Description
Adds multiple guide images from a batch to the latent at corresponding frame indices. Non-black (i.e., non-zero) images in the batch are used as guides. This node is part of the LTXV conditioning pipeline, enabling fine-grained control over per‑frame appearance by injecting reference images directly into the latent space.
Inputs
| Name | Type | Required | Description |
|---|---|---|---|
positive | CONDITIONING | Yes | Positive conditioning tensor (e.g., from a CLIP text encoder or other conditioning setup). |
negative | CONDITIONING | Yes | Negative conditioning tensor. |
vae | VAE | Yes | VAE model used to encode guide images into the latent space. |
latent | LATENT | Yes | Initial latent tensor (e.g., from an empty latent or previous LTXV stage). |
images | IMAGE | Yes | Batch of images. Non‑black images in this batch are treated as guides and applied to corresponding frame indices. Black (all‑zero) images are ignored. |
strength | FLOAT (0‑10) | Yes | Strength factor applied to all guide images. Default: 1.0. Higher values increase the influence of the guides. |
Outputs
| Name | Type | Description |
|---|---|---|
positive | CONDITIONING | Updated positive conditioning with guide information injected. |
negative | CONDITIONING | Updated negative conditioning with guide information injected. |
latent | LATENT | Modified latent tensor containing guide‑injected per‑frame adjustments. |
Usage Notes
- The node expects the
imagesbatch to have the same number of frames as thelatenttensor (or at least up to the number of frames in the latent). Each non‑black image in the batch is used as a guide for the corresponding frame index. - A “black” image is defined as one where all pixel channels are zero. This allows you to include “placeholder” images in the batch that should not contribute any guide signal.
- The
strengthparameter scales the encoded contribution of all guides uniformly. To vary strength per guide, you can pre‑process the images (e.g., blend with black) before passing them to this node. - This node is intended for LTXV‑based workflows where precise per‑frame control via reference images is needed, such as frame‑consistent video generation or stylized animation.
- All three outputs must be connected to continue the conditioning pipeline (typically to a sampler or further LTXV nodes). The updated conditioning tensors carry the guide information, while the latent contains the spatially modified representation.
Comments
Sign in with GitHub to join the discussion.