Transforms still images into smooth, looping cinemagraphs with selective motion while keeping the rest of the scene frozen.

It is based on the LTX-2.3 foundation model.

(This is the blurb from their HF model page, I have no idea how well it actually loops)

NOTE: I've tried this here on tensor, and with tensors 'fast' implementation without CFG settings, more steps and access to proper negative prompting it does not seem to produce static camera. However it DOES produce more subtle localized motion even with tensors 'fast' version of ltx, which is something, and potentially useful given video models tendency to produce excessive intense motion, and it probably does do something like the model page claims if you can run it locally.

Instructions:

Recommended Settings

• LoRA strength / weight: 0.7–3.0

• Spatial Guidance (STG): Use stg_v mode, scale 1.0, targeting block 29 for enhanced motion control and boundary preservation.

• Inference steps: 30 (from training validation)

• Guidance scale: 4.0

• Prompting: See examples below. The model responds well to explicit descriptions of what should move and what should remain still. Emphasize "locked-off static camera" and "seamless natural loop" for best results.