RS Nodes for ComfyUI

A comprehensive custom node pack for ComfyUI focused on LTXV audio-video generation, LoRA training, image generation, prompt engineering, and post-processing. Built for real-world production workflows with emphasis on VRAM efficiency and quality control.

Installation

cd ComfyUI/custom_nodes
git clone https://github.com/richservo/rs-nodes.git
cd rs-nodes
install.bat

The install script handles everything: initializes the LTX-2 submodule (for LoRA training) and installs all Python dependencies.

Linux:

./install.sh

External Dependencies

Ollama — Required for RS Prompt Formatter and dataset captioning (RS LTXV Prepare Dataset). Install and run locally. The nodes auto-pull models on first use.
LTXV Models — The LTXV generation/extension nodes require LTXV model checkpoints and VAEs loaded through ComfyUI's standard model loading nodes.

Node Reference

All nodes appear under the RS Nodes category in ComfyUI's node menu.

Video Generation

RS LTXV Generate

All-in-one LTXV video generation with optional audio, keyframe injection, multimodal guidance, and integrated 2x latent upscaling.

Required Inputs:

Input	Type	Description
`model`	MODEL	LTXV diffusion model
`positive`	CONDITIONING	Positive text conditioning
`negative`	CONDITIONING	Negative text conditioning
`vae`	VAE	Video VAE for encoding/decoding

Optional Inputs — Generation:

Input	Type	Default	Description
`width`	INT	768	Generation width (step: 32)
`height`	INT	512	Generation height (step: 32)
`num_frames`	INT	97	Frames to generate (step: 8). Auto-overridden by audio duration when `audio` + `audio_vae` are connected
`steps`	INT	20	Denoising steps
`cfg`	FLOAT	3.0	Video classifier-free guidance scale
`noise_seed`	INT	0	Random seed
`seed_mode`	ENUM	random	`random`, `fixed`, `increment`, `decrement`
`frame_rate`	FLOAT	25.0	Video frame rate

Optional Inputs — Frame Injection:

Input	Type	Default	Description
`first_image`	IMAGE	—	Keyframe injected at frame 0
`middle_image`	IMAGE	—	Keyframe injected at midpoint
`last_image`	IMAGE	—	Keyframe injected at last frame
`first_strength`	FLOAT	1.0	Preservation strength (1.0 = exact match)
`middle_strength`	FLOAT	1.0	Preservation strength for middle frame
`last_strength`	FLOAT	1.0	Preservation strength for last frame
`crf`	INT	35	LTXV preprocessing quality (0–100)

Optional Inputs — Audio (AV Dual-Tower):

Input	Type	Default	Description
`audio`	AUDIO	—	Input audio (passthrough to output if provided)
`audio_vae`	VAE	—	Audio VAE (enables AV dual-tower mode)
`audio_cfg`	FLOAT	7.0	Audio CFG scale
`stg_scale`	FLOAT	0.0	Spatiotemporal guidance scale (0 = disabled)
`stg_blocks`	STRING	"29"	Comma-separated transformer block indices for STG
`rescale`	FLOAT	0.7	CFG rescaling factor

Optional Inputs — Efficiency:

Input	Type	Default	Description
`attention_mode`	ENUM	auto	`auto` (SAGE if available), `default`, `sage`
`ffn_chunks`	INT	4	FFN sequence chunking (0 = disabled)
`video_attn_scale`	FLOAT	1.03	Video attention scaling + VRAM-efficient forward

Optional Inputs — Upscale (2x Latent):

Input	Type	Default	Description
`upscale`	BOOLEAN	False	Enable 2x spatial upscaling
`upscale_model`	LATENT_UPSCALE_MODEL	—	Latent upscale model
`upscale_steps`	INT	4	Re-diffusion steps at upscaled resolution
`upscale_cfg`	FLOAT	1.0	CFG during re-diffusion
`upscale_denoise`	FLOAT	0.5	Denoise strength (0 = no re-diffuse, 1 = full)

Outputs:

Output	Type	Description
`latent`	LATENT	Video latent `[B, C, T, H, W]`
`audio_latent`	LATENT	Audio latent (if audio_vae used)
`images`	IMAGE	Decoded video frames `[B*T, H, W, C]`
`audio_output`	AUDIO	Output audio (passthrough or decoded)

RS LTXV Extend

Extend an existing video with seamless temporal continuation using overlap blending.

