A simple 2-stage T2V workflow for the LTX2 model.

v1.1 – “Full Sigmas on Stage 1” option inside the Inference subgraph

• True: processes all steps in the first stage (slower)

• False: processes only the appropriate percentage of steps for Stage 1, based on the sigmas denoise value (faster)

Prompt Enhancer

• To use the Prompt Enhancer group, you need to install llama.cpp.

• It is recommended to compile it from source instead of spending time searching for a compatible prebuilt wheel for your OS and environment. The build usually takes ~10 minutes.

• You can use any GGUF model supported by llama.cpp.

• gemma-3n-E4B-it-Q4_0 provides:

  • well-structured results

  • good understanding of complex instructions

  • consistent outputs across runs

  • lightweight and fast performance

Model Setup (Prompt Enhancer)

• Download the model using the link provided in the workflow.

• Place it in:
  ./ComfyUI/models/llm_gguf