IMPORTANT: I'm currently investigating if bundling nodes into subgraphs may be causing the nodes to not operate in correct order. If I will conclude that is the case, I will upload an amended workflow.
This is a modified version of @zardozai's Unlimited Long Video Generation Loop (https://civarchive.com/models/1897323), but the Ollama part has been replaced with CreaPrompt, enabling you to define a prompt for reach segment of the generation by hand instead of relying on an LLM.
Description
FAQ
Comments (7)
Hello. An error occurs when launching WF. Apparently due to the ezXY node. It does not work in new versions of comfyui.
Prompt outputs failed validation: Exception when validating node: validate_inputs() takes 3 positional arguments but 4 were given
Well I'm running 0.3.59 and the workflow works fine for me. Is there any chance you could save a .json file of your setup and pop it my way, so I can troubleshoot?
Hi should this run on 12gb 4070 as I get an out of memory error after the firs generation.
Thanks
Hey! I was able to run the whole thing on my RTX 4070 Ti of 12 GB, so yes, you might wanna try getting some lower quants of the model, if I recall correctly, the best I could run this with locally was Wan 2.2 I2V 14B GGUF Q4 quants, for both the unet and the text encoder.
you should be able to run this, even though i have RTX 4070Ti Super with 16GB, but with GGuf Q4 or Q5 goes with 10GB, just make sure you also have enough storage. The drive storage left > model size.
Great work, but I seem to be missing something. The first group (first segment) works great, and I'm not getting any errors, but it never seems to move on to the next group (loop system, loop processing). Is there a switch somewhere I have to turn on? Something hidden in one of the sub-workflows? It just seems to end at the first generated video.
Thanks!
Hi! ;3 Unfortunately, I don't really have much time these days to work with videos anymore, but the workflow should work right out the box. Make sure all the models are loading (start, loop and end sections have separate model loaders), you have all the required custom nodes installed, and it should go through the whole system automatically, as they are all interconnected. I'm guessing maybe some model isn't loaded in the loop system part loaders, so it just stops there.