LTX-2.5 — Video Generation AI Model

LTX-2.5 generates picture and sound together from a prompt, and an optional image becomes the first frame. Runs the official two-stage distilled recipe: generation at half resolution, a x2 latent upscale, then a short refine at full resolution.

Try in Studio

What it does

LTX-2.5 is Lightricks' open-weight video and audio model. A prompt alone gives a clip with its own synchronized soundtrack; add an image and it becomes the first frame, animated from there.

This node runs the distilled checkpoint as the official DistilledPipeline does: eight steps at half size, a x2 latent upscale, then three steps at full size, with no guidance. Detailed prompts work best; the optional Gemma 4 E2B-it enhancer expands a short one.

When to use it

A video and a matching soundtrack from one prompt
Animating a still, with the image as the video's first frame
Landscape or portrait output: 1920x1088 (default), 1088x1920, 1536x1024 or 1024x1536
Five-second clips (121 frames) at 24 frames per second
Expanding a short idea into a detailed prompt with the Gemma 4 E2B-it prompt enhancer

Details

CategoryVideo Generation
Year2026
LicenseLTX-2.x Community License

Compliance & Provenance

ProviderLightricks (open weights) · Specialized
EU AI Act RiskLimited Risk
Art. 50 TransparencyRequired — AI-generated outputs are marked

Inputs & Outputs

TextInputstring: Text prompt describing the video and its sound
ImageInputimageoptional: Optional first-frame image. When given, it becomes the first frame and the model animates from it (image-to-video); it is resized and centre-cropped to the chosen resolution. Without it the whole clip comes from the prompt (text-to-video).
VideoOutputvideo: Generated 24 fps video with its synchronized soundtrack muxed in

Parameters

resolutionstringdefault 1920x1088 (16:9)options: 1920x1088 (16:9), 1088x1920 (9:16), 1536x1024 (3:2), +1 more

Resolution (W×H, ratio).

Output size, width x height. '1920x1088 (16:9)' (default): Lightricks staff call upscaling 544x960 to 1088x1920 (height x width) 'the most common scenario' (Hugging Face discussion Lightricks/LTX-2.5#53), the model card's two-stage example starts from 544x960, and Lightricks says the model 'is geared towards 1080p and up' (Lightricks/LTX-2.3#59). '1536x1024 (3:2)' is the default output size of the official ltx-pipelines (twice its 768x512 stage-1 size). '1088x1920 (9:16)' and '1024x1536 (2:3)' are those two sizes turned portrait; they are not official examples, but the official constraint is only that both sides are multiples of 32 (64 for two-stage, since stage 1 runs at half size), with no orientation limit.

num_framesintegerdefault 121 frames (~5s @ 24fps)options: 121 frames (~5s @ 24fps)

Frame count (8k+1).

Total frames to generate. LTX needs 8k+1 frames; 121 frames is about 5 seconds at 24 fps: the model card's example value and the default in the official ltx-pipelines parameters (the distilled CLI instead predicts the length when it is omitted, which this node does not use). The soundtrack covers the same duration.

enhance_promptbooleandefault falseadvanced

Expand a short prompt with the prompt enhancer

When true, the separate Gemma 4 E2B-it prompt enhancer rewrites the prompt before generation, expanding a short prompt into richer cinematic instructions (it also sees the image when one is attached). It is loaded and released within the request, adding about 10 GB of weights to read. Off by default, as in the official pipelines where prompt enhancement is opt-in.

seedintegerdefault -1range -1–2147483647advanced

Random seed (-1 = random).

Random seed. -1 (default) picks a fresh random seed every run; set a fixed value to repeat a result.

Limitations

Prompt following depends heavily on prompting style; write a detailed prompt or turn on the prompt enhancer
Only the first frame can be set from an image, and the image is centre-cropped to the chosen resolution
About 71 GB of weights (81 GB with the prompt enhancer) are read from disk again for every clip

Alternatives in Video Generation

Cosmos3 Nano
NVIDIA world model for text-, image- and video-to-video, with optional synced audio.
Helios Base
On-device text-to-video, up to 60s @ 24fps. 3-10 min.
Helios Distilled
Fast distilled text-to-video, fixed 640x384. 1-5 min.
LatentSync 1.6 (Lip Sync)
Re-sync a talking video's lips to a new voice.
LTX-2.5 Audio-to-Video
Make a talking video from one picture and a voice.
MiniMax-H3 Ref2VA
MiniMax-H3 Ref2VA (omni-reference) generates a 5-15s 24fps video with native stereo audio from a prompt plus a reference image and a reference clip, that clip's own soundtrack included. Video and its soundtrack come out of one denoising loop, so lip movement and sound land in sync. ~144GB bf16 weights — offloaded component-by-component, minutes-scale per clip.
Wan2.2 TI2V 5B
Wan2.2 dense 5B unified text+image-to-video. Text only → 5s video from prompt; text + image → image as the first frame of the video.

Resources

All AI models