Always load Random Images for Both First Frame and Last Frame, Load random audio file in audio loader (it wont get used, unless you activate the options)
Please Watch Entire Tutorial Before using the Workflow.
www.youtube.com/watch?v=bxGQp2kylqk
Essential Requirement
added "Llama-CPP-Python_Windows(No_Deps).bat" file and double click it for auto installation.
or
X:\XXX\ComfyUI_Windows_Portable\ComfyUI\python_embeded\python.exe -m pip install llama-cpp-python
or
X:\XXX\ComfyUI_Windows_Portable\ComfyUI\python_embeded\python.exe -m pip install https://github.com/JamePeng/llama-cpp-python/releases/download/xxx_xxx.whl
install correct version of llama-cpp-python wheel
https://github.com/JamePeng/llama-cpp-python/releases/
Before opening the workflow, please update/download following Nodes
https://github.com/monnky/ComfyUI-RT-LTX2-RareTools
https://github.com/kijai/ComfyUI-KJNodes
https://github.com/Lightricks/ComfyUI-LTXVideo
https://github.com/city96/ComfyUI-GGUF
Description
v 1.2
FAQ
Comments (17)
This looks amazing!
To keep things clean would you suggest a standalone Comfy for LTX?
Thanks mate, your work is as ever, really appreciated.
first time to comment ever, is for you man, this is Awesome, i hope you can also make the Motion Control Version Also
I'm having a problem with the comment nodes on your YouTube video; I'm getting the same error.
I just updated ComfyUI to 0.17... do you think I should be able to run your workflow?
hello thank you, this workflow is great.
v1.2 I keep getting an error at the "Load audio" node, regardless of whether the audio switch is set to true or false. Debug: Custom validation failed for node, audio - Invalid audio file: 15004c1a-7406-38dd-620d-5354060330c6.wav
Another Workflow that does not work.
Switches dont work.
T2V - True
First/Last Frame - False
Custom Audio - False
Still asks me to add each section marked as false.
Cant bypass them, another error then comes up.
The switches work for me. The issue that was throwing me off is the image and audio nodes have to be linked to something even if they are not used. If the switch says false the node appears active but the image or audio in the switched off node is not referenced.
you have to fill all the image nodes, audio nodes first even if you won't use them
rtx nodes not installing and a big fat X
I installed Llama-CPP-Python_Windows in the description and it worked for me.
try going into your custom_nodes>ComfyUI-RT-LTX2-RareTools
find Llama-CPP-Python_Windows(No_Deps).bat and double click. It will install automaticly. Restart comfyui.
Works great thank you!!
I wish generation would be faster tho. 12gb vram with 32 gb ram. Using 2.3 22b distilled 4QKM gguf.
5 sec upscaled final>1920p video took close to 8mins.
I would love to have faster generation (2-3mins with lower quality) then upscale the good ones with some other fast workflow too. If anyone have suggestion I would appriciate that.
or I can bypass second pass generation to make it fast but eventually do I have to rerun whole process with same seed to get upscaled video?
what setting ration did you used it?
Low Vram : 640 x 360 -> Final Video after Upscale : 1280 x 720 (720p HD)
Mid Vram : 960 x 544 -> Final Video after Upscale : 1920 x 1088 (Full HD)
High Vram : 1280 x 704 -> Final Video after Upscale : 2560 x 1408 (2K)
V. High Vram : 1920 x 1024 -> Final Video after Upscale : 3840 x 2048 (4K)
@nozilanozila 960x544 with final 1920x1080.
update:
I changed sampler to euler from euler_cfg_pp (default one) and it increased my speed a lot with a little (maybe) quality lost but worth it. from 440sec to 280sec.
1- I have another problem, I cannot change the total fps from 'Video Width + Video Height + FPS + Total Frames' section. Probably Int nodes are not working properly and couldnt find the fix.
2- Also Stage 1+2 subsgraph display settings are not changing when you type something else for ex. it says euler as sampler on subsgraph when you change it still says euler on subsgraph but it changes inside of subsgraph.
How can I fix this?
problem 2 is not a biggie but still wondering how:) but How can I fix the problem 1?
I really want to try your 'RT-LTX-2 Royal Prompt by RareTutor' node for prompts. I watched your full video, and did the settings at the end of your video about prompt generator:
-gemma-3-12b-it-Q4_K_M.gguf
-mmproj-BF16.gguf
After I press Run, it doesnt give me error but 'Show Text' node output is:
Error: Llava15ChatHandler(_init_mtmd_context): Failed to load mtmd context from: C:\Users\Cem\Desktop\ComfyUI_windows_portable\ComfyUI\models\clip\mmproj-BF16.gguf
Also tried some different models but getting same error.
How can I fix this? -_-
Absolutely love the clean interface.
Took a while to get all the nodes and llama settled but finally got everything working and error free.
Unfortunately with I2V or T2V (I.e. the sample prompts used in the base workspace) all video is horribly motion blurred and pixelated :/ At this point I have no idea how to fix it :(
(Using the safetensor versions of ltx 2.3 models, 32G vram)