This is really cool. I wonder if you could use it to generate images instead of this SSH message:
> The authenticity of host 'mint.phcomp.co.uk (78.32.209.33)' can't be established.
RSA key fingerprint is 6a:de:e0:af:56:f8:0c:04:11:5b:ef:4d:49:ad:09:23.
Are you sure you want to continue connecting (yes/no)? no
That way you could check the fingerprint for the server visually. Of course you'd first have to have a way to deterministically generate a description of an image from a sequence of random bytes.
I can’t explain exactly why or how but I feel like this would have an inherent weakness that we are not able to comprehend yet. It would greatly reduce the threshold for cryptographic collisions as the human brain is not good at telling two images apart on the pixel level like a hashing algo would be.
Hello HN! I'm the author of this post and one of the co-creators for this project. We put this out as a fun community project to track prompt engineering in large Transformer models (in this case CLIP) and to showcase our Gradio panel integration. Thank you all for the support!
We were not expecting to make it to the front page, and I think our tiny community server is not able to keep up with all the requests. All your prompts are logged to a queue and processed by a GPU worker (currently just a single T4). The prompts that made it to this project will eventually be generated by the model.
I've seen this method with the brush strokes and it's cool but I don't think it's there yet. The Dall-E type stuff that's coming up is exciting but the open source versions still need some work. I found this site a month ago or so and I'm a big fan of it: https://www.artbreeder.com/
7 comments
[ 4.6 ms ] story [ 32.8 ms ] thread> The authenticity of host 'mint.phcomp.co.uk (78.32.209.33)' can't be established. RSA key fingerprint is 6a:de:e0:af:56:f8:0c:04:11:5b:ef:4d:49:ad:09:23. Are you sure you want to continue connecting (yes/no)? no
That way you could check the fingerprint for the server visually. Of course you'd first have to have a way to deterministically generate a description of an image from a sequence of random bytes.
We were not expecting to make it to the front page, and I think our tiny community server is not able to keep up with all the requests. All your prompts are logged to a queue and processed by a GPU worker (currently just a single T4). The prompts that made it to this project will eventually be generated by the model.
We will work on a solution to handle the number of requests we're getting. In the mean time you can check out the Colab version of this project here https://colab.research.google.com/drive/16nR8_Ida5tID1RSzLy1...
Thank you again for checking out our project! We will post updates soon!