LTX 2.3 Outpaint

LTX 2.3 Outpaint is an IC-LoRA for extending the canvas of an input video. Instead of generating a new clip from a prompt, you letterbox the source video into a target aspect ratio and leave the new regions pure black. The model treats those black regions as the area to generate, then fills the sides, top, bottom, or any combination with content that matches the original footage.

That makes it useful when the creative problem is format adaptation. A horizontal clip can become vertical for short-form video, a tight crop can become wider for a landing page hero, or a square asset can be extended without throwing away the center frame.

Start with the public LTX 2.3 Outpaint Space. Upload a short clip with a clear scene, pick the target aspect ratio, preview the letterboxed frame, and generate. On the first run, watch the boundary between the original video and the generated regions. The useful result is not just a filled border; it is whether motion, lighting, and scene structure stay believable over time.

If you want to reproduce it locally, use the LoRA weights with the IC-LoRA workflow from the LTX ComfyUI repository or the workflow file linked on the model page. This is a GPU workflow. For very dark or night footage, use the model card's gamma correction trick so the pure-black fill area is not confused with real dark content.

The black-region signal is also the main failure point. Very dark scenes can confuse the model unless you brighten the source before generation and reverse the correction after. Local use also depends on the LTX 2.3 stack and a capable GPU.

• Convert horizontal footage into vertical, square, or ultrawide versions without simple cropping.
• Extend product, travel, or event clips when the original framing is too tight.
• Prepare one source video for several social and web placements.
• Test whether video outpainting can preserve motion across the generated border.

Try the demo

View model page