As an amateur AI video creator, very much looking forward for testing these new video models! Only tested Runway so far. Would be fun to try out Sora when available in EU. Veo2 also seems great!
The best technology I discovered today was someone showed Gemini Flash watching him work in Blender and telling him what to do or write Python scripts. I couldn't figure out what tools he used.
In your opinion what is the ideal configuration for working with ComfyUI locally. I am clear about the GPU, RTX 5090 which I hope to release in January but what do you think about the processor and RAM? Thanks for all the information and tutorials.
some of the models, like LTX can already be used fast on 12GB. Also Animatediff creates stunning results on 1.5 models with video to video and motion loras
Idea for a very cool video : Real Photographic Faces of a real person and making flux lora training on that person, all best practices , and best apps to get most photo and most real .... i dont think there is much around, besides a hand full on celebs on civitai. i was given some of their top makers settings to such real great celebs flux training on civitai, but my results all failed miserably, when suing their settings.... its still such a mystical environment....maybe you can un-mystify this?
I know we are in different paths a bit. You have said Comfy UI is where it is at for local gen, but I tried installing that and it didnt work. so i am stuck on Forge, but I do wonder if there is a Forge way of experimenting with Video gen. Maybe you could poke around and see if there is something I have missed for forge? Thanks
please join my discord, linked under the video and ask in the comfyui chat for help. comfyui is selfcontained. so it should work basically out of the box
in the meantime maybe look at all the sales that are going on on different platforms or look into selfhosting a jupyter nodebook (which is a learningcurve but cheap)
@@OlivioSarikas Thank you for the response. The reason I want to run locally is because I keep coming up against censorship. I do VFX for a lot of R rated content and can't afford to get stuck on a platform shutting me down in the middle of a job. I'll definitely look into it, though. I appreciate the recommendations.
It seems the Channel is decreasing in quality. It used to be about techniques for generation of great images, installations, etc.. now this is just marketing clips. I miss your great tutos on SD and MJ…
I was impressed with the video physics engines but I find it hard to believe prompts to AI will ever achieve it. I argued with 01 about how it knew the ball rolling down a ramp would end up in the middle bucket. It obviously had no real answer. I then projected a line from the end of the ramp to the edge of the canvas to show it than even if the ball had 0 velocity it would fall beyond the middle bucket. It still argued that it could end up in the middle bucket. I said only by bouncing into/against the other bucket first. If anyone has AI prompts that can predict balls rolling down ramps I'm interested. AI can easily generate the physics.
I'm not sure about that. AI models like LLMs work based on highest probability. So if you train a model on physics simulation instead of language, i can see how that could work, since physics are math and probability
@@OlivioSarikas I am trying a different approach - but which is equally problematic. I am using coco-ssd to determine the positions of objects in a photo and then AI to look at it and describe them. Joining the two together is problematic. But also getting it to label them. I spent about 24 hrs going around the same loop. AI failed to work it out. In the end it apologised to me. Weird.
AI users in the EU are in a bad place with how the EU is so anti-AI in their law-making (And worst is the law-makers in EU is not even elected but appointed)
With Musk in charge of the US, OpenAI will get a kicking. They should have relocated to avoid Musk controlling them. Musk's revenge on OpenAI may actually bring about the end of it. He could end up controlling Facebook and Google when he has his hands on all the levers of power and persuasion.
EU has said that they won’t miss the opportunity for control over AI as they did when the internet broke. Europe never misses the chance to miss a chance.
It's not anti-AI more pro artist/creator. Trying to make the case that artists/creatives should at least have their consent for their work to be used and to be (possibly) paid for that. I know a lot of companies and people don't give a damn about that though...
Don't worry in the long run it will be useful...soon people will be bored of AI just like they got bored of online classes...because physical will come back...
Thank you for the mention Olivio! Great video 😊
@@Visually_AI you are welcome. Love your content
Excellent reviews! Thanks, Olivio!
Thank you very much :)
As an amateur AI video creator, very much looking forward for testing these new video models! Only tested Runway so far. Would be fun to try out Sora when available in EU. Veo2 also seems great!
check out replicate website, they have a lot of the models on there and you only pay per use
@@OlivioSarikas thx, so many options, only which I had more time, would be great fun to test the different options though
Finally. .we wait for this kind of content
So much news I can't keep up! Thanks for keeping us updated. :)
You are welcome :) This isn't even everything.
Great round up 👍 👌
Thank you very much
Google veo-2 is another level !
Now all we need is good ideas for awesome plots :)
Very nice content Olivio:) Keep it going!
