Lightricks’ open DiT video generator does 4K and synced audio

What it does LTX-Video is a diffusion-transformer model that generates video up to 4K resolution at 50 FPS, with synchronized audio produced in the same pass. It handles text-to-video, image-to-video, video-to-video transformations, and can extend footage forward or backward from keyframes. The repository hosts 2B and 13B parameter weights, including distilled variants that skip classifier-free guidance and need only eight diffusion steps.

The interesting bit The README is unusually candid: it announces that LTX-2 is now the “primary home for LTX development” and directs users there for the latest work. This makes the repo a stable, open-weight checkpoint archive rather than a living trunk—you get working 13B and 2B models with ComfyUI workflows and Diffusers support today, while the successor is explicitly parked in another repository.

Key highlights

• Multi-keyframe conditioning, forward/backward video extension, and generations up to 60 seconds with the 13B model
• Distilled 13B and 2B checkpoints for faster inference; a distilled 13B LoRA adapter that requires only 1 GB of VRAM
• IC-LoRA control models for depth, pose, and Canny edge guidance, plus standard LoRA fine-tuning support
• OpenRail-M commercial license and native integrations for ComfyUI and Diffusers

Caveats

• The highest-quality 13B dev model “requires more VRAM,” though the README never quantifies exactly how much
• Active feature development has shifted to LTX-2; new weights and benchmarks for the successor are slated for release later in 2025

Verdict Grab it if you need an open, commercially usable video generator with fine-grained controls and don’t mind living on the 0.9.x branch. Wait for LTX-2 if you want the successor Lightricks is actively building toward.