ComfyUI Node
RT LTX-2.5 Prompt Enhancer
A ComfyUI node in RareTutor/LTX2.5 with 18 inputs and 3 outputs.
RT LTX-2.5 Prompt Enhancer
- image
- ENHANCED_PROMPT
- NEGATIVE_PROMPT
- DIAGNOSTICS
◄enabledtrue►
◄llm_modelNo supported models (.gguf) found in text_encoders or llm►
◄user_prompt►
◄enhancement_mode01. Single-Shot (Continuous Take - 4-8 Sentences)►
◄camera_movementAuto (Director Discretion)►
◄lighting_atmosphereAuto (Match Scene Mood)►
◄audio_modeFull Audio (Atmosphere + Foley FX + Dialogue in Quotes)►
◄frame_length121 Frames (5.0s @ 24fps - Standard Take)►
◄creativity0.9 - Balanced Cinematic (Official Guide Recommended)►
◄seed-1►
◄max_tokens1024►
◄negative_prompt_generationEnable (Generate Tuned LTX Negative Prompt)►
◄n_gpu_layers-1►
◄n_ctx8192►
◄keep_model_loadedtrue►
◄vision_modelNone / Bypass (Text-Only Mode)►
◄custom_system_instructions►
CategoryRareTutor/LTX2.5
Inputs (18)
| Name | Type | Default | Description |
|---|---|---|---|
| enabled | BOOLEAN | true | Enable or disable prompt enhancement. When disabled, bypasses LLM inference and passes through the user prompt directly. |
| llm_model | COMBO | No supported models (.gguf) found in text_encoders or llm | Select local GGUF LLM model (e.g. Qwen2.5, Llama-3, Mistral, Gemma). |
| user_prompt | STRING | — | |
| enhancement_mode | COMBO | 01. Single-Shot (Continuous Take - 4-8 Sentences) | Select the LTX-2.5 prompting mode matching the official guide. |
| camera_movement | COMBO | Auto (Director Discretion) | Specify camera movement. Choosing 'Static Tripod' locks off the camera and prevents unwanted AI drift. |
| lighting_atmosphere | COMBO | Auto (Match Scene Mood) | Select coherent lighting logic. LTX-2.5 requires a single coherent light logic per shot. |
| audio_mode | COMBO | Full Audio (Atmosphere + Foley FX + Dialogue in Quotes) | LTX-2.5 generates synchronized video and audio. Spoken dialogue is automatically quoted in double quotes. |
| frame_length | COMBO | 121 Frames (5.0s @ 24fps - Standard Take) | Target duration and frame count for LTX-2.5 video generation. |
| creativity | COMBO | 0.9 - Balanced Cinematic (Official Guide Recommended) | Controls LLM temperature. 0.9 is calibrated for cinematic depth. |
| seed | INT | -1-1–18446744073709550000 | Random seed for reproducible prompt expansions. |
| max_tokens | COMBO | 1024 | Maximum output tokens for prompt generation. |
| negative_prompt_generation | COMBO | Enable (Generate Tuned LTX Negative Prompt) | Generates a customized negative prompt preventing LTX-2.5 artifacts (morphing, jitter, muddy audio, plastic skin). |
| n_gpu_layers | INT | -1-1–128 | Number of layers to offload to GPU. -1 offloads all layers (fast CUDA inference). |
| n_ctx | INT | 81922048–32768 | Context window size for LLM. |
| keep_model_loaded | BOOLEAN | true | Keep the LLM model loaded in memory between runs for near-instant generations. |
| vision_modelopt | COMBO | None / Bypass (Text-Only Mode) | Vision projector (mmproj) model for Image-to-Video (I2V) anchoring. If plugged in with an image, the model directly analyzes the initial frame. |
| imageopt | IMAGE | Optional initial frame image for Image-to-Video (I2V) prompt enhancement. | |
| custom_system_instructionsopt | STRING | — |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| ENHANCED_PROMPT | STRING | — |
| NEGATIVE_PROMPT | STRING | — |
| DIAGNOSTICS | STRING | — |