Alternative download:
I2V: https://pastebin.com/A6Qsj8F2
T2V: https://pastebin.com/mS5zaJCb
This is a supplementary page for my Seiichi Kinoshita LTX2 character Lora with a collection of T2V and I2V workflows for ComfyUI which can be immediately used with my Lora.
The following Workflows are attached:
LTX2 Text2Video
LTX2 Image2Video
NOTE: some custom nodes are required to be downloaded - Comfy Manager should take care of it.
The workflows include:
LTX-2 19B DEV FP 8 base model + Gemma3 abliterated text encoder
Distilled Lora rank 175 (from Kijai)
My character Lora
Official LTX2 detailer Lora (optional)
NOTE: You can replace these with your own LTX model files or google the missing file names for download.
Explanation for included features:
DEV FP8 + Distilled Lora is a decent compromise between speed and quality.
All nodes are color-coded. Green nodes - connect to 1st sampler / Red nodes - connect to 2nd sampler (upscaler)
The LTX2 detailer Lora can help with quality but it can also make the anime style worse. Either don't use it or only use it on first sampler
Negative prompts will only work if you set CFG above 1.0 (requires other workflow adjustments not covered in this article)
The workflow contains a context switch between generating only with 1st sampler or with 1st sampler + 2nd sampler.
Using 1st sampler only is useful to quickly check what was generated before upscaling phase
Enable NoUpscale 'no' + Enable Upscaler 'yes' = default full generation
Enable NoUpscale 'yes' + Enable Upscaler 'no' = generates video with only 1st sampler
Recommendation for parameters:
By default the workflow uses Distilled Lora on both 1st and 2nd sampler.
T2V
1st sampler step count: 8
(don't touch 2nd sampler Sigmas if you don't know what you are doing
1st and 2nd sampler CFG: 1.0
1st and 2nd sampler type: Euler
Frame rate: 25
Resolution: wide aspect ratios are preferred (1280x704 / 1920x1024) but square and portrait will also work. Generally speaking higher resolution = better quality
I2V
LTXVPreprocess - img_compression: 0
1st and 2nd sampler LTXVImgToVideoInplace strength: 1.0
1st sampler step count: 8
(don't touch 2nd sampler Sigmas if you don't know what you are doing
1st and 2nd sampler CFG: 1.0
1st and 2nd sampler type: Euler
Frame rate: 25
Character Lora strength: 0.8-1.0
Note: These workflows will of course work for anything LTX2 related outside of my character Lora. Just delete/bypass my character Lora node in that case.