LTX-2.5-A2Vid — Video Generation AI Model

What it does

LTX-2.5 A2Vid from Lightricks turns one image and a voice into a talking video. The voice drives the motion: mouth shapes, head movement and hand gestures follow it, and the original audio is muxed into the result untouched.

By default it picks a vertical (768x1344), landscape (1344x768) or square (1024x1024) canvas to match your image, or you can choose one; it renders at 24 fps, for voices up to 30 seconds. Generation takes a few minutes. Put the subject on a flat green background to composite them later.

When to use it

• A presenter clip for an ad from one photo and a TTS voiceover
• Lip-synced talking heads for product explainers
• A spokesperson video in another language from the same photo
• Green-screen presenters to lay over screen recordings

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

ImageInput · image

First frame: the person who will speak (a flat green background composites best)

SoundInput · sound

The voice to speak (0.5-30 s); kept as-is in the result

TextInput · string · optional

Optional description of the motion and scene

VideoOutput · video

The talking video at 24 fps with the voice muxed in

Parameters

resolutionstring · default auto · options: auto, 768x1344, 1344x768, 1024x1024

Output size (auto follows the image)

auto (default) picks the canvas closest to the image's aspect ratio - the image is centre-cropped to the canvas (fit 'crop'), so a mismatched canvas cuts the subject; fit 'pad' pads it instead. Or force 768x1344 (vertical), 1344x768 (landscape) or 1024x1024.

stepsinteger · default 30 · range 10–60 · advanced

Stage-1 denoising steps (10-60)

Stage-1 denoising steps (10-60, default 30, the checkpoint default). Stage 2 always uses the distilled schedule. Cost is roughly linear in this value (~5 s per step).

seedinteger · default -1 · advanced

Random seed (-1 = random)

Random seed for reproducibility; -1 (default) picks a fresh one.

lead_in_secfloat · default 0.8 · range 0–2 · advanced

Warm-up before the first word so the lips move from the start; trimmed away (0 = off)

Silent warm-up in seconds placed before the voice during generation (0-2, default 0.8). The model keeps roughly the first 0.5-0.8 s close to the still image, so a voice that starts immediately gets almost no lip motion on its first word; with the warm-up the face is already moving when the voice starts, so the lips move from the first frame. The warm-up frames are trimmed away: output timing and length are unchanged. 0 disables it. Voices near the 30 s cap get a shorter warm-up so warm-up + voice stays within 30 s.

fitstring · default crop · options: crop, pad · advanced

Crop the picture to the video frame, or pad it so nothing is cut

How the image meets the canvas. 'crop' (default): the image is centre-cropped to the canvas, so a picture whose aspect differs loses its edges (an 832x1024 character on the 1024x1024 canvas lost the top of its hair). 'pad': the image is first grown to the canvas aspect with its border colour - extra height on top only (a presenter cut off by the bottom edge stays on it), extra width split between the sides - so nothing is cut. Meant for a still on a flat backdrop such as the chroma green from background removal; on a busy photo the padding shows as a plain band.

Tags

• audio to video
• talking head
• lip sync
• avatar
• presenter
• video generation
• ltx

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

  Lightricks LTX-2.5 distilled 22B: text, or text plus a first-frame image, to a video with a synchronized soundtrack generated alongside it.

• 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

• github
• huggingface