Required Inputs:

Input	Type	Description
`model`	MODEL	LTXV diffusion model
`positive`	CONDITIONING	Positive conditioning
`negative`	CONDITIONING	Negative conditioning
`vae`	VAE	Video VAE
`latent`	LATENT	Input video latent to extend

Optional Inputs:

Input	Type	Default	Description
`num_new_frames`	INT	80	New frames to generate (step: 8)
`overlap_frames`	INT	16	Frames of overlap for blending (step: 8)
`steps`	INT	20	Denoising steps
`cfg`	FLOAT	3.0	Video CFG scale
`last_image`	IMAGE	—	Keyframe for end of extension

Also accepts the same Audio, Efficiency, Upscale, and Scheduler inputs as RS LTXV Generate.

Outputs:

Output	Type	Description
`latent`	LATENT	Extended video latent (original + new frames)
`images`	IMAGE	Decoded extension frames
`audio_output`	AUDIO	Decoded audio (if audio_vae used)

RS LTXV Upscale

Standalone 2x video upscaler with optional temporal upscaling and first-frame I2V re-diffusion.

Required Inputs:

Input	Type	Description
`model`	MODEL	LTXV diffusion model
`positive`	CONDITIONING	Positive conditioning
`negative`	CONDITIONING	Negative conditioning
`vae`	VAE	Video VAE
`images`	IMAGE	Input video frames

Optional Inputs:

Input	Type	Default	Description
`upscale_model`	LATENT_UPSCALE_MODEL	—	Spatial 2x upscale model
`temporal_upscale_model`	LATENT_UPSCALE_MODEL	—	Temporal 2x upscale model
`upscale_steps`	INT	4	Re-diffusion steps
`upscale_cfg`	FLOAT	1.0	CFG during re-diffusion
`upscale_denoise`	FLOAT	0.5	Denoise strength
`upscale_lora`	ENUM	none	LoRA for re-diffusion pass
`attention_mode`	ENUM	auto	Attention optimization
`ffn_chunks`	INT	4	FFN chunking for memory

Outputs:

Output	Type	Description
`latent`	LATENT	Upscaled video latent
`images`	IMAGE	Decoded upscaled frames
`audio_output`	AUDIO	Audio passthrough

Key Behaviors:

Encodes input, applies spatial 2x via upscale model, re-diffuses with first-frame I2V guidance
Optional temporal 2x increases frame count before spatial upscaling
Tiled encoding/decoding for memory efficiency

Image Generation

RS Flux2 Generate

All-in-one Flux2.dev image generation with Kontext reference images, empirical mu scheduling, and optional two-pass latent upscaling.

Required Inputs:

Input	Type	Description
`model`	MODEL	Flux2 model
`clip`	CLIP	Text encoder
`vae`	VAE	Image VAE
`prompt`	STRING	Generation prompt

Optional Inputs:

Input	Type	Default	Description
`width`	INT	1024	Image width
`height`	INT	1024	Image height
`steps`	INT	50	Sampling steps
`guidance`	FLOAT	4.0	Guidance scale
`seed`	INT	0	Random seed
`ref_image_1..4`	IMAGE	—	Up to 4 Kontext reference images
`lora`	ENUM	none	LoRA adapter
`attention_mode`	ENUM	auto	Attention optimization
`ffn_chunks`	INT	4	FFN chunking (double blocks only)
`latent_upscale`	ENUM	off	Two-pass latent upscaling

Outputs:

Output	Type	Description
`images`	IMAGE	Generated image

Key Behaviors:

Kontext reference images auto-scaled to optimal aspect ratios and appended to conditioning
Empirical mu scheduling based on image dimensions
Optional two-pass: half-res first pass, then full-res re-diffusion

RS Z-Image Generate

All-in-one Z-Image Turbo generation with Qwen3-4B text encoder, RenormCFG, and system prompt support.

