Stable Diffusion AI Image Generator

I seem to have completely befuddled the Bing image creator by asking for space marines.

It's been 10 minutes. It's still throwing "angazi, boss" at me.
 
saor said:
Snipped this pic from the Adobe Firefly article. What's the best way to go about doing something like this at the moment? Feed an image and get color / material iterations of that image. Are the latest IMG2IMG tools suitable?

View attachment 1496599
Click to expand...
Multi-ControlNET in combination with IMG2IMG is your best friend for this. You might not even need IMG2IMG with the right prompting you could probably get it done with Multi-ControlNET, and TXT2IMG.

Edit... low effort attempt in TXT2IMG with Multi-ControlNet (Canny and Depth models) and it came out as expected. It would require a bit more effort to get better textures on the bands and some variations on the graphics on the face, but nothing more than 30 mins or so of work.

xyz_grid-0000-2857421283-RAW photo of a smartwatch with a blue band.png
 
Last edited:
OnlyOneKenobi said:
Multi-ControlNET in combination with IMG2IMG is your best friend for this. You might not even need IMG2IMG with the right prompting you could probably get it done with Multi-ControlNET, and TXT2IMG.

Edit... low effort attempt in TXT2IMG with Multi-ControlNet (Canny and Depth models) and it came out as expected. It would require a bit more effort to get better textures on the bands and some variations on the graphics on the face, but nothing more than 30 mins or so of work.

View attachment 1496715
Click to expand...
Or even better - combine it with Latent Couple to place things around the canvas with fairly accurate precision.
 
Fjorko said:
Or even better - combine it with Latent Couple to place things around the canvas with fairly accurate precision.
Click to expand...
Latent Couple hasn't worked too well for me, not really an issue though as I usually just composite multiple subjects into an image with Photoshop anyway.
 
saor said:
Click to expand...
Why would you not be able to copyright AI art?
AI is just a more advanced rendering engine, you can copyright 3D renders from blender why not AI?
In both situations you feed the program parameters and then it spits out an image according to its programming.
 
airborne said:
Why would you not be able to copyright AI art?
AI is just a more advanced rendering engine, you can copyright 3D renders from blender why not AI?
In both situations you feed the program parameters and then it spits out an image according to its programming.
Click to expand...
I don't think you're comparing apples with apples. Rendering something in Blender is a LOT more complex than just typing something into an AI generator, especially if you have to do the modeling, texturing, lighting, and everything else. If you just mean downloading a .blend file from the internet and rendering it, I'm not sure you can actually copyright that.
As the video says, if you take elements from an AI render and re-work it into something else, then you CAN copyright the picture itself but perhaps not the character or the design depending on the specifics of what you are doing.
For me, where there's a bit more grey area in terms of AI image generation,is cases where the final art isn't simply generated by (just) typing in a prompt and cherry picking the best looking pictures -- ie, if the (human) artist had to use several techniques such as control net, inpainting, outpainting, upscaling or whatever to produce the final result, then there's a bit more of a human hand in it and it should be copyrightable because the human who produced it actually had to do more than just type a sentence for the AI.
 
Last edited:
This is a pretty darn awesome way to combine LLMs and Automatic1111 to generate images or even play "choose your own adventure" games. The possibilities of creating your own stories and worlds with just a few words and clicks are virtually endless.

 
OnlyOneKenobi said:
This is a pretty darn awesome way to combine LLMs and Automatic1111 to generate images or even play "choose your own adventure" games. The possibilities of creating your own stories and worlds with just a few words and clicks are virtually endless.

Click to expand...
Speaking of LLMs. I just started playing with Oobabooga last week.

Quite easy to use, if a bit slow.

I was considering starting a thread.
 
Vorastra said:
Speaking of LLMs. I just started playing with Oobabooga last week.

Quite easy to use, if a bit slow.

I was considering starting a thread.
Click to expand...
Yeah, surprisingly slow compared to an image generator like Auto1111 which seems to do more, as far as utilising a GPU is concerned. I don't really find these chatbots to be particularly useful though. I suppose once more plugins are available then it would be a different matter.
 
OnlyOneKenobi said:
Yeah, surprisingly slow compared to an image generator like Auto1111 which seems to do more, as far as utilising a GPU is concerned. I don't really find these chatbots to be particularly useful though. I suppose once more plugins are available then it would be a different matter.
Click to expand...
I haven't played with image generation in a while. I hear there's been some large speed improvements with the release of PyTorch2?
Reinstalled Windows about 2 months ago. Haven't had the patience to reinstall everything lol
 
Vorastra said:
I haven't played with image generation in a while. I hear there's been some large speed improvements with the release of PyTorch2?
Reinstalled Windows about 2 months ago. Haven't had the patience to reinstall everything lol
Click to expand...
I haven't installed PyTorch2 yet... too many things that will break if it goes wrong. I'm fine with the speed SD is going at the moment, been playing with ControlNet 1.1 lately - some improved results to the old models and some new models to play with as well.
 
You must log in or register to reply here.
Back
Top
Sign up to the MyBroadband newsletter
X