A simple, color-coded GGUF workflow for running LTXV-2 in ComfyUI.
Designed to be straightforward and lightweight.
Common Issues
Error: LTXVEmptyLatentAudio – AttributeError
Update Kijai Node Pack to v1.2.8
Checkerboard / patterned noise output
Update all nodes + ComfyUI
(Outdated GGUF loader or ComfyUI build)
Required Node Packs
ComfyUI-GGUF
https://github.com/city96/ComfyUI-GGUF
ComfyUI-KJNodes
https://github.com/kijai/ComfyUI-KJNodes (install & update it to nightly)
additional install via manager
📦 Downloads
🧩 Kijai’s Repo (ALL files except diffusion models — choose ONE LoRA)
https://huggingface.co/Kijai/LTX2.3_comfy/tree/main
Includes:
Distilled LoRA (REQUIRED)
Text Projection (embedding connector)
🧠 Diffusion Models (GGUF)
https://huggingface.co/unsloth/LTX-2.3-GGUF/tree/main
Choose based on VRAM:
Q4_K_M→ lower VRAM (recommended)Q8_0→ higher quality, more VRAM
⚠️ Very low quants = basically unusable
Includes:
VAE (audio + video)
GGUF models
✍️ Text Encoders (REQUIRED)
Option 1 – Gemma (recommended for low VRAM)
https://huggingface.co/Comfy-Org/ltx-2/tree/main/split_files/text_encoders
Option 2 – Full version (higher VRAM, better quality)
➡️ You still ALSO need:
Text Projection (from Kijai’s repo)
🔍 Latent Upscaler
https://huggingface.co/Lightricks/LTX-2/tree/main/
Files:
ltx-2-spatial-upscaler-x2-1.0.safetensorsltx-2-temporal-upscaler-x2-1.0.safetensors
⚠️ Important Notes
Use DEV GGUF diffusion models (no distilled base model yet)
Distilled LoRA is still REQUIRED
Most issues = outdated nodes or ComfyUI
.
Description
FAQ
Comments (13)
VRAMdebug: freed memory: 10,169,016,448
!!! Exception during processing !!! tuple index out of range
Traceback (most recent call last):
File "N:\ComfyUI-aki-v1.7dsadas\ComfyUI-aki-v1.7\ComfyUI\execution.py", line 525, in execute
output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "N:\ComfyUI-aki-v1.7dsadas\ComfyUI-aki-v1.7\ComfyUI\execution.py", line 334, in get_output_data
return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "N:\ComfyUI-aki-v1.7dsadas\ComfyUI-aki-v1.7\ComfyUI\custom_nodes\comfyui-lora-manager\py\metadata_collector\metadata_hook.py", line 168, in async_map_node_over_list_with_metadata
results = await original_map_node_over_list(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "N:\ComfyUI-aki-v1.7dsadas\ComfyUI-aki-v1.7\ComfyUI\execution.py", line 308, in _async_map_node_over_list
await process_inputs(input_dict, i)
File "N:\ComfyUI-aki-v1.7dsadas\ComfyUI-aki-v1.7\ComfyUI\execution.py", line 296, in process_inputs
result = f(**inputs)
^^^^^^^^^^^
File "N:\ComfyUI-aki-v1.7dsadas\ComfyUI-aki-v1.7\ComfyUI\comfy_api\internal\__init__.py", line 149, in wrapped_func
return method(locked_class, **inputs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "N:\ComfyUI-aki-v1.7dsadas\ComfyUI-aki-v1.7\ComfyUI\comfy_api\latest\_io.py", line 1764, in EXECUTE_NORMALIZED
to_return = cls.execute(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "N:\ComfyUI-aki-v1.7dsadas\ComfyUI-aki-v1.7\ComfyUI\comfy_extras\nodes_lt.py", line 675, in execute
audio_latent["samples"] = latents[1]
~~~~~~~^^^
IndexError: tuple index out of range
Prompt executed in 447.09 seconds
not sure why this is happening but it seems an issue with the audio vae and audio latent, make sure KJ nodes is updated to nightly and youre using comfyui desktop that might help it..
@The_frizzy1 This is happening to me as well, made sure to clear some space on the drive and downloaded all models exactly like you show in the WF, in the correct locations, even though the DualClipLoader saw text encoders in the CLIP folder I moved them to TE folder as instructed, everything up to date... 🤷♂️
@The_frizzy1 to be clear, I'm using the I2V version, I see your video examples are T2V so I can't really download to compare, but your T2V looks maybe a bit more polished than the I2V I am messing with.
@Pity_the_Foo Hmm weird ususally i2v generations look better even for me. I will try to update this but I think I found it try connecting the secound samplers audio latent not the first that might be the issue for i2v.
@The_frizzy1 Can you DM me a screenshot of what you're talking about if you get a chance? Your workflow is clean but I'm not very good at ComfyUI and haven't had ample enough time to actually generate stuff as well as learn the ropes lol. In any event thanks for your response and for looking into it, much appreciated.
My problem with LTX are generation times. Even if I lower to steps, cfg, 5 seconds. Takes 20 to 30 minutes. I am running the latest version of everything and have a 3080TI V12GB, 32GB. I also did a clean install of comfy. Any tweaking suggestions? I do not have these issues with WAN2.2.
im not really sure maybe use a smaller quant or try to cahnge the attn youre using or use libraries like triton. For me out the box wan2.2 takes longer and takes more recources than LTX2 by alot..
Uh, it should be shorter than WAN2.2. I can render a 30 second video in about 3 minutes. A 3080TI isn't th efastest thing on the planet but 20 to 30 minutes is very long for 5 seconds.
Are you sure your rendering at a resolution that isn't too high? Some workflows people upload start off with resolutions that are insanely high sometimes.
Thanks for this. Just a heads up, your note with the model links still links to the LTX 2 files
will fix that today thank you!
Es genial, solo podrías hacer la aclaración de que para personas con 6, 8 de vram, hay que desactivar todos los nodos de audio, liberar el subgrafo del sampler, y agregar otro muestreador , o siempre dara errores, pero en resumen esta genial, gracias amigo!
no need to do it in the new version all fixed up now, gracias!