3 ms·
You know, I've been trying to use some of these new generative AI and LLM models and I think I'm sort of reflecting off of them. I've gotten a few good things o
by bane 3y ago
You know, I've been trying to use some of these new generative AI and LLM models and I think I'm sort of reflecting off of them. I've gotten a few good things out chatGPT in terms of writing advice and such, but I find the generative art bits really frustrating, and typing in long prompts to be also kind of not worth the effort, I can usually just make what I want in less time than it takes to refine my prompts down.
However...I'm a very visual thinker, when I'm thinking or speaking, I often see images in my head of what I'm trying to convey.
I wonder if aligning ChatGPT and DALL-E or something similar so that I can "see" an image of what the computer is saying as well as the text might be a great next step towards making me feel more engaged with the technology.
That and native speech to text would be nice so I can just talk at it and have it just sitting on the side as a helper-bot instead of being the main point of my focus while I work on or do other things.
- jonplackett 3y agoI’m also excited if they can make an image generator you can chat to and iterate. That would be awesome. You might know Thai already but have you tried using ControlNet to make images with stable diffusion? It allows you to input an image that it follows along with the prompt, and there’s a bunch of ways to influence it. You can even give it a hand drawn sketch. Or have it pick out position of limbs, or use the edges of objects and keep those. If you have something specific you want to create then it’s amazingly helpful. The only easy to use site I know of that offers it is happyaccidents.ai (or you can run yourself if you have SD installed)
- bane 3y agoThanks for the tip! I'll be checking it out. This conversation reminds me of the old Star Trek:The Next Generation episodes in the Holodeck, only they're talking to "the computer" to iterate on a holodeck scenario design. https://youtu.be/vaUuE582vq8 https://youtu.be/vaUuE582vq8 https://youtu.be/NXX0dKw4SjI https://youtu.be/NXX0dKw4SjI
- mcintyre1994 3y agoIt's quite a bit more narrow, but FusionAI is a really cool example of this too! https://www.producthunt.com/posts/fusionart-ai https://www.producthunt.com/posts/fusionart-ai
- abound 3y ago> but I find the generative art bits really frustrating, and typing in long prompts to be also kind of not worth the effort I think this might just be inherent in the domain - the state space for images is so much larger than it is for text, so there's just a lot more ways to interpret a text prompt. Sane "defaults" help, but it might just be inherently true that it takes a longer prompt to get close to what you're seeing in your head.
- bane 3y agoThey do seem to really struggle with prepositions...at least for what I've been trying.