I understand the curiosity around AI etc... But on this example given that the car is modeled so its very easy and much FASTER to put shader presets and render everything on UE5 while having 100% of consistency, including for splash n smoke, there are presets for that which work very well. For me it's a waste of time and not a gain, in addition to being completely unusable in production.
Hey! Thank you so much for the video, can I get the workflow? Lately I was working on the consistency, and I'm finding it very difficult to maintain it. Will be very helpful if I get the workflow.
Great stuff, thanks for the overview! I guess for the time being, you will stick with SDXL for all your CG related workflows rights? Not sure far the development is right now for controlNets, IPadapters, etc for SD3.
These are great, but as a novice, I feel you need to explain each node as the data travels and why you use it and where it plugs in. Feels like you are speaking to someone that knows everything you know already and just give detail about your specific settings. It would help if you explain why each node is used and especially the function of that node, otherwise it is just you showing where you plug what in without understanding the reasoning and thinking behind your workflow.
HI Great Video! What GPU do you have? 4090 I presume? If so does this make the difference for real? let say you want to upscale the "final" porche image at 5 K how long would it take do do it? I am trying to figure out how to upgrade my system:)
Thanks for putting the time in to create the tutorial it was really helpful. It's my first time using ComfyUI and it took me a couple of weeks to resolve issues with installation, and then to get a feel for what' going on. I thought I'd never get this scene working but got there in the end. It would be great to know what hardware you're using to better understand whats required to get decent processing speeds. I've got a Macbook pro with M3 and it's currently pretty slow.
I have a 3090 and 96 gigs of ram on a relatively older machine, building a new one with 4090 soon. also ComfyUI is getting more vram efficient. also depends on what nodes are you using, more controlnets and ipadapters the slower the render would be, generally speaking!!!
Looks incredible! Thanks for sharing the file, thats huge. Still working on making it run on my end, but looking forward to play with it. Really useful! Thanks
I appreciate all the workflows you share but unfortunately this end result is Ai blended morphed garbage. I can't see any specific use for it in any production besides the most abstract of clips. Love your houdini to ai workflows though
You should never judge things solely based on the end result you don't like. It's important to consider the process, the technology, and how you can leverage it to your advantage. Otherwise, you'll always be waiting for others to provide something useful, which is a lazy approach!
Really cool. I love the water splash and smoke. Was that purely prompt based result or did you guide it with any simple particle setups? The interior was the most stable and looked most realistic. Was that because that shot was the most stable (as in angle not changing?) or did you try different settings in animatediff vs the exterior car. A week or two ago a new node came out for comfyui to train a lora within comfy. Have you tried that yet? I've managed to train from Kohya but not getting the greatest results. really great test, man. Thanks, and keep em coming! inspiring.
Splash and smoke come from the text prompt only! As for the exterior, the reason is that the resolution of my source image is not enough for the line art to produce a very defined guide for the animation, and the car is moving very fast. I'm not trying to fix every problem since I know how fast technology is evolving, and there might be a solution for all of this in a month or less! That Lora trainer in Comfy is very basic and needs an upgraded PyTorch, which really messes up every other tool in Comfy, but it seems like a good, fast solution! lmk if you ever get it to work tho, Ill release a video for kohya, it was a pian for me too but got it to work!!
Hey, I gladly accept your offer to ask questions in the comments. I'm honestly quite confused because I don't understand this whole ComfyUI theme. I downloaded it with Pinokio and can use the basic form. If I understand correctly, I need modules for good results. For example, Animatediff. I wonder how safe it is to download the modules through the ComfUI Manager. Because every time I click on the link, it shows that the module has been downloaded 0 times in the last month. Isn't that a security risk? Is there any way to see more of you in detailed tutorials? Patreon? Anything? Because I get the impression that you are becoming a pioneer in this topic, which is being far too under-addressed.
Great CG-Comfy tutorial. Is there a way to increase the accuracy of the generated image using the wireframe so that it is almost 100% true to the 3d model? Thinking specifically of the small details in the rims, front grill, etc.
Thank you. It could potentially match your model 110%, but If your CG model is a 2021 Porsche 911 and your text prompt describes a yellow 1972 Porsche Carrera, then certainly some elements will need to be adjusted to fit the contours of your line art generated by your CG renders. this is where the deformity comes from!!
This is absolutely fantastic. I saw your clip on LinkedIn and smashed “subscribe.” This was such an incredibly clever manipulation of materials and assets; just brilliant! Thank you for sharing your knowledge. People like you elevate the rest of the tide. Bless!
Pretty dope, thanks for the insights on your process! Just wished it would produce less artifact on the fish, hopefully at some point Sora model will get shared
THANK You. You've definitely inspired me to jump right in. Have been going at it for two weeks and its finally making sense. Feels like a new combination of Nuke and Blender with AI magic.