Hey folks! Back with another LTX 2.3 drop — this time we’re going advanced with workflows that let you generate video using timbre reference, plus either storyboard frames or character design sheets.
This is all built on the workflow pack I shared last time, using ID-LoRA and IC-LoRA, just with tweaked reference inputs to fit these new use cases. There are 2 workflows in total:
One for generating video from storyboard frames + timbre reference
A dual-character dialogue workflow, using 2 character reference sheets + timbre reference
Don’t get me wrong — closed-source models still blow this out of the water when it comes to rock-solid character and voice consistency. But I wanted to share this as a working option for anyone sticking to open-source tools.
Quick breakdown of what I found in testing:
The storyboard method gives you way better character lock, but it lives and dies by how good your storyboards are, and making those takes a ton of time.
The dual-character dialogue workflow skips all the storyboard and scene setup work entirely. It’ll keep your characters consistent for a full 20-second video in one single generation, but you’ll see more consistency drift if you re-generate multiple times.
On the timbre reference side: in my tests, the voice tone shifts a lot depending on the emotion of the dialogue, so it only picks up the original reference partially right now.
Open-source models have their limits for sure, but I’m super excited to see where LTX goes with future updates. I truly believe open-source video models are only going to get better and better, especially when it comes to locking in reference consistency.
And of course, all the workflows are 100% free, no login needed, instant download. You can test them out online right away, or grab the local files to run on your own machine — just hit the download button in the top-right corner.
Timbre Reference (ID-LoRA):https://www.runninghub.ai/post/2043718764537057282?inviteCode=rh-v1495
Dual Characters and Timbre Reference:https://www.runninghub.ai/post/2042618627672772609?inviteCode=rh-v1495
ID-LoRA-TalkVid:https://huggingface.co/AviadDahan/LTX-2.3-ID-LoRA-TalkVid-3K
IC-LORA-Dual-Character:https://www.modelscope.cn/models/fxj1131/LTX2.3-IC-LORA-Dual-Character
ChineseXianxiaStyle Lora:https://www.modelscope.cn/models/fxj1131/ChineseXianxiaStyle
I put together a full walkthrough video with all the nitty-gritty details, so check that out if you want a deep dive.
![[ComfyUI]LTX 2.3 Advanced Workflows | Video Generation from Character Reference & Timbre Reference](https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/84b5e967-5226-4078-ac21-7b33e580b3b2/width=1320/竖版封面en.jpeg)