>>19909987
A few more tips.
You'll have to tell your workflow how many frames you want to gen. Generally, it's 16 frames per second. I wouldn't go over 6-7 seconds with the lightning loras since everything gets a little wacky the longer you go.
If you find that you enjoy this hobby, upgrade to 128 GB of ram. Some processes can fill up 64 GB of ram quickly. If it gets too high, comfy-ui will crash in the middle of your workflow. 64GB is definitely enough for most things, but it may be good to add some more memory to avoid having to think about it. If you notice your ram getting high, restarting comfy ui will clear it.
Block Swapping. Learn how to use it. Generally, I keep all blocks on system RAM since I only have a 16GB GPU. I've played around with it and just haven't seen many performance improvements by swapping 6-10 blocks to GPU. The only thing I found is that my system starts to lag if I try to use too many blocks on the GPU since it's maxing the VRAM available.
Start off small. Don't add in a ton of loras. Believe it or not, wan 2.2 is kinda uncensored and can actually do some nsfw stuff with just the model and the lightning lora. Adjust parameters in your workflow to see what effect it has on the final product. Really dive deep into the tuning. If you do that, you'll have a good idea of what went wrong if your gens aren't working the way they should be.
Resize your images properly. If you don't resize your images in a way that the model expects, you'll get random errors. Generally, models like their input images to be a multiple of 16 or 32 or 64 or whatever. I recommend using Resolution Master with the resize image v2 node from kjnodes.
Local AI gen is exploding right now. We've got a ton of new models to play with and there's always a new process or improvement coming out that you can apply to your workflows. Join the comfyui subreddit. Lot's of good info on there and it's a solid place to ask questions.
