting updates to share:
Huge Whisper Cost Savings
Whisper is a popular speech transcription model that turns audio files into text. We’re excited to share that running Whisper on OctoAI for audio transcription costs 6X less than on OpenAI, based on internal benchmarks using OctoAI’s A10G tier. More detailed benchmarking of Whisper on OctoAI is in progress, so stay tuned for a full latency and cost analysis coming soon!
Two New ControlNet Models
If you work with Stable Diffusion, you already know ControlNet, which uses a prompt + image framework to give greater control over your image generations. This week, we released two new ControlNet endpoints in OctoAI: Canny and M-LSD. Each works a little bit differently (Canny uses edge maps, while M-LSD uses straight line detection) so go give them both a try at octoai.cloud.
ControlNet Canny in action, teasing an upcoming collab 👀
Run Automatic 1111 in OctoAI
Automatic 1111 is one of the most popular web UI available for customizing Stable Diffusion, including features like txt2img, img2img, Inpainting, Prompt Matrix, and SD Upscale. Now you can quickly spin up an Automatic 1111 instance directly in OctoAI. Read this tutorial to learn how in 5 minutes.