HN Explorer

19 comments

[ 4.2 ms ] story [ 82.0 ms ] thread
crowwork
This project brings stable diffusion models to web browsers. Everything runs inside the browser with no server support. Please check out our GitHub repo to see how we did it. There is also a demo which you can try out.
sberens
> As WebGPU does not support FP16 at this moment, the memory consumption of running the demo is about 7GB. For Apple silicon Mac with only 8GB of unified memory, it may take longer (a few minutes) to generate an image. This demo may also work for Mac with AMD GPU.

Ouch. It will be interesting what this number will look like in 2-3 years though.

crowwork
on apple M2max, it takes 20sec, and with the fp16 support, there will be opportunities for improvement likely soon this eyar
sberens
I'm looking forward to an OpenAI quality embedding that I can use locally. I think eliminating the latency/cost from calling their API will allow for some novel UX patterns.
crowwork
The current demo uses CLIP model from openai, which is likely what you are looking for
sberens
Oops I meant text embeddings
cyx6
This is an AMAZING work! Just type in one line description and click, I get a photo by stable diffusion on my macbook in minutes. And it never needs any setup, just having fun. :)
l3jin
Cooool! We can use Stable Diffusion models on chrome.

It can generate any image I want within 2mins:)

jinhongyii
Great work! It's exciting to see such a large model running on my local web browser.
junrushao1994
This actually works on my non-M1/M2 macbook with AMD GPUs pretty smoothly. It's a bit slow for now (~70sec), but I expect this to be faster with fp16. Any roadmap to support fp16 on WebGPUs?
crowwork
FP16 is already in the spec and hopefully will ship this year we believe
ziheng
Cool project. The main point I guess is this doesn't require user to do any environment setup. All your need to do is just run it in the browser.

One thing I am curious, how's the performance of WebGPU approach compared to the local server approach?

crowwork
On the current setup(M2 and metal), it should run as fast as the local native environment.
zxybazh
Very cool demo, got it working on M1 macbook, generates each pic in 140s. Fun experience without specific setup.

Does it also work for cuda device? Is there any plan to support other models soon?

misakamisuzu
Really impressive! Very smooth experience on chrome.
yongwww
Great project! It is able to run on my local Mac M2, the generated result looks good to me! Looking forward to seeing more models are enabled
wuweil
super cool! it also works on my MacBook with AMD GPU. this is a great demonstration of the power of web GPU
yecomb
huggingface would save a lot of money if they know this :)