LTX-2.5: Lightricks Launches 22B Open Video Model for ComfyUI
Lightricks releases LTX-2.5, a 22B open-weights audio-video foundation model with native multishot, 4K HDR, and official ComfyUI workflow templates.
Overview
LTX-2.5 is the next generation of Lightricks' open video foundation family, succeeding LTX-2.3. It is built on a 22B-parameter asymmetric dual-stream diffusion transformer that generates video and audio jointly through bidirectional cross-attention with modality-aware classifier-free guidance. The text encoder is a Gemma 4 12B model with a learned projection.
The model ships as both a full dev checkpoint and a distilled few-step variant, and supports text-to-video, image-to-video, video-to-video, text-to-audio, and audio-to-video generation. Lightricks says a 10-second clip from an image can be generated in roughly 6-8 seconds on NVIDIA superchips with the distilled model.
Official LTX-2.5 hero demo (video from ltx.io)
What's New in LTX-2.5
Native multishot. The model can generate connected scenes — wide, medium, and close-up shots — in a single generation while holding character, environment, lighting, and voice consistent across cuts.
Auto duration. Clip length is predicted from the described action, so scenes land at the right duration without manual frame-count fiddling.
Native 4K HDR. The model targets high-resolution, high-dynamic-range output, with HDR pipelines built into the official workflows.
Diffusion Fidelity Rendering (DFR). A new rendering pass aimed at real footage, editable results, and cinema-grade EXR output for production pipelines.
Weights and Variants
The Hugging Face repository provides the full set of official weights:
- Base:
ltx-2.5-22b-dev-transformer-bf16.safetensors - Distilled:
ltx-2.5-22b-distilled-transformer-bf16.safetensors - ComfyUI-ready quantizations: int8 convrot and NVFP4 variants of the distilled transformer
- Text encoder:
gemma4-12b-with-proj-ltx-2.5(BF16 and ComfyUI int8 convrot) - VAEs: separate video VAE and audio VAE
- Upscalers: 2x latent spatial and 2x latent temporal upscalers
- Add-ons: a distilled LoRA (450) and a duration-head model patch
ComfyUI Support
LTX-2.5 is supported in ComfyUI on day one. Official workflow templates for text-to-video, image-to-video, and first/last-frame-to-video are available in the ComfyUI workflow templates library:
Lightricks also updated their official ComfyUI-LTXVideo wrapper with a full set of 2.5 example workflows, covering single-stage and two-stage (upscaled) T2V/I2V, ICLoRA inpainting and outpainting, motion tracking, union control, and text-to-audio generation.
Links
- Model page: ltx.io/model
- Weights: Lightricks/LTX-2.5 on Hugging Face
- ComfyUI wrapper: Lightricks/ComfyUI-LTXVideo
- Workflow templates: Comfy-Org/workflow_templates
Comments
Sign in with GitHub to join the discussion.