LTX 2.5
LTX 2.5
LTX 2.5 is Lightricks' new open-source 22B video model — an open world model generating synchronized, high-fidelity video and audio from text, image, and video inputs.
Native multishot scenes, a sharper diffusion video decoder, and a Gemma 4 12B text encoder make LTX 2.5 the strongest release in the LTX family. Generate connected multi-scene clips in seconds — free to try online.
Generate for FreeNative Multishot Generation
LTX 2.5 generates connected scenes in a single pass: multiple shots that hold character identity, environment, lighting, voice, and visual style across cuts.
- One prompt drives a full multi-shot sequence, no manual stitching
- LTX 2.5 keeps character, environment, and voice consistent between cuts
- Built for narratives, ads, and any LTX 2.5 project that cuts between shots
Diffusion Fidelity Rendering
Instead of locking every scene to one compression rate, LTX 2.5 dynamically allocates compute by scene complexity — detail where it matters, efficiency everywhere else.
- LTX 2.5's new diffusion video decoder replaces the VAE reconstruction stage
- Sharper faces, textures, and on-screen text with fewer artifacts
- Cleaner motion in demanding scenes with complex camera moves
Gemma 4 12B Text Encoder & Prompt Enhancer
LTX 2.5 pairs a custom Gemma 4 12B text encoder with a built-in prompt enhancer, so complex prompts — multiple characters, camera moves, lighting, actions — survive encoding intact.
- The LTX 2.5 encoder keeps longer, denser prompts intact through generation
- The LTX 2.5 prompt enhancer expands short prompts into cinematic instructions
- An optional duration predictor sets the clip length from the prompt
Fast Distilled Checkpoint & Quantized Builds
The improved LTX 2.5 distilled model runs a fixed 8-step schedule at CFG 1, keeping most of the full model's quality while cutting generation time dramatically.
- 8-step distilled inference with far lower compute per clip
- ComfyUI int8 and NVFP4 builds plus runtime FP8 casting for consumer GPUs
- Community GGUF quantizations of the LTX 2.5 transformer for low-VRAM setups
What You Can Create with LTX 2.5
LTX 2.5 suits everyone who needs connected scenes, synced audio, and open weights.
Multi-Scene Narratives for Storytellers
Short-film and web-series creators can prompt a full sequence of connected shots in one generation, with character, lighting, and voice holding across every cut.
Multi-Beat Ads for Marketing Teams
Ad teams draft multi-shot product stories fast — establishing shot, close-up, callout — as a single LTX 2.5 generation instead of several stitched clips.
Self-Hosted Generation for Developers
Independent developers run the open LTX 2.5 weights on their own infrastructure with no per-generation billing.
Product Animation for E-Commerce Teams
Animate still product shots into moving scenes with image-to-video, keeping lighting and environment consistent with the original.
Audio-to-Video for Music Creators
Feed a soundtrack and LTX 2.5 generates synchronized video and audio together, with lip-sync, motion, and sound aligned in one pass.
Consistent Character Work for Animators
With LTX 2.5, animators testing character designs hold the same face, wardrobe, and voice across multiple shots in a single run.
How to Use LTX 2.5
Generate connected AI video with LTX 2.5 in three simple steps — online here or locally via ComfyUI.
Enter Your Prompt
Describe the video you want in natural language. For multi-shot scenes, describe each shot in sequence — the LTX 2.5 prompt enhancer keeps complex instructions together.
Generate Your Clip
Click generate and the LTX 2.5 distilled schedule returns your clip quickly. Multishot prompts produce the full connected sequence in a single run, audio included.
Download and Refine
Download your LTX 2.5 clip ready to publish, or iterate on the prompt to refine composition, motion, and pacing.
LTX 2.5 vs Other AI Video Models
How LTX 2.5 compares to previous LTX releases and other open-weight video models.
LTX 2.5 vs LTX 2.3
LTX 2.3 generates a single continuous shot per run. LTX 2.5 adds native multishot — connected shots holding character, environment, lighting, voice, and style across cuts — plus a sharper decoder, a new Gemma 4 12B text encoder, and a faster distilled checkpoint. Note that 2.5 files and LoRAs do not carry over from 2.3.
LTX 2.5 Distilled vs Dev
The LTX 2.5 distilled checkpoint runs a fixed 8-step schedule at CFG 1 for fast everyday generation; the dev checkpoint is fully trainable for fine-tuning. Most LTX 2.3 LoRAs run on LTX 2.5, but validate each adapter before production use.
LTX 2.5 vs MiniMax H3
MiniMax H3 is a 33B omni-modal model with native stereo audio at up to 2K. LTX 2.5 differentiates on native multishot continuity and its split, Comfy-aligned open-weight package. Both are open-weight audio-video models under community licenses.
LTX 2.5 Quantized Builds by Hardware
LTX 2.5 ships ComfyUI-only int8 builds, an NVFP4 build for Blackwell GPUs, and runtime FP8 casting, with community GGUF quantizations for low-VRAM cards.
크리에이터들의 LTX 2.3 사용 후기
수천 명의 영화 제작자, 마케터, 개발자가 매일 LTX 2.3 모델로 전문적인 비디오를 생성하고 있습니다.
Jordan Vance
“LTX 2.3은 후처리 없이 바로 타임라인에 넣을 수 있는 시네마틱 B롤을 생성합니다.”
Aisha Patel
“앱 목업을 업로드하면 이미지-비디오 모드가 몇 분 만에 완성된 워크스루 데모로 애니메이션화해 줍니다.”
Rachel Kim
“기존 스튜디오 비용의 일부만으로 카탈로그 규모의 제품 비디오를 생성합니다. ROI가 엄청납니다.”
Jordan Vance
“LTX 2.3은 후처리 없이 바로 타임라인에 넣을 수 있는 시네마틱 B롤을 생성합니다.”
Aisha Patel
“앱 목업을 업로드하면 이미지-비디오 모드가 몇 분 만에 완성된 워크스루 데모로 애니메이션화해 줍니다.”
Rachel Kim
“기존 스튜디오 비용의 일부만으로 카탈로그 규모의 제품 비디오를 생성합니다. ROI가 엄청납니다.”
Linda Wu
“이 AI 모델로 비디오 에셋을 생성하기 시작한 이후 광고 캠페인 런칭 속도가 10배 빨라졌습니다.”
Marcus Rivera
“LTX 2.3이 제품 사진 한 장을 완전한 광고 영상으로 만들어 주었습니다. 스톡 영상 구독을 해지했습니다.”
Tom Eriksen
“이미지-비디오 기능이 스토리보드 프레임을 극장급 모션과 완벽한 연속성으로 살려냈습니다.”
Linda Wu
“이 AI 모델로 비디오 에셋을 생성하기 시작한 이후 광고 캠페인 런칭 속도가 10배 빨라졌습니다.”
Marcus Rivera
“LTX 2.3이 제품 사진 한 장을 완전한 광고 영상으로 만들어 주었습니다. 스톡 영상 구독을 해지했습니다.”
Tom Eriksen
“이미지-비디오 기능이 스토리보드 프레임을 극장급 모션과 완벽한 연속성으로 살려냈습니다.”
LTX 2.5 FAQ
More from LTX 2.5 AI
Try LTX 2.5 Free
Generate connected, audio-synced AI video with the LTX 2.5 model family — free credits for new users.