note: there is a bug on civitai where you cannot see total number of downloads or likes. trust me its getting downloaded. comments are also not appearing for some reason. comment on my youtube video if you have any issues/questions.
MY PC SPECS (for comparison):
(RTX 3070+I-7) 8gb VRAM + 16gb System RAM = 24gb in total. it works pretty well for such high memory requirements.
LINKS
Dev GGUF (pick a lower quant than your vram requirement, the total workflow requirements are HUGE)
https://huggingface.co/unsloth/LTX-2-GGUF/tree/main
Text Encoder (runs alongside the second encoder in dual clip loader gguf node)
https://huggingface.co/GitMylo/LTX-2-comfy_gemma_fp8_e4m3fn/blob/main/gemma_3_12B_it_fp8_e4m3fn.safetensors
embeddings connector (runs in dual clip loader gguf node with gemma encoder)
Distill LoRA (for upsampling later in workflow)
https://huggingface.co/Lightricks/LTX-2/blob/main/ltx-2-19b-distilled-lora-384.safetensors
DETAIL LoRA (adds fidelity to lower resolution generations. trust me it helps.)
https://huggingface.co/Lightricks/LTX-2-19b-IC-LoRA-Detailer/blob/main/ltx-2-19b-ic-lora-detailer.safetensors
vaes (audio AND video)
audio
https://huggingface.co/Kijai/LTXV2_comfy/blob/main/VAE/LTX2_audio_vae_bf16.safetensors
video
https://huggingface.co/Kijai/LTXV2_comfy/blob/main/VAE/LTX2_video_vae_bf16.safetensors
latent upsampler (place in latent_upscale_models folder)
https://huggingface.co/Lightricks/LTX-2/blob/main/ltx-2-spatial-upscaler-x2-1.0.safetensors
IMPORTANT!!!!
you need the following custom nodes:
-KJ Nodes (available in custom nodes manager, update and use nightly)
-Comfyui-GGUF nodes (available in custom nodes manager, update and use nightly)
-a custom node from VantageWithAI named "Vantage GGUF Unet Loader" node
this node is experimental but more efficient than the regular unet GGUF nodes. (there is a PR pull missing from the GGUF nodes at the moment and until it is merged fully the node wont work without updating it in your command prompt yourself. i dont suggest as it is highly unusable. trust me, use the Vantage node.)
TO GET THE VANTAGE NODES, open a command prompt in your custom nodes folder and copy in this command and hit enter:
git clone https://github.com/vantagewithai/Vantage-Nodes.git
(i removed the "cd comfyui/custom_nodes" portion from the youtube video tutorial as its not needed)
use this command once its finished:
cd Vantage-Nodes
pip install -r requirements.txt
this installs the nodes fully. restart comfyui and you drag in my workflow and you have a fully functioning LTX-2 workflow!
I HIGHLY RECOMMEND UPDATING YOUR BAT FILE WITH THESE FLAGS:
--lowvram --disable-xformers --use-pytorch-cross-attention --reserve-vram 2 --disable-smart-memory
this workflow is very resource intense. even for the lowest quant my pc struggled until i lowered resolution to 480p. i suggest editing your bat file in notepad and just adding the flags to the code line that contains the "--windows-standalone" code and just save the file as a copy and use that one for LTX-2 ONLY. (rename it that if it helps you remember) instructions on how to do it are in the youtube video if you dont know how!
TUTORIAL VIDEO
Description
notes will be uploaded shortly.
FAQ
Comments (24)
I couldn't find source for the gguf node
read the description bud. its plain as day how to get it.
@realrebelai Installation Error: Failed to clone repo: https://github.com/M1kep/ComfyLiterals
Work great~ The bat file is important. First I didn't change and it WOW. take a long time to generate. After change, it take 5mins done.
Where to set the video length? Where to randomize the seed? It doesn't want to auto run because nothing is changing that I can tell. I'm getting basically a static image with the audio behind it :S
yeah so the frame count is in the first level of the workflow before you enter the subgraph. you can input your frames there. also, be careful what you input into negative prompt because thats what was causing issues for me being static outputs, and the workflow not running multiple runs its because of the workflow not randomizing the seed after generation, the native workflow was incorrectly strung that i pulled from comfy and edited to produce this workflow. you need to reassign the slots in the subgraph to remove the "control after generate" and "seed" options from the subgraph then re-enter them into the subgraph and it will correctly reassign them to slots so you can use them without entering the subgraph entirely. if you click the subgraph the options bar comes up and you just click the tab with the 2 horizontal lines with a ball on opposing ends (thats the option to assign slots in the subgraph) then remove the "control after generate" and "seed" options and then reassign them back in and itll fix it. i noticed even when i changed them to randomize and inputted a number, it was still fixed. i went into the subgraph and changed them myself and it corrected it with this method. i just didnt realize it was an issue until i started stress testing the model after i posted the workflow that they had incorrectly strung it together.
you can also just go into the subgraph and randomize the seed yourself if you dont feel like re-assigning the slots in the subgraph to the nodes
@realrebelai cant you just fix it and reupload it?
left message on youtube, but after about 2 hours I got: VantageGGUFLoader
expected str, bytes or os.PathLike object, not NoneType
Though I did not think to update and use nightly on the gguf nodes, I'll do that and try again. I'm on a 5060ti, 16gb vram, 32gb ram, win11, py3.12, cu128
hmm thats strange, my LLM is telling me you most likely didnt activate the model in the node, meaning you need to select the gguf model in the node so it can run it, the workflow wont run unless you call on your own personal files because its catered to my files technically until you implement yours (and it cant find mine because you dont have mine if that makes sense)
is your dev model in your unet folder? it cant be the distilled model it needs to be the dev
Works great once I got everything running, the only thing I can't seem to get to work is gemma3 which is fine since I like to write my own prompts anyway. Good work and thanks!
yeah i didnt try gemma 3 tbh, figured it would be too straining on the workflow with everything else, wanted to keep it as condensed as possible!
LTXVEmptyLatentAudio
'VAE' object has no attribute 'latent_frequency_bins'
help
you need to update kj nodes
ohh... follow the directions... bingo! you da man, thanks
Getting OOM on the VAE decode, it suggests using tiled decode...There is an LTX Tile Decode node, would that be better? What settings on that?
So this is only i2v??
Interesting - works flawlessly... EXCEPT I2V the image is completely ignored... what gives? Any clue?
works great.
Where is this text encoder? I have gemma but have no idea about the second one. The models I've seen from ltx are made into shards.
embeddings connector (runs in dual clip loader gguf node with gemma encoder)
https://huggingface.co/Kijai/LTXV2_comfy/blob/main/text_encoders/ltx-2-19b-embeddings_connector_dev_bf16.safetensors
It's this one.
Theyre both listed in the description
Can I pool gpus with this like two 8gb gpus?? And have it work on same generation ?
Theres seperate comfy versions for twin gpu use