# LTX-2.5 > LTX-2.5 is Lightricks' open-weight audio-video model, released on 11 August 2026: a 22B-parameter diffusion transformer that succeeds LTX-2.3 with a diffusion video decoder, native multi-shot scenes and output up to 4K. On Kenerate AI it runs online as LTX 2.5 — text or image to video, 5–20 s, 720p, 1080p, 2K or 4K, 16:9 or 9:16, with a start and end frame — for 30 credits a second at 720p, so a 5-second 720p clip costs 150 credits. URL: https://kenerateai.com/models/lightricks/ltx-2-5 Publisher: Kenerate AI (https://kenerateai.com) Last updated: 19 October 2026 ## Key facts - Official name: LTX-2.5 - Model ID: Lightricks/LTX-2.5 - Maker: Lightricks - Released: 11 August 2026 - Size: 22B parameters, dev + distilled - Clip length on Kenerate: 5–20 s - Resolution on Kenerate: 720p · 1080p · 2K · 4K - Aspect ratios: 16:9 · 9:16 - Frames: Start frame + optional end frame - Billing: 30 / 42 / 63 / 99 credits a second ## Steps 1. Open the video studio — Go to /app/video and pick LTX 2.5 (the button on this page selects it). Video needs any one-time credit pack. 2. Choose the mode — Text to video for a written shot, or image to video with a start image — add a second image as the end frame if you want the clip to land on it. 3. Write the prompt — Describe subject, action, camera move, lighting and mood in one or two sentences. A prompt is required in both modes. 4. Pick length, resolution and ratio — Choose 5–20 seconds, 720p to 4K and 16:9 or 9:16. The credit cost shows before you generate. ## In depth ### LTX 2.5 vs LTX 2.3 vs LTX 2 on Kenerate All three LTX models on Kenerate make 5–20 second clips from text or a start image, in 16:9 or 9:16. They differ in resolution, frames and price — prices below are for a 5-second clip. - LTX 2.5 — 150 credits at 720p, 210 at 1080p, 315 at 2K, 495 at 4K. Start and end frame. Final renders and high-resolution work. - LTX 2.3 — 30 at 480p, 45 at 720p, 60 at 1080p. Start frame, plus extend by up to 20 seconds. - LTX 2 — 18 at 480p, 24 at 720p, 36 at 1080p. The cheapest way to test an idea. ### What changed in LTX-2.5 Lightricks swapped LTX-2.3's VAE reconstruction stage for a diffusion video decoder, which it says gives sharper faces, textures and on-screen text, at the cost of more decode time and memory when run locally [2]. The model can also plan several connected shots — wide, medium, close-up — in one generation while keeping the character, setting, lighting and voice consistent, and a duration predictor estimates clip length from the described action [1][2]. The text encoder is now a bundled Gemma 4 12B instead of a separate download [2]. ### Running LTX-2.5 online instead of locally Locally, LTX-2.5 needs the 22B checkpoints, the Gemma 4 encoder and ComfyUI's LTXVideo nodes; Lightricks quotes a 16 GB VRAM floor with int8 or NVFP4 builds and CPU offload [2]. On Kenerate the model runs on hosted GPUs: no download, no node graph, and the same credit balance covers Seedance, Kling, Wan and Veo. The trade-off is fewer controls — no LoRAs, audio-to-video, video-to-video, steps or CFG — so if you need those, run the open weights yourself. References: [1] LTX-2.5: Lightricks launches 22B open video model for ComfyUI — ComfyUI Wiki: https://comfyui-wiki.com/en/news/2026-08-11-ltx-2-5-open-weights-release [2] LTX-2.5 adds native multi-shot video and a diffusion decoder — NYU Shanghai RITS: https://rits.shanghai.nyu.edu/ai/ltx-2-5-adds-native-multi-shot-video-and-a-diffusion-decoder/ [3] Lightricks/LTX-2.5 — Hugging Face: https://huggingface.co/Lightricks/LTX-2.5 ## FAQ Q: What is LTX-2.5? A: LTX-2.5 is Lightricks' open-weight AI video model, released