Viridian's AI Experiments
-
- Posts: 1590
- Joined: Wed Apr 15, 2009 10:03 am
Re: Viridian's AI Experiments
Another AI photo set. I'm rolling some of my process back. The problem with progression is that AI photography is a little TOO good, especially when using AI upscaling and Hires Fixer tools. The fault is that they fix all the imperfections that might photography realistic, whereas the current AI models are getting better at including blemishes,such as irregular faces, skin moles and sweat. The use of film grain prompts also adds a fuzzy look that AI removes to make perfect studio photos.
You do not have the required permissions to view the files attached to this post.
Viridian @ deviantART: http://viridianqs.deviantart.com/
- cerberus
- Posts: 482
- Joined: Thu Apr 16, 2009 7:34 pm
Re: Viridian's AI Experiments
They're going to displace a lot of mud, aren't they! Nice fun set.
Cerberus
Johnny Dowd: Be content with your life. It may not get any better!
Johnny Dowd: Be content with your life. It may not get any better!
-
- Posts: 1590
- Joined: Wed Apr 15, 2009 10:03 am
Re: Viridian's AI Experiments
Bringing some old sets up to date. Here's a comparison through the year.
You do not have the required permissions to view the files attached to this post.
Viridian @ deviantART: http://viridianqs.deviantart.com/
-
- Posts: 1590
- Joined: Wed Apr 15, 2009 10:03 am
Re: Viridian's AI Experiments
I went on a massive update spree and remastered virtually all of my AI images. Here's a set of five.
You do not have the required permissions to view the files attached to this post.
Viridian @ deviantART: http://viridianqs.deviantart.com/
-
- Posts: 1590
- Joined: Wed Apr 15, 2009 10:03 am
Re: Viridian's AI Experiments
Another updated set.
You do not have the required permissions to view the files attached to this post.
Viridian @ deviantART: http://viridianqs.deviantart.com/
- Theo
- Posts: 698
- Joined: Sat Feb 13, 2021 4:12 am
- Location: Sinking Fantasyland
Re: Viridian's AI Experiments
Missed these somehow, nice sets Viridian!
Finally broke down and went to see a psychologist. She told me I just have an overactive imagination—and it really excited her. We're going sinking next weekend. Theo's AI Quicksandbox New stuff every weekend (unless life gets in the way)
-
- Posts: 1590
- Joined: Wed Apr 15, 2009 10:03 am
Re: Viridian's AI Experiments
A slight update to one my tools has made a few more things possible. Here are some re-imagined versions of the traps that Susan has triggered in her desert expeditions.
As a bonus, RunwayXL has made significant updates to their tools, and it seems the rendering software is getting better at maintaining cohesion.
As a bonus, RunwayXL has made significant updates to their tools, and it seems the rendering software is getting better at maintaining cohesion.
You do not have the required permissions to view the files attached to this post.
Viridian @ deviantART: http://viridianqs.deviantart.com/
- sixgunzloaded
- Posts: 922
- Joined: Tue May 05, 2015 8:16 pm
Re: Viridian's AI Experiments
Love these! Far more realistic than anything 3D can do. It's difficult to pick a favorite here, so I won't! I'll just say "Thank you" and enjoy them all!
When you say, 'maintaining cohesion', are you saying it's essentially able to do a consistent story line now? With the same character and setting and everything?
When you say, 'maintaining cohesion', are you saying it's essentially able to do a consistent story line now? With the same character and setting and everything?
How long did Tarzan watch before deciding to save Jill..?
-
- Posts: 1590
- Joined: Wed Apr 15, 2009 10:03 am
Re: Viridian's AI Experiments
sixgunzloaded wrote:When you say, 'maintaining cohesion', are you saying it's essentially able to do a consistent story line now? With the same character and setting and everything?
No, at least not yet. Its not yet possible specifically train one character model on Stable Diffusion tools. You have to deal with the "close enough" effect. You also can't use key frames to make the AI animation tools smoothly blend in motion.
By "cohesion", I mean the output has less of a tendency to morph, which was a huge problem. The entire scene would suddenly transform into a completely different scene because the AI picked up the wrong visual cues, so someone's face might turn into a grotesque monster, or an arm might turn into a plus-sized woman. RunwayXL has expanded its custom motion brushes so that you can isolate specific sections and control their direction and scale, which also largely prevents the unwanted morphing.
Viridian @ deviantART: http://viridianqs.deviantart.com/
- cerberus
- Posts: 482
- Joined: Thu Apr 16, 2009 7:34 pm
Re: Viridian's AI Experiments
Viridian wrote:sixgunzloaded wrote:When you say, 'maintaining cohesion', are you saying it's essentially able to do a consistent story line now? With the same character and setting and everything?
No, at least not yet. Its not yet possible specifically train one character model on Stable Diffusion tools. You have to deal with the "close enough" effect. You also can't use key frames to make the AI animation tools smoothly blend in motion.
By "cohesion", I mean the output has less of a tendency to morph, which was a huge problem. The entire scene would suddenly transform into a completely different scene because the AI picked up the wrong visual cues, so someone's face might turn into a grotesque monster, or an arm might turn into a plus-sized woman. RunwayXL has expanded its custom motion brushes so that you can isolate specific sections and control their direction and scale, which also largely prevents the unwanted morphing.
This is the same for SD as well. I suppose you could train a LORA for a specific face and use that in the prompt to get consistency, but I'm not sure how well that would work. At least with the EpicRealism model you do get quite a lot of consistency with faces. Clothing, well, that's a different matter, I think you did well to get those three images so consistent.
It sounds as though the animation tool you are using has AnimateDiff at the back of it, I experienced the same problem you've been getting. In Automatic1111 you can see what is going on though. AnimateDiff will generate a max of 32 frames, the problem is, it will use one seed for the first 16 then change to a different (but close) seed for the second 16! I can't find a way to stop this and suspect it may be deliberate to stop people doing too much with it. The other issue I found was that you could find a nice seed to use as a starting point and specify that, but it wouldn't use it. I read somewhere that it takes your seed and picks one within about 100 of it to use.
All very frustrating. Still, I'm watching your results with interest, at some point I may wish to switch to SDXL, but at the moment I've still got a lot to experiment with on SD and don't want to change too much at once.
PS: I sent you a private message, if you are like me you don't get many so may not have noticed.
Cerberus
Johnny Dowd: Be content with your life. It may not get any better!
Johnny Dowd: Be content with your life. It may not get any better!
Who is online
Users browsing this forum: No registered users and 1 guest