mirror of
https://github.com/huggingface/diffusers.git
synced 2026-10-23 09:09:01 +03:00
7e4383c0a2
* [LTX-2] Add `conditioning_frame_rate` to decouple the positional time base from the playback frame rate LTX-2 places video tokens on a time axis in seconds (`pixel_frame / fps` in the RoPE coordinates), derived from `frame_rate`. Adapters such as the LTX-2.5 Slow-Motion-Control LoRA are conditioned on a different rate (`fps / speed`) than the video plays at. Add an optional `conditioning_frame_rate` to the LTX-2 classic pipelines and modular blocks: it drives the video, keyframe and reference coordinates and the transformer's `fps`, while the audio length, the duration head and the returned video keep following `frame_rate`. `None` resolves to `frame_rate`, leaving current outputs unchanged. * [LTX-2] Replace `conditioning_frame_rate` with `motion_speed` `frame_rate` keeps one meaning, the playback rate; `motion_speed` (default 1.0, must be positive) is the speed of the motion relative to real time, and the positional time axis is conditioned on `frame_rate / motion_speed`. * [LTX-2] Add an end-to-end `motion_speed` example to the LTX-2 doc Image-to-video with the LTX-2.5 Slow-Motion-Control LoRA at motion_speed 1.0 and 0.2 (same seed and caption), plus a note that two-stage generation must pass the same motion_speed to both stages.