ComfyUI Node: LTX25 Ultimate Upscale

LTX25 Ultimate Upscale:

The LTX25UltimateUpscale node is designed to enhance video content by leveraging the LTX2.5 model, which is a sophisticated 2x latent spatial upscaler. This node is particularly beneficial for AI artists looking to upscale video content while maintaining high quality and consistency across frames. It integrates a comprehensive pipeline that includes temporal and spatial processing, allowing for seamless upscaling and interpolation to target dimensions. The node supports advanced features such as temporal splitting, spatial stitching, and the use of reference guides to ensure scene consistency. By utilizing this node, you can achieve high-quality video upscaling with minimal manual intervention, making it an essential tool for enhancing video projects.

LTX25 Ultimate Upscale Input Parameters:

upscale_model

This parameter specifies the LTX2.5 latent spatial upscaler model, which is responsible for the 2x upscaling of video content. It is crucial to place the appropriate model files in the latent_upscale_models folder to ensure proper functionality.

vae

The VAE (Variational Autoencoder) input is used for normalizing and un-normalizing per-channel statistics during the upscaling process. It is important to use the video VAE, not the audio VAE, to ensure accurate video processing.

width

This parameter defines the target overall pixel width of the upscaled frame. The value is snapped to a multiple of 32, aligning with the LTX VAE 32x grid. The default value is 1280, with a minimum of 64 and a maximum of 4096. It must match the conditioning's generation size to ensure proper upscaling.

height

Similar to the width parameter, this defines the target overall pixel height of the upscaled frame. It is also snapped to a multiple of 32, with a default value of 704, a minimum of 64, and a maximum of 4096. Consistency with the conditioning's generation size is required for optimal results.

LTX25 Ultimate Upscale Output Parameters:

upscale_param

This output provides the LTX2.5 upscale settings that are consumed by the LTX25UltimateUpscale node. It encapsulates the configuration used during the upscaling process, ensuring that the settings are correctly applied to achieve the desired video enhancement.

LTX25 Ultimate Upscale Usage Tips:

• Ensure that the upscale_model and vae inputs are correctly configured with the appropriate model files to avoid processing errors.
• Match the width and height parameters with the conditioning's generation size to maintain consistency and achieve optimal upscaling results.
• Utilize the reference guides feature to maintain scene consistency across independently sampled chunks, especially when working with complex video projects.

LTX25 Ultimate Upscale Common Errors and Solutions:

" LTX25 Ultimate Upscale expects an LTX2.5 AV latent (nested video [B,128,T,H,W] + audio)"

• Explanation: This error occurs when the input does not match the expected LTX2.5 AV latent format.
• Solution: Ensure that the input is a properly formatted LTX2.5 AV latent, including both video and audio components.

" LTX25 Ultimate Upscale expects a single-video latent (batch 1)"

• Explanation: The node is designed to process a single video latent at a time.
• Solution: Adjust the input to ensure that only one video latent is being processed in the batch.

"Upscale width ({up_w}) must be >= tile_width ({tile_w})"

• Explanation: The target upscale width is smaller than the spatial tile size.
• Solution: Increase the upscale width to be equal to or greater than the tile width.

"Upscale height ({up_h}) must be >= tile_height ({tile_h})"

• Explanation: The target upscale height is smaller than the spatial tile size.
• Solution: Increase the upscale height to be equal to or greater than the tile height.