This is a heavily modified workflow from LTX-2.3 All-In-One workflow for RTX 3060 with 12 GB VRAM + 32 GB RAM - v1.0 | LTX Video Workflows | Civitai.
All credits go to the creator for his excellent workflow.
I modified it using the settings and models from the official 10Eros workflow TenStrip/LTX2.3-10Eros_Workflows at main
Plus I added the "Prompt Relay Encode (Timeline)" and some other Q.O.L. settings.
All instructions and downloads are in the workflow.
Description
FAQ
Comments (19)
very nice WF. what are you using to write your prompts?
Hi, I use this https://huggingface.co/SulphurAI/Sulphur-2-base with LMstudio is for img to prompt, read the instructions on the model card for how to setup is very easy to setup, but some manual refinement is needed.
Hi, i just wanna know if smooth transition from image input 1 to 2 is possible? I tried some other workflows and there is always a problem that when second image generation starts it doesn't have reference from first input generation the same as first generation doesn't target 2 input image, so it kinda results in video 1 + video 2
Hi I have used this LoRA, is in the lora links in my workflow with good success valiantcat LTX-2.3 Transition LORA - v1.0 | LTX Video LoRA | Civitai
Is all t2v horrible from 10eros? It just seems to always be this ugly looking thing.
Hi, Try euler for generation and use res_2s or dpmpp_2m for the upscale
Hello, my entire workflow ran perfectly in the early stage without any SageAttention errors. But after I entered the "processing video" stage, the samplerCustomAdvanced node threw an error: torch.OutOfMemoryError: Allocation on device. This error indicates your GPU memory has run out. TIPS: If the workflow worked previously, you may have accidentally set batch_size to an overly large value.
My hardware setup is RTX 4070 Super paired with 32GB system RAM. I’ve tried disabling SageAttention and lowering the Video Height parameter to 1024, neither of which resolved the issue. Also, I haven’t enabled any preview functions. Could you please help me troubleshoot this?QAQ
So I suppose that your 4070 is the same as mine with 12GB VRAM, all the videos that I have upload are with this workflow. First tip increase the paging file of windows to 122880MB (make a google search for how to do this, is easy), set this number to initial and maximum size, is better to use an SSD and even better a different SSD from that windows is installed. If that does not fix it, there is a minimized unused node under the LoRA node, called "LTXV Chunk FeedForward" it has a model input and output, put it between "model_2nd_pass" and "Set_model2nd". Finally edit your "run_nvidia_gpu.bat" with e text editor and add this "--lowvram --windows-standalone-build --use-sage-attention --windows-standalone-build --listen 0.0.0.0 --port 8188" after the "main.py". Hope that helps happy video generating :-)
@panosgr Thanks for your quick reply! Do you mean turning on the chunked rendering function? I can’t locate the two nodes "model_2nd_pass" and "Set_model2nd". Should I plug the model output from the Power LoRA Loader into both input and output sides of the chunk node?
@2730286336794 Under the "Patch Sage Attention KJ" node there is a "subgraph" named "New Subgraph", it has two outputs use the "model_2nd_pass" output to the input of "LTXV Chunk FeedForward" mode and the output to the input ""Set_model2nd" node under the "Distilled Lora" node. Hope that helps. Let me know how it goes.
@panosgr I’ve tried really hard to locate the Patch Sage Attention KJ node, but I can’t find it anywhere in the v6.1 workflow. If you’re referring to the Patch Flash Attention KJ node from KJNodes, it’s not being used in this workflow at all. Maybe I just don’t have enough experience with ComfyUI,T-T
@panosgr I'm using ltx2310eros_v1.safetensors,I don't think it has anything to do with this model
@2730286336794 you are not using my workflow then, my workflow is V1.0, and uses another variant of 10Eros, (all the links for the models are in the workflow, download them in the right folders), so download my workflow and drag and drop it in ComfyUI. Rgards
@panosgr Oh fk, I wouldn’t have caught that without your heads-up. Sorry for bothering you, my bad for missing it. I’ll follow your workflow moving forward. :(
@2730286336794 Hi, No problem, we all must learn somehow, we should help each other when we can, use my wf, and give me an update if you want. Regards
@panosgr ok! salute to u, bro. have a nice day.
Wanted to give it a try, but I give up finding the correct LayerUtility: AnyRerouter.
Try the installation instructions at the original github repo : chflame163/ComfyUI_LayerStyle: A set of nodes for ComfyUI that can composite layer and mask to achieve Photoshop like functionality.
Download this node and replace the existing one, please note that there are hidden nodes behind the old one, so move the old node before replace it with the new one "https://www.transfernow.net/dl/20260706WBPkcoJ0/u4C2HY8l"