LTXVideo

Adds a conditioning frame or a video at a specific frame index. This node is used to add a keyframe or a video segment which should appear in the generated video at a specified index. It resizes the image to the correct size and applies preprocessing to it.

Adds a conditioning frame/video at a specific frame index with per-guide attention strength control. Same preprocessing as LTXVAddGuideAdvanced, plus attention_strength and optional spatial attention_mask.

Adds a conditioning frame or a video at index 0. This node is used to add a keyframe or a video segment which should appear in the generated video at index 0. It resizes the image to the correct size and applies preprocessing to it.

Applies image conditioning to the first frames of an existing latent. Creates a noise mask to control conditioning strength.

Concatenates two video latents along the frames dimension. latents1 and latents2 must have the same dimensions except for the frames dimension.

Dilates a latent by a grid size.

Selects a range of frames from the video latent. start_index and end_index define a closed interval (inclusive of both endpoints).

Applies multiple masks to a video latent. masks can be 2D, 3D, or 4D tensors. If there are fewer masks than frames, the last mask will be reused.

Fixed minimal placeholder video latent for audio-only (text-to-audio) generation - a single 64x64 frame. LTX-2 needs a video latent at input index 0, but with LTXV Audio Only Model it is never attended to, so the dimensions are fixed (and hidden) to avoid confusion. Join it with the audio latent via LTXVConcatAVLatent.

Run the LTX-2 audio/video model in audio-only mode for text-to-audio. Skips the video stream and disables the audio<->video cross-attention, so the generated audio does not depend on the video latent. The sampler still needs a minimal dummy video latent (EmptyLTXVLatentVideo at 64x64 length 1, joined with LTXVConcatAVLatent); it is never attended to. Remove or bypass this node to restore normal joint audio+video generation.

The APG Guider implements Adaptive Projected Guidance (APG). Reference: https://arxiv.org/abs/2410.02416.

Enhance text prompts using Gemma 3 VLLM for improved video generation.

Defines the blocks to apply the STG to.

Enhances text prompts for image generation using LLMs. Optionally incorporates reference images to create more contextually relevant descriptions.

Implements Spatiotemporal Skip Guidance (STG), a training-free method enhancing transformer-based video diffusion models by selectively skipping layers during sampling. This approach improves video quality without sacrificing diversity or motion fidelity.Reference: https://arxiv.org/abs/2411.18664.

The Advanced STG Guider implements sophisticated techniques for controlling the denoising process: It creates a dynamic mapping from scheduler-defined sigma values to CFG and STG (Spatio-Temporal Skip Guidance [1]) parameters. This approach establishes distinct sigma value ranges that operate independently of step numbers, allowing precise control over: • CFG scale • STG scale and rescale factors • STG attention layer skipping patterns The guider also supports: • CFG-Zero* [2] rescaling, which dynamically adjusts negative predictions based on the dot product between positive and negative signals • Threshold-based noise prediction zeroing for steps with sigma values exceeding a specified threshold For example if the sigma ranges are defined as [1.0, 0.9, 0.85, 0.6] and the CFG values are defined as [4, 3, 2, 1] and STG scale values are defined as [2, 2, 2, 1] and STG rescale values are defined as [1, 1, 1, 1] and STG layers indices are defined as [[14, 17], [14, 16], [14], [14]], then the guider will: - use CFG=4, STG scale=2, STG rescale=1 and STG layers indices = [14, 17] for sigma in the range (0.9, 1.0] - use CFG=3, STG scale=2, STG rescale=1 and STG layers indices = [14, 16] for sigma in the range (0.85, 0.9] - use CFG=2, STG scale=2, STG rescale=1 and STG layers indices = [14] for sigma in the range (0.8, 0.85] - use CFG=1, STG scale=1, STG rescale=1 and STG layers indices = [14] for sigma in the range (0.6, 0.8] The guider will use the same parameters for the same sigma values, regardless of the step number. References: [1] https://arxiv.org/abs/2411.18664 [2] https://arxiv.org/abs/2503.18886

Implements Spatiotemporal Skip Guidance (STG), a training-free method enhancing transformer-based video diffusion models by selectively skipping layers during sampling. This approach improves video quality without sacrificing diversity or motion fidelity.Reference: https://arxiv.org/abs/2411.18664.

Preprocess masks to be used for masking latents in the LTXVideo model.

Loads an LTXV Audio VAE checkpoint with dependency support. Connect 'dependencies' to a previous loader's output to ensure sequential loading and reduce peak VRAM usage.

Loads a diffusion model checkpoint with dependency support. Connect 'dependencies' to a previous loader's output to ensure sequential loading and reduce peak VRAM usage.

Sets the audio and video mask by time.