Text-to-Video Workflow for Beginners

This tutorial shows you how to generate a video with synchronized audio entirely from a text prompt, using the built-in LTX-2.5 Text-to-Video template in ComfyUI.

When to Use

Text-to-Video is the right starting point when you want to explore a concept from scratch, test a style or mood, or generate a scene where you don’t have a specific frame to anchor from. If you have a source image you want to animate, see the Image-to-Video guide instead.

Step-by-Step Guide

This guide assumes ComfyUI is already installed. If not, see the ComfyUI download page and our ComfyUI setup guide, and check the system requirements to confirm your hardware is supported.

1. Load the Template and Download Models

Open ComfyUI.
Click the Templates button and search LTX-2.5.
Select the Text-to-Video template (video_ltx2_5_t2v). The workflow loads as a node graph with everything pre-configured.
Open the Workflow Overview panel. On first use it lists Missing Models.
Click Download all to fetch the model files directly in ComfyUI.

You only download the models once — they’re reused on every run.

The template uses these files (also on the LTX-2.5 HuggingFace repository if you prefer to download them manually):

File	Purpose	Placement
`ltx-2.5-22b-distilled-transformer-bf16.safetensors`	Distilled model	`ComfyUI/models/diffusion_models/`
`gemma4-12b-with-proj-ltx-2.5-bf16.safetensors`	Gemma 4 text encoder	`ComfyUI/models/text_encoders/`
`gemma4_e2b_it_bf16.safetensors`	Gemma 4 prompt enhancer	`ComfyUI/models/text_encoders/`
`ltx-2.5-video-vae-bf16.safetensors`	Video VAE	`ComfyUI/models/vae/`
`ltx-2.5-audio-vae-bf16.safetensors`	Audio VAE	`ComfyUI/models/vae/`
`ltx-2.3-spatial-upscaler-x2-1.1.safetensors`	Spatial upscaler (2×, reused from LTX-2.3)	`ComfyUI/models/latent_upscale_models/`

2. Write Your Prompt

The prompt is the most important input in a Text-to-Video workflow. Without visual guidance, the model relies entirely on your description to build the scene. See the Prompting Guide for detailed tips.

The template turns on Prompt Enhance by default, which expands a short prompt into a more detailed one before it’s encoded. Turn it off to use your prompt exactly as written.

3. Set Resolution, Length, and Frame Rate

The template’s defaults are a fast, low-VRAM starting point:

Parameter	Default	Notes
Width	768	Base resolution. Must be divisible by 32.
Height	512	Base resolution.
Length	97 frames	Frame count must be `1 + a multiple of 8` (e.g. 97 = 1 + 96).
Frame rate	24 fps	24 for cinematic, 25 standard, 30 for smoother motion. 97 frames at 24 fps ≈ 4 seconds.

The pipeline generates at the base resolution, then upscales 2× in the second stage — so the final output is double the width and height you set (768×512 → 1536×1024 by default).

Higher resolution and longer length need more VRAM. Start near the defaults for testing and increase once you know your hardware handles it.

4. Generate

Click Run. The template runs a two-stage pipeline automatically:

Stage 1 — generates video and audio together at the base resolution.
Upscale — the video latent is upscaled 2× by the spatial upscaler.
Stage 2 — refines the upscaled video at full resolution.

Audio is generated jointly with the video in Stage 1 and carried through to the final output.

5. Review and Iterate

The output saves as an MP4 with synchronized audio. To iterate:

Change the prompt and re-run to explore different scenes.
Adjust length if the video is too short or long for your content.
Try different resolutions to match your target format (landscape, portrait, square).

The Stage 1 seed randomizes by default, so each run differs. To reproduce a result, note its seed and switch the seed control from randomize to fixed.

How the Pipeline Works

Understanding the two-stage pipeline helps when troubleshooting or fine-tuning results.

Model loading: The template loads the distilled transformer directly (no separate base model or distilled LoRA), the Gemma 4 text encoder and prompt enhancer, and the video and audio VAEs. A negative prompt ("pc game, console game, video game, cartoon, childish, ugly") is applied automatically.

Stage 1 (base resolution): Empty video and audio latents are created and concatenated into a single audio-video latent, then sampled together with the euler_ancestral sampler and a dual-CFG guider (CFG 1 for both video and audio) on a fixed distilled sigma schedule. Sampling them jointly is what keeps audio and video in sync. The audio and video latents are then separated.

Upscale: The video latent passes through the spatial upscaler, doubling its resolution.

Stage 2 (full resolution): The upscaled video latent is recombined with the Stage 1 audio latent and refined — sharpening detail without regenerating the composition.

Decode: Video decodes through tiled VAE decoding (to keep VRAM down), audio through the audio VAE, and the two are merged into the final file.

Advanced Techniques

The built-in template gets you generating quickly with sensible defaults. For more control, move to a fuller workflow.

Recommended Next Step

Once you’re comfortable with the template, the Two-Stage Generation workflow uses the same two-stage structure but runs the full-precision base model with the distilled LoRA applied, and adds a higher-quality Full sampler path plus direct control over settings the template handles automatically. Everything you’ve learned here still applies.

Distilled vs. Full

The template uses the distilled model — optimized to produce good results in a few steps, which makes it fast and ideal for iteration. The full model (in the Two-Stage Generation workflow) takes more steps but can yield finer detail and more nuanced motion. Match the model and workflow to best meet your needs and available hardware.

Using LoRAs

Add a LoRALoader node to apply style, motion, or character LoRAs on top of the base model. See the LoRA guide.

Python

Text-to-Video is also available through the PyTorch API for programmatic use. See the PyTorch API documentation.