Required Inputs:

Input	Type	Description
`model`	MODEL	Z-Image Turbo model
`clip`	CLIP	Qwen3-4B text encoder
`vae`	VAE	Image VAE
`prompt`	STRING	Generation prompt

Optional Inputs:

Input	Type	Default	Description
`width`	INT	1024	Image width
`height`	INT	1024	Image height
`steps`	INT	10	Sampling steps (4 for turbo)
`system_prompt`	ENUM	superior	System prompt variant
`cfg`	FLOAT	1.5	CFG scale
`renorm_cfg`	FLOAT	1.0	RenormCFG norm limit (0=off)
`lora`	ENUM	none	LoRA adapter

Outputs:

Output	Type	Description
`images`	IMAGE	Generated image

LoRA Training

RS LTXV Prepare Dataset

Scans a folder of videos/images, optionally detects and crops faces, identifies speakers by voice, generates captions via Ollama with woven-in dialogue, and preprocesses latents for LoRA training.

Required Inputs:

Input	Type	Description
`media_folder`	STRING	Path to folder of videos and/or images
`model_path`	CHECKPOINT	LTX-2 checkpoint
`text_encoder_path`	STRING	Gemma-3 HF directory (auto-download available)
`output_name`	STRING	Name for preprocessed output folder

Optional Inputs — Captioning:

Input	Type	Default	Description
`resolution_buckets`	STRING	"576x576x49"	WxHxF resolution buckets (semicolon-separated)
`lora_trigger`	STRING	""	Trigger word prepended to all captions
`caption_mode`	ENUM	ollama	`ollama`, `skip`, `auto_filename`
`caption_style`	ENUM	subject	`subject`, `subject + style`, `style`, `motion`, `general`, `multi_character`
`ollama_model`	STRING	"gemma4:26b"	Vision model for captioning
`skip_id_pass`	BOOLEAN	False	Skip cast/location ID pass — send all refs directly to captioner

Optional Inputs — Face Detection & Character ID:

Input	Type	Default	Description
`face_detection`	BOOLEAN	True	Enable InsightFace (antelopev2) face detection + ArcFace embedding
`target_face`	IMAGE	—	Reference face for single-character identity matching (uses `lora_trigger` as the character name)
`character_refs_folder`	STRING	""	Multi-character mode: folder of reference face images. Filename stem = character trigger
`location_refs_folder`	STRING	""	Location reference images folder
`face_similarity`	FLOAT	0.40	Face match threshold
`face_padding`	FLOAT	0.6	Padding around detected face for face-crop mode
`crop_mode`	ENUM	face_crop	`face_crop`, `pan_and_scan`, `full_frame`

Optional Inputs — Speech Transcription & Voice Attribution:

Input	Type	Default	Description
`transcribe_speech`	BOOLEAN	False	Enable Whisper transcription with Demucs vocal isolation
`whisper_model`	ENUM	large-v3	Whisper model size (`tiny`/`base`/`small`/`medium`/`large-v2`/`large-v3`/`large-v3-turbo`)
`voice_refs_folder`	STRING	""	Folder of voice reference clips (audio OR video). Filename stem must match `character_refs` entries. Enables per-line speaker attribution via speechbrain ECAPA-TDNN

Optional Inputs — Other:

Input	Type	Default	Description
`target_fps`	FLOAT	0.0	Target framerate (0 = source fps). Drops/keeps frames to match without affecting audio speed
`max_samples`	INT	0	Max total character appearances. Quotas split evenly across characters (e.g. max=200, 4 chars → 50 each)
`skip_start_seconds`	FLOAT	0.0	Skip first N seconds of every video
`skip_end_seconds`	FLOAT	0.0	Skip last N seconds of every video
`conditioning_folder`	STRING	""	IC-LoRA conditioning inputs folder
`clip`	CLIP	—	Text encoder for in-process condition encoding
`vae`	VAE	—	VAE for in-process latent encoding
`clip_vision`	CLIP_VISION	—	Optional CLIP Vision model for matching non-human characters (puppets, props)

