XODA-LTXV-Video-Pack
0.9.5 base model update:
TBA version 8 workflows released!
0.9.1 base model update:
Initial Release video:
click show more, full details are below!
Consistent Video Variations can be provided by using my String Painter Node!
A comprehensive workflow pack for ComfyUI implementing the LightTricks LTXV video generation model, complete with example workflows, prompts, and custom tools for streamlined video generation.
Overview
This pack brings together essential workflows, tools, and resources for working with the groundbreaking LTXV real-time video generation AI model from LightTricks. It includes official example workflows, custom prompt generation tools, and extensive prompt lists to help users get started quickly.
Contents
Example Workflows
The pack includes the official ComfyUI maintainer and LTXV author-provided workflows:
AUTHOR-ltxvideo-i2v.json
AUTHOR-ltxvideo-v2v.json
COMFY-ltxv_image_to_video.json
COMFY-ltxv_text_to_video.json
Resources and Tools
Getting Started Guide
Comprehensive setup instructions
Required models and dependencies
Available at: CivitAI Article #9115
LTXV PromptGen LLM Module
Custom LLM module specifically designed for LTXV video generation
Repository: LLM-Base-Prompts/LTXV-PromptGen
Curated Prompt Lists
110 professionally crafted prompts
Organized into 11 categories
Located at: LTXV-PromptGen/LTXV-PromptLists
Integration with DJZ-Nodes
Custom nodes for workflow automation
Includes Zenkai-Prompt nodes for automated prompt handling
Seamless integration with provided prompt lists
Located at: DJZ-Nodes/Prompts
Getting Started
Clone this repository
Follow the setup guide at CivitAI Article #9115
Install required models and dependencies
Load example workflows in ComfyUI
Features
Official LTXV example workflows
Custom prompt generation tools
Extensive prompt library
Automation capabilities
Integration with existing tools
Requirements
Please refer to the Getting Started Guide for a complete list of required models and dependencies.
Description
FAQ
Comments (9)
Wonderful! Everything works fine, but I can't understand why I always get pixelated videos with little squares, like the one I uploaded of the rainbow wave generated with 50 steps. any advice?
it might be that this model must be run with 30 steps, we tried all kinds of different setups, i added in the V2 release a \testing\ folder containing "Sampler Garden, CFG Garden and Steps Garden" workflows to investigate alternate setups and almost all of them were on par with the defaults or worse in fact. Confirmed by many users now. Seems like the defaults it comes with are the best.
TLDR - drop the steps back down to 30 and see if you still see pixalation, and check you are using the T5XXL_FP16, those are my best suggestions :) everything else is the prompt, so mimic the examples from their github.
@driftjohnson thank you! after trying and testing all night, i can say that it works very well but only for some types of scenes. it surely has few videos in its "pool" in fact it does very well on some scenes rather than others. the more you try to do surreal things the worse the result will be both in quality and adherence. i tried various prompts even my own custom and adapted for other sites like kling, i must say that it manages the prompts for minimax very well.
ty. the garden things were really handy for testing. also those frame selector and video combine. i may swap my nodes and implement those to simplyfy my workflow a little 🖤 i already have found some tricks that can be usefull but at this early stage of this model everythin gis a test i guess
also havent found yet a workflow that uses those PixArt-XL-2-1024-MS files. i just downloaded them now, i was using LTX normally without those 🙄. no idea how to try them or load them up
@LatentDream iirc those are for the Author workflows, these are slightly different from the Native ComfyUI workflows which mine are all based on. You can find them in the \examples\ folder, but are no longer needed afaik.
you know what i'll drop some discoveries here in case you want to test them out, cause i'm nt gonna publish anything more yet about LTX untill bigger news.
-LTXconditioning frame rate: lowered to half (12,5) can help when image are frozen. i actually tend to leave that at that value, untill i see that the video looks like speed up too much
-LTXScheduler Max_shift: lowered to 1.5 can help when the video try to go crazy at the end
-i tryed to add Skimmed CGF and raise the native CFG to 10 to 20, sometimes got nice results
@LatentDream thanks for sharing !! i'll add those to my tester gardens and see what happened
I admit i was busy with some other stuff this week, so even though my V2 pack came out - i'll just to a roundup video with V3 release. Lots of cool new workflows on the github side, if you want to grab them early. Cause i'm adding so many every day it doesn't make sense to do daily big video update atm.
Moch1 Fine Tuning arrives, Flux Daemon added to my Foda pack and ltx-tricks with new Roda-ReactiveAudio pack almost ready! all that kept me busy all day.
@driftjohnson Don't worry, I know. With all these new features, it's already hard to sleep well. 😢 We’d need multiple clones of ourselves to test everything, but we do what we can. I just released a new workflow (It's Alive V8) in case you want to take what you need from it.