Year: 2022

@ThirdEyeXyz @rainisto 1/ No, not fully automated. Eventually, thanks to the progress we are making with computer vision, large language models, and synthetic audio, voice UIs will become viable. Like Star Trek TNG “Computer” viable.

Once you have that, you can have a dialogue with your computer…

@ThirdEyeXyz @rainisto 2/ …similar to the one you have experienced recently with #ChatGTP. It won’t be just you saying “do this” but
You: do this
AI: done
You: ok, what about changing x
AI: sure
You: nice. Now let’s make this other change…

@ThirdEyeXyz @rainisto 3/ So the activity that Roope is showing in his original tweet, will (hopefully) become a dialogue between human and AI that collaborate to bring an idea to life.

For that, you first need a speech-to-text AI that translates your utterance into written instructions…

@ThirdEyeXyz @rainisto 4/ …then you pass that written instruction to a large language model to “interpret” and translate into commands for your media editor of choice, either via API calls or other means…

@ThirdEyeXyz @rainisto 5/ …then that media editor (A1111, InvokeAI, Photoshop, Krita, Gimp, Blender) will invoke the latent diffusion model (or what will come next) to execute the commands.

And all of this will be an interactive back and forth like you would be talking to a human designer.

@ThirdEyeXyz @rainisto 6/ You can certainly automate certain workflows. For example the InvokeAI team is working on a node editor, similar to the one in Blender, that will allow you to create and customize a generative pipeline. And that will help studios streamline production…

If you are generating #stablediffusion images (or want to start) and never tried #InvokeAI, you are really missing out.

This video shows you the incredible power of #GenerativeAI and the astonishing quality of this open source software:
https://t.co/FWpEM5zGIO

The only thing #InvokeAI is missing is support for Stable Diffusion 2.x, which will come with version 2.3 as I understand it.

Hopefully, @lincolndstein and team will release it soon.

@camenduru @huggingface I’m getting “https://t.co/65qTv3jfjh refused to connect.” since yesterday.

(I didn’t replicate it yet)

@a__spaceman Yes. Unfortunately, for now, it only generates 256x256px images. However:

https://t.co/nQRDMIKCJB