Outputs:

Output	Type	Description
`preprocessed_path`	STRING	Path to preprocessed data root
`dataset_json_path`	STRING	Path to dataset JSON

Key Behaviors:

Extraction & Selection:

Two-phase processing: clip generation → latent preprocessing
Face detection via InsightFace (SCRFD + ArcFace, antelopev2 model pack)
Multi-character mode: drop face refs in character_refs_folder — auto-identifies cast per clip, balances quotas across characters
Location mode: separate folder for set/location references — Gemma matches location per clip
Pan & scan: face-aware cropping at any aspect ratio while keeping the face in frame
Frame-count safety: clip extraction always overshoots target frame count and trims, so clips never come up short of the bucket minimum
Incremental & resumable: per-clip atomic saves — interrupted runs continue from where they stopped; transcripts and conditions persist as they're generated
Quarantine, never delete: clips that fail (hallucination, no usable speech, no matching characters, etc.) move to <output_dir>/rejected_clips/<reason>/ instead of being deleted, so false positives can be reviewed and restored

Speech Transcription:

Demucs vocal isolation runs first to strip music/SFX before Whisper transcribes
Default model is large-v3 for best accuracy on character voices, accents, and unusual vocabulary; configurable down to tiny for quick tests
Silent clips get an empty transcript "" (recorded once, never re-attempted on subsequent runs)

Voice Attribution (when voice_refs_folder is set):

Each character's voice reference is embedded with speechbrain ECAPA-TDNN (192-d L2-normalized vector, downloaded automatically — no HuggingFace token needed)
For each Whisper segment in a clip, ECAPA produces a per-segment embedding that's matched against enrolled voice prints by cosine distance
Face-detection hint: the on-screen character (from face detection) gets a small distance bonus, biasing ambiguous matches toward the visible speaker without overriding strong off-screen voice evidence
Each segment is tagged with the matched character (or unknown for short utterances / unenrolled voices)
Tagged segments are passed to the captioner as a DIALOGUE CONTEXT block, and Gemma weaves the lines naturally into the caption ("X stands by the window and says, '...'") instead of appending dialogue as a separate field

Caption Encoding (in-process when clip is connected):

Captions stay clean in dataset.json (visual prose only — dialogue is woven in by the captioner)
At text-encode time, captions are normalized: quote-wrapping single quotes are stripped (so 'pee-wee' becomes pee-wee without affecting Pee-wee's or it's); known character names are case-normalized using the canonical trigger (so a stray Cowboy curd is from a Whisper mistranscription still encodes as Cowboy Curtis)
Original captions in dataset.json are never mutated — the normalization only affects the encoded .pt tensors

VRAM Management:

All prepper-loaded models (Whisper, Demucs, ECAPA-TDNN, InsightFace) are unloaded between phases, with a final unload sweep before prepare() returns so unattended workflows that follow with training start with a clean GPU

Voice Reference Files:

Format: any audio (.wav, .mp3, .flac, .m4a, .ogg, .opus) or video (.mp4, .mkv, .mov, .avi, .webm, .m4v) — audio is auto-extracted from video via ffmpeg
Naming: filename stem must match the character_refs_folder entry (e.g. cowboy curtis.wav matches cowboy curtis.jpg)
Length: 10-30 seconds of clean speech is the sweet spot; under 5 seconds risks unstable embeddings
Variety helps: different sentences/intonations beat one repeated phrase. Concatenating short clips of the same character with ffmpeg works fine — each character can be built from 4-6 short single-speaker clips spliced together
Demucs runs on enrollment too, so background music/effects in your reference clip get stripped automatically

RS LTXV Train LoRA

In-process LoRA training for LTX-2, reusing ComfyUI's already-loaded transformer. Includes a standalone training monitor that opens in a separate browser tab for full-size, resizable loss charts.