Thank you so much :)
You never realize how long it's been since you had good pancakes until you see pancakes. Now I want nothing but pancakes.
right? I love pancakes soooooo much
Lol i just asked you for text to video ai few hours ago and you made a a video , BTW love you content man, always easy to follow
thank you very much. I put a lot of work into this video :)
The best technology I discovered today was someone showed Gemini Flash watching him work in Blender and telling him what to do or write Python scripts. I couldn't figure out what tools he used.
really? that sounds very cool!
In your opinion what is the ideal configuration for working with ComfyUI locally. I am clear about the GPU, RTX 5090 which I hope to release in January but what do you think about the processor and RAM?
Thanks for all the information and tutorials.
Hey :3 How often will you do an episode? I like the new style btw 👍
thank you. not sure yet. This is a first try :)
When do you think video AI will be easily available locally for those of us with only 12 GB of RAM?
some of the models, like LTX can already be used fast on 12GB. Also Animatediff creates stunning results on 1.5 models with video to video and motion loras
Do we have 3d modeling AI yet? Character \ Creature\ 3d objects?
Yes, Meshy can do that
How to running dual GPU/VGA in comfyui?
Спасибо!
Idea for a very cool video : Real Photographic Faces of a real person and making flux lora training on that person, all best practices , and best apps to get most photo and most real .... i dont think there is much around, besides a hand full on celebs on civitai. i was given some of their top makers settings to such real great celebs flux training on civitai, but my results all failed miserably, when suing their settings.... its still such a mystical environment....maybe you can un-mystify this?
I know we are in different paths a bit. You have said Comfy UI is where it is at for local gen, but I tried installing that and it didnt work. so i am stuck on Forge, but I do wonder if there is a Forge way of experimenting with Video gen. Maybe you could poke around and see if there is something I have missed for forge? Thanks
please join my discord, linked under the video and ask in the comfyui chat for help. comfyui is selfcontained. so it should work basically out of the box
Still waiting for the email since 3 Days, not in junk also!
Please contact them on their discord. They seem to have some technical issues
My old video card has got me feeling a bit behind. Hoping I can upgrade to the 5000 series in 2025.
in the meantime maybe look at all the sales that are going on on different platforms or look into selfhosting a jupyter nodebook (which is a learningcurve but cheap)
@@OlivioSarikas Thank you for the response. The reason I want to run locally is because I keep coming up against censorship. I do VFX for a lot of R rated content and can't afford to get stuck on a platform shutting me down in the middle of a job. I'll definitely look into it, though. I appreciate the recommendations.
Thanks for covering Whisk, I've been obsessed with it lately!
I wish I had access to it already. But it's country locked
@@OlivioSarikas Invoke 5.5 Desktop с flux tool canny depth
review?
👋 hi
Hi :)
nice transparent shirt
lol, good thing i'm not Kermit, or you couldn't see me at all ;)
I've already submitted a video to Project Odyssey. How do you want me to tag you on the video?
It seems the Channel is decreasing in quality. It used to be about techniques for generation of great images, installations, etc.. now this is just marketing clips. I miss your great tutos on SD and MJ…
I was impressed with the video physics engines but I find it hard to believe prompts to AI will ever achieve it. I argued with 01 about how it knew the ball rolling down a ramp would end up in the middle bucket. It obviously had no real answer. I then projected a line from the end of the ramp to the edge of the canvas to show it than even if the ball had 0 velocity it would fall beyond the middle bucket. It still argued that it could end up in the middle bucket. I said only by bouncing into/against the other bucket first.
If anyone has AI prompts that can predict balls rolling down ramps I'm interested. AI can easily generate the physics.
I'm not sure about that. AI models like LLMs work based on highest probability. So if you train a model on physics simulation instead of language, i can see how that could work, since physics are math and probability
@@OlivioSarikas I am trying a different approach - but which is equally problematic. I am using coco-ssd to determine the positions of objects in a photo and then AI to look at it and describe them. Joining the two together is problematic. But also getting it to label them. I spent about 24 hrs going around the same loop. AI failed to work it out. In the end it apologised to me. Weird.
AI users in the EU are in a bad place with how the EU is so anti-AI in their law-making (And worst is the law-makers in EU is not even elected but appointed)
With Musk in charge of the US, OpenAI will get a kicking. They should have relocated to avoid Musk controlling them. Musk's revenge on OpenAI may actually bring about the end of it. He could end up controlling Facebook and Google when he has his hands on all the levers of power and persuasion.
EU has said that they won’t miss the opportunity for control over AI as they did when the internet broke. Europe never misses the chance to miss a chance.
The EU needs to be abolished.
It's not anti-AI more pro artist/creator. Trying to make the case that artists/creatives should at least have their consent for their work to be used and to be (possibly) paid for that. I know a lot of companies and people don't give a damn about that though...
Don't worry in the long run it will be useful...soon people will be bored of AI just like they got bored of online classes...because physical will come back...