What frame counts are supported by LTX-V2?

LTX-V2 requires a frame count following the 8n + 1 constraint. Common values include 49, 81, 97, 121, and 161 frames.

Which text encoder does this skill use?

This skill uses the Gemma 3 12B text encoder, which provides superior natural language understanding for video scene progression.

What is the benefit of the two-stage upscale pattern?

It allows you to generate a base video at a lower resolution (like 768x512) to save memory, then uses a spatial upsampler and a second sampling pass to reach high-definition resolutions like 1920x1088.

Can I run LTX-V2 on a 24GB VRAM GPU?

Yes, but it is highly recommended to use FP8 quantized checkpoints, the FP4 Gemma 3 text encoder, and tiled VAE decoding to avoid Out of Memory (OOM) errors.

How do I control camera movement in the generated video?

You can apply official camera control LoRAs such as Dolly Left, Dolly Right, or Jib Up at a strength of 1.0. These handle the movement without needing specific descriptions in your text prompt.

LTX-V2 Video Generation

Name: LTX-V2 Video Generation
Author: artokun

byartokun

•

データサイエンスとML

Builds high-fidelity 19B parameter video workflows for LTX-V2 text-to-video and image-to-video generation.

The LTX-V2 Video skill enables the creation of advanced video generation pipelines using Lightricks' 19-billion parameter DiT-based foundation model. It provides specialized implementation patterns for distilled 8-step generation, two-stage spatial upscaling, and cinematic camera control via dedicated LoRAs. By automating the complexities of frame-rate conditioning, VRAM optimization through tiled decoding, and strict frame-count constraints (8n+1), this skill allows developers to integrate professional-grade video synthesis directly into their Claude Code environment.

主な機能

01Support for both Text-to-Video (T2V) and Image-to-Video (I2V)

02Two-stage spatial upscaling for high-resolution output

03Distilled 8-step fast video generation pipeline

04Cinematic camera control LoRAs (Dolly, Jib, Static)

05VRAM-optimized spatio-temporal tiled VAE decoding

0611 GitHub stars

ユースケース

01Generating high-resolution cinematic clips from natural language descriptions

02Animating static images with precise camera movements for marketing content

03Building automated AI video production pipelines with 2x latent upscaling

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add artokun/comfyui-mcp ltxv2-video

For use in Claude.ai and ChatGPT

Download Skill