FmlfLtx23

## FMLFLTX 2.3 Prepares video and audio latents for LTX Video 2.3 from up to 6 guide images. Images are automatically placed at evenly-spaced positions: - **image_1** → frame 0 (First Frame). - **image_N** → frame (length - 1) (Last Frame — hard anchor, re-embedded by sampler in Pass 2). - **image_2 … image_N-1** → evenly spaced at `round((length-1) * i / (N-1))` for i = 1..N-2. Connect `width`, `height`, `length`, and `fps` from **LTX Resolution Selector**. `segment_lengths` output is calculated between all frames except the last anchor, and can be connected directly to **PromptRelayEncode**. --- ### Inputs | Pin | Description | |---|---| | `video_vae` | Video VAE for image encoding. | | `audio_vae` | Audio VAE for empty audio latent creation. | | `img_compression` | JPEG pre-compression strength (0 = disabled). Matches LTXVPreprocess. | | `width` | Latent width in pixels — connect from LTX Resolution Selector. | | `height` | Latent height in pixels — connect from LTX Resolution Selector. | | `length` | Frame count — connect from LTX Resolution Selector. | | `fps` | Frames per second — used for audio latent duration. Connect from LTX Resolution Selector. | | `batch_size` | Batch size (default 1). | | `image_1 … image_6` | Guide images (optional). Connect only the slots you need. | | `strength_1 … strength_6` | Conditioning strength per image (1.0 = fully conditioned, 0.0 = ignored). | ### Outputs | Pin | Description | |---|---| | `latent` | Combined video + audio NestedTensor — for direct sampler input (legacy). | | `video_latent` | Video latent only → connect to **SamplerLTXV_2.3** `video_latent`. | | `audio_latent` | Empty audio latent → connect to **SamplerLTXV_2.3** `audio_latent`. | | `segment_lengths` | Pixel-space frame counts per segment → connect to **PromptRelayEncode**. |