Required Inputs:

Input	Type	Description
`model`	MODEL	LTX-2 model (from CheckpointLoaderSimple)
`output_name`	STRING	Name for the output LoRA file
`preprocessed_data_root`	STRING	Path from RS LTXV Prepare Dataset
`model_path`	CHECKPOINT	LTX-2 checkpoint (for EmbeddingsProcessor/VAE)

Optional Inputs:

Input	Type	Default	Description
`vae`	VAE	—	VAE for validation (avoids loading from checkpoint)
`preset`	ENUM	subject	`custom`, `subject`, `style`, `motion`, `subject + style`, `all video`, `audio + video`
`lora_rank`	INT	16	LoRA rank
`lora_alpha`	INT	16	LoRA alpha
8 module toggles	BOOLEAN	varies	Select which layers to train (self-attn, cross-attn, FFN, audio)
`learning_rate`	FLOAT	1e-4	Learning rate
`epochs`	INT	3	Number of passes through the full dataset
`auto_stop`	BOOLEAN	False	Ignore epoch count — train until divergence detection stops
`optimizer`	ENUM	adamw8bit	`adamw8bit`, `adamw`, or `rose`
`rose_stabilize`	BOOLEAN	True	ROSE only: CV Trust Gating (try False for some conditions)
`scheduler`	ENUM	linear	LR schedule: `linear`, `constant`, `cosine`, `cosine_with_restarts`, `polynomial`
`lr_cycle_steps`	INT	0	LR schedule cycle length (0 = one cycle per epoch)
`lr_cycle_decay`	FLOAT	1.0	Multiply LR by this factor each cycle reset (1.0 = no decay)
`quantization`	ENUM	fp8-quanto	`fp8-quanto`, `int8-quanto`, `int4-quanto`, `none`
`strategy`	ENUM	text_to_video	`text_to_video` or `video_to_video` (IC-LoRA)
`clip`	CLIP	—	Text encoder for validation prompt
`validation_prompt`	STRING	""	Prompt for validation video generation
`validation_interval`	INT	250	Steps between validations
`checkpoint_interval`	INT	500	Steps between checkpoints
`diverge_detect_steps`	INT	150	Steps above threshold before entering monitoring
`diverge_stop_steps`	INT	300	Steps in monitoring without recovery before stopping
`diverge_threshold`	FLOAT	15.0	% above lowest EMA loss to trigger divergence
`resume`	BOOLEAN	False	Resume from latest checkpoint

Outputs:

Output	Type	Description
`status`	STRING	Training status message
`lora_path`	STRING	Path to saved LoRA file

Key Behaviors:

In-process: reuses the loaded 22B transformer — no reload, no double memory
Training monitor: click "Open Training Monitor" on the node to open a full-page loss chart in a new browser tab. Shows raw loss dots, EMA-smoothed line, color-coded trend line, step timing, and divergence status. Loads history from loss_history.json on refresh.
ROSE optimizer: stateless optimizer — no momentum buffers, lower memory. ROSE-specific settings auto-hide when another optimizer is selected.
Divergence detection: monitors EMA loss vs minimum. If loss rises above threshold% for too long, saves a checkpoint and attempts LR reset. If unrecoverable, stops training and rewinds to the pre-divergence checkpoint.
LR cycle decay: progressive learning rate reduction across scheduler cycles (e.g., 0.9 = 10% reduction per cycle)
Loss history persistence: saves loss_history.json every 50 steps for chart continuity across resume/restart
Layer offloading: streams transformer blocks CPU↔GPU one at a time (~0.5 GB VRAM instead of ~11 GB)
FP8 quantization: reduces model memory while preserving LoRA weights in float
Resume support: restores optimizer state, RNG, loss history, and rebuilds LR schedule
Presets: auto-configure module toggles and rank for common use cases (subject, style, motion)
Quantization modifies the transformer in-place — reload checkpoint after training

Guidance & Control

RS IC-LoRA Guider

Structural control for LTXV using IC-LoRA (In-Context LoRA) with preprocessed control images.

