LTX 2.5 is here

LTX-2.5 is the newest version of LTX's open video model, with day 0 support in ComfyUI. Weights are downloadable and it runs fast on local GPUs. Native 4K, synchronized audio and video, and frame rates up to 50 FPS. Read the full announcement on the Comfy blog.

LTX-2.5 improves the full generation stack rather than a single stage. New in this release: Diffusion Fidelity Rendering, a diffusion video decoder, a custom text encoder, a reworked distilled variant, a prompt enhancer, and a base checkpoint built for adaptation. Native 4K, synchronized audio, and up to 50 FPS carry over from 2.3. Read the full breakdown on the Comfy blog.

The core change in this release. Instead of spending compute evenly across a scene, the model allocates it by complexity. Structure comes first: motion, composition, and framing generate in an 8x temporally compressed latent space, alongside a set of high-fidelity keyframes, more for complex scenes and fewer for simple ones. A dedicated pixel-diffusion stage then renders the final video from structure and keyframes together. Textures, materials, intricate objects, and faces resolve with pixel-level precision, and busy shots draw more rendering compute than static ones.

Open weights you run yourself: LTX-2.5 dev is the main model, and LTX-2.5 distilled is a smaller, faster variant that now carries more quality, prompt adherence, and motion than earlier distilled releases. Through Partner Nodes: LTX-2.5 (Fast) covers the wider envelope at 2 to 20 seconds, 720p through 4K, landscape or portrait, at 24, 25, 48, or 50 FPS. LTX-2.5 (Pro) runs 2 to 10 seconds at 720p or 1080p, at 24, 25, or 50 FPS.

Update ComfyUI to 0.32.0, or open Comfy Cloud. Download the LTX-2.5 weights and place them in your models directory. Load the LTX-2.5 template from the Templates panel: Text to Video, Image to Video, or FLF2V. Add your prompt and input images, then run. Full setup steps are in the docs.

Through Partner Nodes, Fast runs 2 to 20 seconds, with clips over 10 seconds capped at 720p or 1080p and 24 or 25 FPS. Pro runs 2 to 10 seconds. LTX-2.5 also ships an experimental duration head that reads the action in your prompt and sets clip length before diffusion starts.

Yes. Synchronized audio and video carry over from LTX-2.3, and multi-shot generations hold voice across cuts.

Yes. One generation produces multiple connected shots, holding character, environment, lighting, voice, and style across the cuts. You get a sequence from a single run instead of matching separate generations afterward.

Yes. The base checkpoint is built for adaptation, so you can fine-tune the raw model on your own data.

Open weights are free to download and run on your own hardware. Running LTX-2.5 through Partner Nodes uses credits. See the docs for setup details.