ComfyUI Node
VRGDG LTX Audio Video LoRA Train Chunk
Runs one LTX-2.3 audio-video LoRA training chunk using musubi-tuner on short videos with embedded audio.
VRGDG LTX Audio Video LoRA Train Chunk
- model
- model
- latest_state_path
- log_path
- video_filename_prefix
- output_name
- completed_steps
- total_target_steps
◄dataset_videos_dir►
◄workspace_dir►
◄run_nameLTXAVChunkRun►
◄output_nameLTXAVChunkRun►
◄av_profilefast_short_clip►
◄use_profile_defaultstrue►
◄resolution_width768►
◄resolution_height512►
◄target_frames17,33►
◄frame_extractionhead►
◄max_frames49►
◄target_fps25.0►
◄steps_per_run100►
◄total_target_steps400►
◄network_dim16►
◄network_alpha16►
◄blocks_to_swap2►
◄separate_audio_bucketstrue►
◄clear_memory_before_gemmatrue►
◄lora_target_presett2v►
◄fp8_basetrue►
◄fp8_scaledtrue►
◄learning_rate_preset1e-4►
◄learning_rate0.0001►
◄num_repeats8►
◄cache_strategyauto►
◄copy_latest_to_comfy_lorastrue►
◄keep_only_comfy_lorafalse►
◄strength_model1.00►
◄create_captionsfalse►
◄caption_text►
◄add_trigger_wordfalse►
◄trigger_text►
◄musubi_rootA:/MUSUBI/musubi-tuner-ltx2►
◄ltx2_checkpointA:/MUSUBI/models/ltx2/ltx-2.3-22b-dev.safetensors►
◄gemma_rootA:/MUSUBI/models/gemma3►
◄gemma_load_in_4bitfalse►
◄gemma_recovery_modefalse►
CategoryVRGDG/Training
Inputs (39)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | Base model to return downstream with the latest trained LoRA optionally applied. | |
| dataset_videos_dir | STRING | Folder containing your training videos, or a parent folder that will be organized into a videos subfolder. | |
| workspace_dir | STRING | Working folder for cache, logs, config files, checkpoints, and training state. | |
| run_name | STRING | LTXAVChunkRun | Name prefix used for the log file. |
| output_name | STRING | LTXAVChunkRun | Name prefix used for saved LoRA files and state folders. |
| av_profile | COMBO | fast_short_clip | Training recipe switch. fast_short_clip is the default quick test for one short video. step200 keeps the same fast_short_clip recipe but sets a 200-step chunk budget with lighter repeats. balanced is a stronger middle ground. full_video trains on the whole clip and is the slowest. overnight_full_video is tuned for a longer clip using smaller frame windows, lower resolution, and light swapping so it can run overnight on a midrange GPU. |
| use_profile_defaults | BOOLEAN | true | When enabled, the selected profile controls the AV recipe bundle instead of the manual fields below. Turn this off if you want to hand-tune every value. |
| resolution_width | INT | 76864–8192 | Training bucket width in pixels. 768 is a good fast default for a single short clip. Smaller values like 512 are faster; larger values cost more VRAM and time. |
| resolution_height | INT | 51264–8192 | Training bucket height in pixels. 512 pairs well with 768x512 for a quick short-clip run. Use a lower height for faster experiments. |
| target_frames | STRING | 17,33 | Comma-separated frame buckets to train on. For a 10-second clip, 17 and 33 frames are a good fast starting point. The trainer rounds values to the required N*4+1 frame buckets. |
| frame_extraction | COMBO | head | How frames are selected from each video. head uses only the first target_frames window and is the fastest choice for a short single-clip LoRA. full uses the entire video, but is much slower and should be capped with max_frames. |
| max_frames | INT | 491–4096 | Safety cap used when frame_extraction is full. 49 keeps the run short and prevents a 10-second clip from turning into a long heavy sample. It does not affect head/chunk/slide/uniform. |
| target_fps | FLOAT | 25.01–240 | Target FPS used during caching/training. 25.0 is the standard LTX rate and works well for most short clips with audio. |
| steps_per_run | INT | 1001–100000 | How many steps to train before saving and stopping this chunk. 100 is a fast check-run default for a single short clip. |
