How to Generate 1-Minute AI Videos Using LTXV 0.9.8 ComfyUI Workflow with LoRA Detailer

Let’s talk about the LTXV 0.98 model — and why it’s a pretty big upgrade if you’re working inside ComfyUI.

So yeah, here’s what dropped: the LTX team released a new base model (v0.9.8), a detail-enhancing LoRA, and an upscale model — all tuned to work together. It’s designed to handle both text-to-video and image-to-video, and supports generation up to 1 minute at 24 FPS.

There’s a distilled variant that actually improves prompt understanding and detail generation, without pushing your VRAM over the edge.

What You Get with the LTXV 0.98

There are three components to know:

• A main video model (this is the one you use for generation)
• An IC LoRA Detailer to punch in finer details (optional)
• A new LTXV Spatial Upscaler, which cleans things up at the end

And the main generation model itself comes in a few variants — depending on your hardware:

So you’ve got flexibility here. Just match the version to your system specs and drop it into your models/checkpoints folder like usual.

For the upscale model, download LTXV Spatial Upscale 0.98 and place that one into your ComfyUI upscale_models folder.

As for the LoRA, grab the LTXV 0.98 IC LoRA Detailer (ComfyUI version) and save that into your LoRA directory. If you’re using ComfyUI regularly, this setup will feel familiar

Workflow

Once you’ve got everything saved in the right folders, the workflow starts like this:

That’s your base. From there, you’ve got two optional paths:

Honestly, both work — but they give different looks.

There’s also an “Extend” section in the workflow. This lets you push past the 1-minute cap using an overlap-style method that stitches extra frames in without glitches. It’s surprisingly smooth — more on that in a bit.

Generating Longer Videos with VAE Decode Title Node

Now if you’re trying to generate a full 1-minute video in one go, you’ll want to switch to a different node.

Specifically: use VAE Decode Title. That’s the one that allows longer clips.

I gave it a try. The good news? It worked — I got the full minute without crashing ComfyUI or hitting a memory wall. But there was one issue: the car didn’t move.

The output looked clean, but the motion felt stuck. Like the object was frozen in place and the camera was just drifting around it. Not ideal.

Fixing Motion with the Extend Sampler Trick

So here’s a better way to do it.

Instead of trying to generate all 721 frames in one pass, I switched over to the Extend Sampler method.

The result? Way smoother transitions, and the object motion stayed consistent. No weird frame warping or sudden flickers.

I honestly didn’t expect it to work that cleanly, but it held up. The quality was a big jump from the original 1-minute clip I made earlier.

Final Thoughts

This model really impressed me with its ability to generate smooth, high-quality videos up to 1 minute long. The new upscaling model and IC LoRA Detailer make a big difference in the final output quality.

If you’re working with video generation in ComfyUI, I’d definitely recommend giving LTXV 0.9.8 a try. The workflow might take a bit of tweaking to get exactly right, but once you dial it in, the results speak for themselves.