Nodes/VRGameDevGirl Video Enhancement Nodes/VRGDG LTX Speed Character Lora Training
ComfyUI Node

VRGDG LTX Speed Character Lora Training

Runs the LTX trainer with a fast character-LoRA preset using dynamic IMAGE and caption inputs.

By vrgamegirl19·Created about a year ago·Updated 22 days ago· 533
VRGDG LTX Speed Character Lora Training
  • model
  • image1
  • image2
  • image3
  • image4
  • image5
  • image6
  • image7
  • image8
  • image9
  • image10
  • image11
  • image12
  • image13
  • image14
  • image15
  • image16
  • image17
  • image18
  • image19
  • image20
  • model
  • latest_state_path
  • log_path
  • video_filename_prefix
  • output_name
  • completed_steps
  • total_target_steps
workspace_dirA:/MUSUBI/Training/SpeedCharacterLoraTraining
run_nameSpeedCharacterLoraTrainingRun
output_nameSpeedCharacterLoraTraining
image_count4
resolution_width1256
resolution_height1256
blocks_to_swap0
clear_memory_before_gemmatrue
cache_strategyauto
strength_model1.00
musubi_rootA:/MUSUBI/musubi-tuner-ltx2
ltx2_checkpointA:/MUSUBI/models/ltx2/ltx-2.3-22b-dev.safetensors
gemma_rootA:/MUSUBI/models/gemma3
gemma_recovery_modefalse
gemma_load_in_4bitfalse
caption_1
caption_2
caption_3
caption_4
caption_5
caption_6
caption_7
caption_8
caption_9
caption_10
caption_11
caption_12
caption_13
caption_14
caption_15
caption_16
caption_17
caption_18
caption_19
caption_20
CategoryVRGDG/Training

Inputs (56)

NameTypeDefaultDescription
modelMODEL
workspace_dirSTRINGA:/MUSUBI/Training/SpeedCharacterLoraTrainingWorkspace folder for cache, output, logs, config, and the managed dynamic dataset.
run_nameSTRINGSpeedCharacterLoraTrainingRunRun name used for logs.
output_nameSTRINGSpeedCharacterLoraTrainingLoRA output name used for checkpoints and downstream preview naming.
image_countINT41–20How many dynamic image inputs and caption fields to show.
resolution_widthINT125664–4096Training bucket width. Pick the resolution preset you want to train at.
resolution_heightINT125664–4096Training bucket height. Pick the resolution preset you want to train at.
blocks_to_swapINT00–64How many transformer blocks to swap to CPU. 0 is fastest if VRAM allows it.
clear_memory_before_gemmaBOOLEANtrueClears Comfy and CUDA memory before the Gemma cache stage.
cache_strategyCOMBOautoCache behavior. auto reuses cache when present, force rebuilds, skip bypasses cache creation.
strength_modelFLOAT1.00-100–100Strength used when applying the newest trained LoRA back onto the returned MODEL.
musubi_rootSTRINGA:/MUSUBI/musubi-tuner-ltx2Root folder of your musubi install.
ltx2_checkpointSTRINGA:/MUSUBI/models/ltx2/ltx-2.3-22b-dev.safetensorsPath to the LTX-2.3 DiT checkpoint.
gemma_rootSTRINGA:/MUSUBI/models/gemma3Path to the Gemma model root used by this preset.
gemma_recovery_modeBOOLEANfalseExperimental. If enabled, the node will try alternate Gemma cache settings after the normal path fails.
gemma_load_in_4bitBOOLEANfalseLoads Gemma in 4-bit mode during text encoder caching. This lowers VRAM more than 8-bit, but can be slower or less stable.
image1optIMAGE
image2optIMAGE
image3optIMAGE
image4optIMAGE
image5optIMAGE
image6optIMAGE
image7optIMAGE
image8optIMAGE
image9optIMAGE
image10optIMAGE
image11optIMAGE
image12optIMAGE
image13optIMAGE
image14optIMAGE
image15optIMAGE
image16optIMAGE
image17optIMAGE
image18optIMAGE
image19optIMAGE
image20optIMAGE
caption_1optSTRING
caption_2optSTRING
caption_3optSTRING
caption_4optSTRING
caption_5optSTRING
caption_6optSTRING
caption_7optSTRING
caption_8optSTRING
caption_9optSTRING
caption_10optSTRING
caption_11optSTRING
caption_12optSTRING
caption_13optSTRING
caption_14optSTRING
caption_15optSTRING
caption_16optSTRING
caption_17optSTRING
caption_18optSTRING
caption_19optSTRING
caption_20optSTRING

Outputs (7)

NameTypeDescription
modelMODEL
latest_state_pathSTRING
log_pathSTRING
video_filename_prefixSTRING
output_nameSTRING
completed_stepsINT
total_target_stepsINT