Required Inputs:

Input	Type	Description
`model`	MODEL	LTXV diffusion model
`positive`	CONDITIONING	Positive conditioning
`negative`	CONDITIONING	Negative conditioning
`vae`	VAE	Video VAE
`control_image`	IMAGE	Preprocessed control map (e.g., canny edges)
`ic_lora`	ENUM	IC-LoRA safetensors file

Outputs:

Output	Type	Description
`guider`	GUIDER	Connect to the `guider` input of RS LTXV Generate

RS LTXV TTM Guider

Time-to-Move (TTM) motion control guider for LTXV. Enforces reference motion via dual-clock denoising with mask-based control.

Required Inputs:

Input	Type	Description
`model`	MODEL	LTXV diffusion model
`positive`	CONDITIONING	Positive conditioning
`negative`	CONDITIONING	Negative conditioning
`vae`	VAE	Video VAE
`reference_video`	IMAGE	Reference video for motion
`mask`	IMAGE	Motion mask (1.0 = enforce reference, 0.0 = free generation)

Optional Inputs:

Input	Type	Default	Description
`ttm_strength`	FLOAT	0.5	Fraction of steps to apply TTM
`cfg`	FLOAT	3.0	CFG scale
`stg_scale`	FLOAT	0.0	Spatiotemporal guidance

Outputs:

Output	Type	Description
`guider`	GUIDER	Connect to RS LTXV Generate's guider input
`width`	INT	Generation width
`height`	INT	Generation height

RS Canny Preprocessor

Canny edge detection with automatic LTXV-safe resolution (128-aligned).

Inputs:

Input	Type	Default	Description
`image`	IMAGE	—	Input image
`width`	INT	768	Target width (step: 128)
`height`	INT	512	Target height (step: 128)
`low_threshold`	INT	100	Canny low threshold
`high_threshold`	INT	200	Canny high threshold

Outputs:

Output	Type	Description
`image`	IMAGE	Edge map (grayscale to RGB)
`width`	INT	128-aligned width
`height`	INT	128-aligned height

Prompt Engineering

RS Prompt Parser

Parse structured dialogue scripts with [s]tyle, [a]ction, and [d]ialogue tags into separate video and audio prompts.

Inputs:

Input	Type	Default	Description
`script`	STRING	""	Multiline script with tags
`dialogue_mode`	ENUM	individual	`individual` or `all`
`dialogue_index`	INT	1	Which dialogue line to select

Script Format:

[s] cinematic, natural lighting, handheld camera
[a] a man walks into a room and sits down
[d] Hello, how are you doing today?

Outputs:

Output	Type	Description
`video_prompt`	STRING	Combined visual prompt
`audio_prompt`	STRING	TTS text
`dialogue_count`	INT	Total dialogue segments
`current_index`	INT	Current dialogue index
`dialogue_list`	STRING	Numbered list of all dialogue

RS Prompt Formatter

AI-powered prompt enhancement using a local Ollama model with reference image support and output caching.

Inputs:

Input	Type	Default	Description
`prompt`	STRING	""	Raw prompt to enhance
`system_prompt`	STRING	(built-in)	Instructions for the model
`model`	STRING	"gemma3:12b"	Ollama model name
`reference_image`	IMAGE	—	(optional) Image for visual context

Outputs:

Output	Type	Description
`formatted_prompt`	STRING	Enhanced prompt text

Key Behaviors:

Streams responses from Ollama with live token printing
Auto-pulls missing models, strips <think> blocks
Caches prompt + output as JSON — skips Ollama when input unchanged

Requires: Ollama running locally.

RS Prompt Formatter Local

Ollama-free prompt formatter reusing Gemma3 12B from ComfyUI's DualCLIPLoader. Supports optional reference images via vision embeddings.