on 11 August 2026. It's a 22B-parameter diffusion transformer that generates video and audio together, and it replaces LTX-2.3's VAE decoding stage with a diffusion video decoder for sharper detail. It ships as a dev checkpoint and a faster distilled one. Q: When was LTX 2.5 released? A: Lightricks released LTX-2.5 on 11 August 2026, with weights on Hugging Face (Lightricks/LTX-2.5), code in the Lightricks/LTX-2 GitHub repo and ComfyUI workflows on day one. It followed LTX-2.3, which came out in March 2026. Q: How much does LTX 2.5 cost on Kenerate? A: It's billed per second of output: 30 credits at 720p, 42 at 1080p, 63 at 2K and 99 at 4K. A 5-second clip costs 150, 210, 315 or 495 credits; a 20-second 720p clip costs 600. Credits come in one-time packs from $15 for 1,300, with no subscription. Q: Is LTX 2.5 free? A: The weights are free to download under the LTX-2.x Community License (companies with $10M or more in annual revenue need a paid licence), but running the 22B model locally needs a strong GPU. On Kenerate, video isn't on the free plan — any one-time credit pack unlocks it, and the cheapest LTX 2.5 clip is 5 seconds at 720p for 150 credits. Q: LTX 2.5 vs LTX 2.3 — what's the difference? A: LTX-2.5 adds a diffusion video decoder, native multi-shot scenes, a duration predictor and a bundled Gemma 4 12B text encoder. On Kenerate, LTX 2.5 goes to 2K and 4K and takes an end frame, while LTX 2.3 tops out at 1080p but costs much less (9 credits a second at 720p) and can extend a clip. Q: What resolution does LTX 2.5 support? A: Lightricks says LTX-2.5 can render native 4K HDR at up to 50 fps. On Kenerate you can choose 720p, 1080p, 2K or 4K, in 16:9 or 9:16. Test prompts at 720p first — 4K costs over three times as much per second. Q: How long can LTX 2.5 videos be? A: On Kenerate, 5 to 20 seconds per clip, in whole seconds, at every resolution. Lightricks doesn't state a hard maximum for the open model; its own speed figures use 10-second clips. Q: Can LTX 2.5 use a start and end frame? A: Yes. In image to video on Kenerate, the first image is the start frame and an optional second image is the end frame, and the model animates the path between them. LTX 2.3 and LTX 2 on Kenerate take a start frame only. Q: Does LTX 2.5 make sound? A: The open model generates audio and video together, and Lightricks says multi-shot output keeps the same voice across cuts. Kenerate's LTX 2.5 has no sound setting of its own, so describe the sounds you want in the prompt and check the result. Q: How do I write a good LTX 2.5 prompt? A: Write it like a shot description: subject, action, camera move, lighting and mood, in plain sentences. For image to video, describe the motion rather than re-describing the picture, and use an end frame when you need the clip to finish on a precise shot. ## Related pages - LTX 2.3 video generator: https://kenerateai.com/ltx-2-3-video-generator - Wan 2.7: https://kenerateai.com/models/alibaba/wan-2-7 - Kling 3.0: https://kenerateai.com/models/kwaivgi/kling-v3-0 - 4K AI video generator: https://kenerateai.com/4k-ai-video-generator-free - Image to video AI: https://kenerateai.com/image-to-video-ai - All AI video models: https://kenerateai.com/ai-video-models ## Sources - LTX-2.5: Lightricks launches 22B open video model for ComfyUI — ComfyUI Wiki: https://comfyui-wiki.com/en/news/2026-08-11-ltx-2-5-open-weights-release - LTX-2.5 adds native multi-shot video and a diffusion decoder — NYU Shanghai RITS: https://rits.shanghai.nyu.edu/ai/ltx-2-5-adds-native-multi-shot-video-and-a-diffusion-decoder/ - Lightricks/LTX-2.5 — Hugging Face: https://huggingface.co/Lightricks/LTX-2.5