A simple workflow for using the spatial upscale stage in LTX 2.3 to upscale any video. The LTX2.3 model upscales videos but adds its minor details and variations to the video.
V2.0
Played around with sigmas for the upscale. Consistency to original video has been improved significantly.
No need to bypass individual nodes anymore. Use the Groups bypasser node to enable or disable the generate audio module
Works better with prompts describing the video or the audio you want
LoRA dependency is reduced, but will do better with them for specific scenarios
V1.0
LTX2.3 model adds realism to the video while upscaling with added details
The workflow can upscale to 1.5x or 2x, depending on the spatial model used (Download from LTX2.3 Huggingface repo)
One of these nodes needs to be bypassed depending on whether the original video has audio - LTXV Audio VAE Encode or LTXV Empty Latent Audio
A LoRA might be needed based on the content of the video
A brief description of video for positive prompt or can be left empty
Description
V2.0
Played around with sigmas for the upscale. Consistency to original video has been improved significantly.
No need to bypass individual nodes anymore. Use the Groups bypasser node to enable or disable the generate audio module
Works better with prompts describing the video or the audio you want
LoRA dependency is reduced, but will do better with them for specific scenarios
FAQ
Comments (5)
Looks interesting.. will this also add audio to a soundless video while it is upscaling it?
If Generate audio group is enabled, LTXV generated audio will be added. If Generate audio group is disabled and the input video has audio, the output will have same audio.
@Ravix_of_fourhorn amazing, two birds one stone, will definitely try this, thanks for the WF!
@Ravix_of_fourhorn btw, 1.1 version of the distilled transformer was released.. if I use that in this workflow, i have to bypass the distilled lora, right?
@artificialotaku Only 1 distilled model - whether its the distilled transformer model or LoRA model