Inputs:

Input	Type	Default	Description
`text_encoder`	STRING	—	Gemma3 text encoder file
`prompt`	STRING	""	Prompt to format
`system_prompt`	STRING	—	System prompt for generation
`first_image`	IMAGE	—	(optional) Opening image
`middle_image`	IMAGE	—	(optional) Mid-scene image
`last_image`	IMAGE	—	(optional) Ending image
`max_tokens`	INT	1024	Maximum output tokens
`temperature`	FLOAT	0.8	Sampling temperature

Outputs:

Output	Type	Description
`formatted_prompt`	STRING	Enhanced prompt text

Key Behaviors:

No Ollama dependency — loads Gemma3 12B weights directly
Vision support for up to 3 reference images
JSON caching (text-only prompts)

Audio

RS Audio Concat

Concatenate up to 20 audio clips with per-clip trimming and configurable pauses.

Inputs: For each clip: audio file, start/end trim, pause after.

Outputs:

Output	Type	Description
`audio`	AUDIO	Concatenated waveform

RS Audio Save

Export audio to disk with format selection (wav, flac, mp3, ogg).

RS MOSS TTS Loader

(Optional — requires transformers)

Load MOSS-TTS model variants for text-to-speech generation.

RS MOSS TTS Batch Save

(Optional — requires transformers)

Generate TTS audio from a dialogue list with automatic segmentation, per-clip trimming, and batch export. Supports one_shot, all, and single generation modes with Whisper-based word alignment.

Post-Processing & Utilities

RS Film Grain

Add realistic film grain with color variation and luminance-aware highlight protection.

Inputs:

Input	Type	Default	Description
`images`	IMAGE	—	Input video frames
`intensity`	FLOAT	0.05	Grain strength (0–1)
`grain_size`	FLOAT	1.5	Grain frequency
`color_amount`	FLOAT	0.3	Color noise ratio
`highlight_protection`	FLOAT	0.5	Protect bright/dark areas

Outputs:

Output	Type	Description
`images`	IMAGE	Grained video frames

RS Video Trim

Trim video frames and/or audio by time range.

Inputs:

Input	Type	Default	Description
`fps`	FLOAT	24.0	Frame rate
`in_point`	FLOAT	0.0	Start time (seconds)
`out_point`	FLOAT	0.0	End time (0 = end of clip)
`images`	IMAGE	—	(optional) Video frames
`audio`	AUDIO	—	(optional) Audio

Outputs:

Output	Type	Description
`images`	IMAGE	Trimmed frames
`audio`	AUDIO	Trimmed audio

RS Free VRAM

Passthrough utility that forces VRAM cleanup between pipeline stages.

Input	Type	Description
`any_input`	* (wildcard)	Any data — passed through unchanged

Unloads all models, runs garbage collection, clears CUDA cache.

RS Counter

Persistent incrementing counter. State stored in counter_state.json across workflow executions.

Inputs:

Input	Type	Default	Description
`start`	INT	0	Starting value (used on reset)
`step`	INT	1	Increment amount
`reset`	BOOLEAN	False	Reset to start value

Outputs:

Output	Type	Description
`value`	INT	Current counter value (increments after output)

Workflow Examples

Basic LTXV Video Generation

[Load LTXV Model] ──┐
[CLIP Text Encode] ──┼──→ [RS LTXV Generate] ──→ [Save Video]
[CLIP Text Encode] ──┤
[Load VAE] ──────────┘

LoRA Training Pipeline

[Load LTXV Model] ──┐
[Load VAE] ──────────┼──→ [RS LTXV Prepare Dataset] ──→ [RS LTXV Train LoRA]
[Load CLIP] ─────────┘                                         │
                                                    (click "Open Training Monitor")

Audio-Driven Video with TTS

[RS Prompt Parser] ──→ [RS Prompt Formatter] ──→ [CLIP Text Encode] ──┐
        │                                                              │
        └──→ [RS MOSS TTS Loader] ──→ [RS MOSS TTS Batch Save] ──┐    │
                                              │                    │    │
                                      [RS Audio Concat] ──────────┼────┼──→ [RS LTXV Generate]

IC-LoRA Structural Control

[Load Image] ──→ [RS Canny Preprocessor] ──→ [RS IC-LoRA Guider] ──→ [RS LTXV Generate]

Video Extension + Upscale

[RS LTXV Generate] ──→ [RS LTXV Extend] ──→ [RS LTXV Upscale] ──→ [Save Video]

Tips & Troubleshooting

VRAM Management

Place RS Free VRAM between heavy inference nodes and post-processing to reclaim GPU memory.
Use ffn_chunks (default: 4) and video_attn_scale (default: 1.03) to reduce VRAM during generation.
Enable upscale_tiling for temporal tiling during upscale on long videos.
LoRA training uses layer offloading to stream blocks one at a time (~0.5 GB instead of ~11 GB).

