This ComfyUI workflow turns a written prompt into a short video using the LTX-0.9.5 text-to-video checkpoint. It starts by loading the model and text encoder with CheckpointLoaderSimple and CLIPLoader, then converts your prompt and optional negative prompt into embeddings via CLIPTextEncode. Those embeddings are adapted for video with LTXVConditioning so the model can synthesize temporally coherent motion from your text description.

Next, EmptyLTXVLatentVideo defines the canvas of your clip—frame count, width/height, FPS, and seed—while LTXVScheduler and the Custom sampler block (SamplerCustom + KSamplerSelect) drive the denoising process across all frames. After sampling, VAEDecode turns latent frames into RGB images. CreateVideo stitches the frame sequence into a playable file, and SaveVideo writes it to disk. The workflow is organized into clear groups (Step1 - Load models, Step2 - Video size, Step3 - Prompt, Custom sampler) so you always know where to adjust models, resolution, timing, and sampling.

What makes this useful is the control you get at each stage: descriptive prompting (via CLIPTextEncode) for content and motion; precise duration and aspect (via EmptyLTXVLatentVideo); and predictable results with seeds and sampler choices (KSamplerSelect). The author notes that LTX-0.9.5 benefits from long, detailed prompts—concise prompts often produce weaker results—so plan to describe subjects, actions, scene context, and camera motion for best quality.