| total_target_steps | INT | 4001–1000000 | Total training budget across all chunks. 400 is a fast starter value for one short video; increase later only if you need more adaptation. |
| network_dim | INT | 161–2048 | LoRA rank. 16 is a good quick-training default for one short clip. Use 32 if the result is too weak. |
| network_alpha | INT | 161–2048 | LoRA alpha. Usually keep this equal to the rank for a simple fast baseline. |
| blocks_to_swap | INT | 20–64 | How many transformer blocks to offload to CPU. Lower is faster if you have enough VRAM. Raise this only if you run out of memory. |
| separate_audio_buckets | BOOLEAN | true | Keeps audio and non-audio items in separate batches. Safe to leave on and useful if your dataset ever mixes audio and silent clips. |
| clear_memory_before_gemma | BOOLEAN | true | Unloads ComfyUI models and clears memory before Gemma caching. Leave this on unless you know your workspace is already empty. |
| lora_target_preset | COMBO | t2v | Which transformer layers get LoRA adapters. t2v is the fastest useful baseline. v2v adds FFN layers. full is the heaviest and produces the largest adapter. |
| fp8_base | BOOLEAN | true | Use the FP8 base-model loading path. Leave this on for speed and lower VRAM. |
| fp8_scaled | BOOLEAN | true | Quantize non-FP8 checkpoints into FP8 at load time. Turn this off if your checkpoint file is already FP8, usually visible from the filename. |
| learning_rate_preset | COMBO | 1e-4 | Quick preset for the training learning rate. 1e-4 is a strong starter value for a short 10-second clip. |
| learning_rate | FLOAT | 0.00011e-8–1 | Custom learning rate used only when the preset is set to Custom. |
| num_repeats | INT | 81–1000 | How many times each video-caption pair is repeated in the dataset. Higher repeats help a single clip produce enough training signal without needing a huge number of source videos. |
| cache_strategy | COMBO | auto | Auto rebuilds cache only when the node detects the cache is missing or stale. Force always rebuilds. Skip assumes the cache is already correct. |
| copy_latest_to_comfy_loras | BOOLEAN | true | Copies the latest Comfy-compatible LoRA into the ComfyUI loras folder after training. |
| keep_only_comfy_lora | BOOLEAN | false | If enabled, deletes the standard .safetensors LoRA files after a matching .comfy.safetensors file exists. |
| strength_model | FLOAT | 1.00-100–100 | Strength used if the node applies the latest LoRA back onto the output model. |
| create_captions | BOOLEAN | false | Creates missing .txt caption files automatically from the caption_text field. Leave this off if your videos already have captions. |
| caption_text | STRING | Fallback caption text used only when create_captions is on and a video has no caption file. | |
| add_trigger_word | BOOLEAN | false | Prepends trigger_text to every caption. Useful when you want a fixed concept token. |
| trigger_text | STRING | Trigger word or phrase to prepend to captions when add_trigger_word is enabled. | |
| musubi_root | STRING | A:/MUSUBI/musubi-tuner-ltx2 | Root folder of your musubi-tuner-ltx2 install. |
| ltx2_checkpoint | STRING | A:/MUSUBI/models/ltx2/ltx-2.3-22b-dev.safetensors | Path to the base LTX audio-video checkpoint used for caching and training. |
| gemma_root | STRING | A:/MUSUBI/models/gemma3 | Folder containing the Gemma model files used for text encoder caching. |
| gemma_load_in_4bit | BOOLEAN | false | Loads Gemma in 4-bit mode instead of 8-bit. This reduces VRAM further, but can be slower or a little less stable. |
| gemma_recovery_mode | BOOLEAN | false | Experimental. If enabled, the node will try alternate Gemma cache settings after the normal path fails. |
Outputs (7)
| Name | Type | Description |
|---|---|---|
| model | MODEL | — |
| latest_state_path | STRING | — |
| log_path | STRING | — |
| video_filename_prefix | STRING | — |
| output_name | STRING | — |
| completed_steps | INT | — |
| total_target_steps | INT | — |