Generation Quality

cfg: 2.5–4.0 works well for video; higher values can cause artifacts.
audio_cfg: 5.0–9.0 is typical for audio.
stg_scale: Start at 0. Small values (0.1–0.5) can improve temporal consistency.
rescale: 0.7 is a good default.
Upscaling: upscale_denoise=0.3–0.6 balances sharpness vs. faithfulness.

LoRA Training

Use fp8-quanto quantization (no C++ build tools needed).
Subject LoRAs: enable self-attention + cross-attention. Captions should describe everything except the subject.
Style LoRAs: enable self-attention + feed-forward. Captions should describe the visual style in detail.
ROSE optimizer: stateless, lower memory than AdamW. LR may need to be ~2x higher than AdamW (model-dependent — experiment).
Divergence detection: uses EMA loss distance from minimum. Threshold of 15% is a good default. Auto-saves checkpoint at detection, attempts LR reset, stops and rewinds if unrecoverable.
Resume: restores optimizer state, loss history, and LR schedule. The training monitor loads history from loss_history.json automatically.
Training Monitor: open via the link on the training node. Full-page chart with raw loss dots, EMA-smoothed line, trend line (green=decreasing, red=increasing), step timing, and divergence status. Resizes with the browser window.

Prompt Workflow

Write scripts with [s], [a], [d] tags and feed into RS Prompt Parser.
RS Prompt Formatter caches to JSON — if the input prompt hasn't changed, Ollama is skipped.
RS Prompt Formatter Local avoids Ollama entirely by loading Gemma3 12B directly.

License

See LICENSE for details.

Name		Name	Last commit message	Last commit date
LTX-2 @ c27c3b9		LTX-2 @ c27c3b9
Reference		Reference
Workflow		Workflow
luts		luts
nodes		nodes
runpod		runpod
tools		tools
utils		utils
web		web
.gitattributes		.gitattributes
.gitignore		.gitignore
.gitmodules		.gitmodules
IC_LORA_GUIDER_REFERENCE.md		IC_LORA_GUIDER_REFERENCE.md
IC_LORA_USAGE_GUIDE.md		IC_LORA_USAGE_GUIDE.md
README.md		README.md
__init__.py		__init__.py
install.bat		install.bat
install.sh		install.sh
requirements.txt		requirements.txt

Latest commit

History

Folders and files

Repository files navigation

RS Nodes for ComfyUI

Table of Contents

Installation

External Dependencies

Node Reference

Video Generation

RS LTXV Generate

RS LTXV Extend

RS LTXV Upscale

Image Generation

RS Flux2 Generate

RS Z-Image Generate

LoRA Training

RS LTXV Prepare Dataset

RS LTXV Train LoRA

Guidance & Control

RS IC-LoRA Guider

RS LTXV TTM Guider

RS Canny Preprocessor

Prompt Engineering

RS Prompt Parser

RS Prompt Formatter

RS Prompt Formatter Local

Audio

RS Audio Concat

RS Audio Save

RS MOSS TTS Loader

RS MOSS TTS Batch Save

Post-Processing & Utilities

RS Film Grain

RS Video Trim

RS Free VRAM

RS Counter

Workflow Examples

Basic LTXV Video Generation

LoRA Training Pipeline

Audio-Driven Video with TTS

IC-LoRA Structural Control

Video Extension + Upscale

Tips & Troubleshooting

VRAM Management

Generation Quality

LoRA Training

Prompt Workflow

License

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages