How should I prompt LTX‑2.3 for the best motion?

Describe Core Actions as they unfold over time (e.g., “the camera slowly dolly‑zooms in as leaves drift across the foreground”) and list concrete Visual Details you want to preserve (composition, lighting, color, subject attributes). Keep directives concise and unambiguous.

What image inputs work best?

Clean, well‑lit images with clear subject separation tend to animate more coherently. Use a resolution that matches your VRAM and the workflow’s defaults, and keep the aspect ratio consistent across input and output to avoid stretching.

How do I control clip length and smoothness?

Clip length is a function of the number of generated frames and the SaveVideo FPS. Increase frames for longer clips and adjust FPS for playback feel. If the LTX‑2.3 node exposes frame count or duration, change it there; then set the desired FPS in SaveVideo.

My results change between runs. How can I make them repeatable?

Set and reuse a fixed seed in the LTX‑2.3 node (if available), keep the prompt identical, and avoid changing other generation parameters. Changing the seed is useful when you want controlled variations.

AI on the Lot 2026 - LTX-2.3 - i2v

This ComfyUI workflow turns a single still image into a short, animated video using Lightricks’ LTX‑2.3 image‑to‑video model. The pipeline is straightforward: LoadImage brings your source frame into the graph, a custom LTX‑2.3 node (2454ad83-157c-40dd-9f19-5daaf4041ce0) synthesizes motion conditioned on that image and a text prompt, and SaveVideo assembles the generated frames into a playable clip. MarkdownNote nodes inside the graph provide in‑context guidance and prompt tips so you can follow along during the AI on the Lot 2026 ComfyUI 101 workshop.

Technically, the LTX‑2.3 node encodes your input image into the model’s latent space and generates a temporally coherent sequence guided by your prompt. You control the character and pace of motion by describing events over time ("Core Actions") and specifying concrete "Visual Details" you want to preserve, as recommended in the LTX‑2.3 prompting tips. Typical controls exposed by the node include the text prompt, seed for reproducibility, and clip characteristics such as number of frames and/or FPS (exact parameters depend on the node build). SaveVideo then writes the frames to disk at your chosen frame rate. For model weights, use Lightricks/LTX‑2.3 from Hugging Face (see the links provided), and follow the node’s loader field to point to the correct files.

FAQ

Frequently Asked Questions

View all workflows