lightricks/ltx-2.3-fast

Lightning-fast video generation with portrait support, camera controls, and synchronized audio. Up to 20 seconds at 1080p, 4K at 50 FPS.

940 runs

LTX 2.3 Fast

LTX 2.3 Fast is the speed-optimized variant of Lightricks’ LTX-2.3 video generation model. It generates videos with synchronized audio faster than real-time — ideal for rapid prototyping, mobile workflows, and high-volume production.

What’s new in 2.3

  • Portrait video — native 9:16 aspect ratio alongside 16:9 landscape
  • More frame rates — 24, 25, 48, and 50 FPS
  • Longer videos — up to 20 seconds at 1080p (with 24 or 25 FPS)
  • End frame interpolation — provide both a start and end frame image to generate video that transitions between them
  • Camera motion controls — dolly, jib, static, focus shift, and more
  • Improved quality — better VAE, larger text connector for improved prompt adherence, fewer audio artifacts

Inputs

Input Description Default
prompt Text description of the video to generate (required)
image First frame image for image-to-video None
last_frame_image End frame for interpolation (requires image) None
resolution Video resolution: 1080p, 2k, or 4k 1080p
duration Length in seconds: 6, 8, 10, 12, 14, 16, 18, or 20 6
aspect_ratio 16:9 (landscape) or 9:16 (portrait) 16:9
fps Frame rate: 24, 25, 48, or 50 25
camera_motion Camera effect: dolly_in, dolly_out, dolly_left, dolly_right, jib_up, jib_down, static, focus_shift, or none none
generate_audio Generate synchronized audio true

Durations longer than 10 seconds are only available at 1080p with 24 or 25 FPS.

Output

MP4 video file with synchronized audio (when generate_audio is enabled).

Prompting tips

  • Write one continuous paragraph in present tense
  • Be specific about camera angles, lighting, and movement
  • Include atmospheric details: lighting quality, weather, textures
  • Describe actions with active verbs: “walks”, “pans slowly”, “zooms in”
  • Longer, more detailed prompts produce better results

Also available

  • LTX 2.3 Pro — higher visual fidelity, plus audio-to-video, retake, and extend capabilities
Model created
Model updated