The only model in this workspace that renders true 4K at up to 50 fps with the audio already locked to the picture — and the only one that animates to a soundtrack you upload. Up to 20 seconds per take, camera moves you choose rather than hope for, and a Fast tier that costs less per second than most 1080p models. Type a scene or drop in a photo; the shot comes back finished, not as a draft.
Quicker renders for iteration and previews. Up to 4K and 20s clips at 24/25 FPS.
0 credits available
Your clip appears here
Hit Generate to start the render on our cloud GPUs. The result plays back here as soon as it is ready.
Compliance update: we replace restricted model routes where required, or retain model access only with a full moderation evidence package, including input/output review and blocking controls.
Two real generations, zero editing — picture and sound rendered together in one pass.
Brief it like you would brief a cinematographer — lens, light, pacing, mood — and the LTX 2.5 video generator hands back the shot with its own ambient sound already in place. No storyboard, no crew, no second tool for audio.
Drop in a still, say what should move, and LTX 2.5 keeps the subject, palette, and framing you chose while adding real motion — cloth settles, liquid pours, the camera travels through the scene instead of faking a zoom.
Six upgrades in the 2.5 release you will feel on the very first render — sharper 4K, prompts that stick, whole scenes instead of clips, and sound that arrives with the picture.
LTX 2.5 swaps the old VAE for a diffusion video decoder, and you can see it: faces stay faces, fabric has weave, on-screen text is legible. Render at 4K and it still looks like footage, not an upscale — which is exactly what an AI video generator has to get right before anything else matters.
Two characters, a camera move, a lighting change, and a timing cue in one sentence? The Gemma 4 text encoder inside LTX 2.5 keeps all of it. You spend less time re-rolling a generic clip and more time getting the shot you wrote.
Ask for a wide, a medium, and a close-up and LTX 2.5 renders them as one connected sequence — same character, same lighting, same voice across every cut. You stop stitching clips together and start directing scenes.
Every generation comes back with synchronized audio: waves break when they hit the shore, footsteps land when feet do, engines rev as the car pulls away. No separate audio model, no manual sync, no silent draft that still needs a session in an editor.
Leave the length on auto and LTX 2.5 ends the clip where the action ends. Diffusion Fidelity Rendering puts extra compute on the frames that need it — crowds, fast motion, dense detail — so your hardest shots are the ones that hold together.
LTX 2.5 is a 22B-parameter open-weights model from Lightricks, published on Hugging Face under the LTX-2.x Community License. Here it runs on our GPUs: open a browser, type, generate. No downloads, no ComfyUI graph, no card to rent.
LTX 2.5 is Lightricks' latest open-weights AI video model: a 22B-parameter Diffusion Transformer with a Gemma 4 text encoder and a new diffusion video decoder. It generates video and synchronized audio in a single pass from text, images, or audio, and adds native multi-shot generation, automatic duration, and Diffusion Fidelity Rendering on top of the LTX-2 line.
Three things you notice on the first render: the picture holds up at 4K, the sound is already there and in sync, and the prompt you wrote is the shot you get. Add multi-shot sequences and auto duration and you are directing scenes, not fishing for a usable clip.
No. LTX 2.5 runs on our cloud GPUs and everything happens in the browser — no downloads, no local GPU, no command line. Generate from a laptop, tablet, or phone.
720p, 1080p, 1440p, and 4K (2160p) at 24, 25, 48, or 50 FPS, in landscape (16:9) or portrait (9:16). Portrait is generated natively, not cropped from a landscape frame, so vertical clips for Reels, Shorts, and TikTok come out properly composed.
Up to 20 seconds at 720p or 1080p with 24 or 25 FPS; up to 10 seconds at 48 or 50 FPS or at 1440p and 4K. Or leave duration on auto and let LTX 2.5 pick the length that fits the action you described.
Yes. LTX 2.5 generates video and audio in the same forward pass — ambient sound, music beds, and effects timed to what is on screen. The clip you download is ready to post, not a silent draft.
Yes. Upload a photo, product shot, portrait, or concept frame, describe what should move, and LTX 2.5 generates motion from that frame while keeping the subject, colors, and composition you started with. Combine image and text for full control of the shot.
It is open-weights. Lightricks publishes the weights on Hugging Face and the code and ComfyUI workflows on GitHub (github.com/Lightricks/LTX-2) under the LTX-2.x Community License — free for individuals and companies under $10M in annual revenue, with a commercial agreement above that. None of that applies to using it here: generation runs on our infrastructure.
Write one line or drop in a photo. The LTX 2.5 AI video generator takes it from there — 4K, synchronized sound, portrait or landscape, rendered on our GPUs while you watch.
