Should I use VACE-14B or VACE-1.3B?

Use 14B for the best detail and temporal consistency (supports 480p and 720p), but it’s slower and more VRAM‑hungry. Use 1.3B when you need faster drafts at 480p or have limited hardware.

Why does my video look shaky or blurry with the LoRA?

The CausVid LoRA trades some fidelity for speed. Try lowering LoRA strength to 0.3–0.5, increasing steps slightly, or bypassing the LoRA for the final render. Also keep cfg conservative (around 1.0 with LoRA) to reduce instability.

How do I control length and frame rate?

Set fps in WanVaceToVideo, and pick a target number of frames there or finalize it with TrimVideoLatent. For example, 48 frames at 24 fps yields a 2‑second clip. CreateVideo and SaveVideo will honor the fps passed from the pipeline.

What hardware do I need?

Requirements vary by resolution and model size. The 14B model at 720p typically needs a high‑VRAM GPU; 480p on 1.3B is more forgiving. If you hit OOM errors, switch to 1.3B, reduce resolution or frames, and avoid high LoRA strengths.

Wan2.1 VACE Text to Video - ComfyUI Workflow

This workflow turns a short text prompt into a coherent video using Wan 2.1 VACE. It encodes your prompt with CLIPLoader + CLIPTextEncode, then drives the Wan VACE UNet (UNETLoader) through a diffusion loop (KSampler) configured for the model’s sampling behavior via ModelSamplingSD3. The WanVaceToVideo node handles time-aware latent generation, and TrimVideoLatent lets you shorten or crop the sequence to an exact frame count. After sampling, VAELoader + VAEDecode convert latents into RGB frames. Finally, CreateVideo compiles frames and SaveVideo writes MP4 output; SaveAnimatedWEBP is provided for lightweight previews.

Two model sizes are supported: VACE-14B (higher quality, slower; 480p and 720p) and VACE-1.3B (faster; 480p only). The graph is organized into clear groups: Load models here, Prompt, Sampling & Decoding, Save Video (Mp4), and Save Video (WebP), plus separate 14B and 1.3B sections for quick switching. A CausVid LoRA is included via LoraLoader to speed up generation; use it at moderate strength (0.3–0.7) and reduced steps (2–4, cfg ~1.0) for drafts. For best final quality, disable or lower the LoRA and use the default 14B settings (around 20 steps, cfg ~6.0).

API

Use this workflow from code

Every Comfy workflow is a JSON graph. The payload below is this workflow, exactly as ComfyUI runs it — fetch it from the URL, keep it in version control, or load it in ComfyUI and run it node by node.

video_wan_vace_14B_t2v.json

Fetching workflow JSON…

Run it from TypeScript or Python with the Comfy SDK. The same code targets Comfy Cloud or a ComfyUI you host yourself — only the base URL changes.

// Install (beta)
npm i @comfyorg/sdk

// Run "Wan2.1 VACE Text to Video" (TypeScript)
import { Comfy } from "@comfyorg/sdk";

const client = new Comfy({ apiKey: "comfyui-..." });

// This workflow, exported in API format (see note below)
const wf = await client.workflows.fromFile("video_wan_vace_14B_t2v_api.json");

wf.setInput("6", "text", "your prompt here");  // CLIPTextEncode

const job = await client.run(wf);
await job.getOutputs("70")[0].toFile("output.png");  // SaveAnimatedWEBP

The SDK takes a workflow in API format: open this workflow in ComfyUI and use File → Export Workflow (API).

Comfy Cloud API access requires a plan with an API key.

SDK docs Get an API key

FAQ

Frequently Asked Questions