Do I have to use the LoRA trigger word?

Yes. Each LTX 2.3 LoRA has a trigger keyword (e.g., F3ltCut0u7) that activates its style. Include it in the positive prompt along with your subject for consistent results.

What controls improve temporal consistency?

Keep prompts simple, use a fixed seed, avoid rapid camera moves, and limit clip length (e.g., 24–48 frames). Increasing steps can help detail, while moderate guidance/CFG avoids jittery overcorrection.

Where do I put the models and VAE files?

Place the base model (LTX-2.3-22B-DISTILLED-1.1-GGUF) and LTX2_video_vae_bf16.safetensors in the paths expected by the core LTX node. If the node reports a missing file, update its model and VAE fields to your actual paths. The audio VAE is not used by default.

How do I change the playback speed without re-generating frames?

Use VHS_VideoCombine. Re-run only that node and change the FPS or format; it will re-encode the existing frames into a new video without redoing diffusion.

LTX 2.3 - Cozy Felt - ComfyUI Workflow

LTX 2.3 - Cozy Felt is a stylized text-to-video workflow that wraps VRGameDevGirl84’s LTX 2.3 LoRAs around the LTX-2.3-22B-DISTILLED-1.1 model to generate short, adorable "felt" animations from a prompt. The core generation is driven by a custom LTX node (ID 45d6b185-beee-4467-b4c2-debdf835cb35) configured to load the LTX2 video VAE (LTX2_video_vae_bf16.safetensors) for latent-space video sampling. You pick a LoRA from a dropdown, include its trigger word in your prompt (for example, F3ltCut0u7 for the cozy felt look), and the node produces a sequence of stylized frames. The MarkdownNote node embedded in the canvas provides quick instructions and lists the LoRA trigger words so you don’t have to leave the workspace.

After diffusion, frames are assembled into a playable clip with VHS_VideoCombine, where you set the frame rate and output format. This separation (generate frames first, then encode) makes iteration fast and predictable: you can tweak prompts, LoRA strength, seed, frame count, or resolution on the generation node, then re-run only the combine step to change FPS or container. The result is a reliable, beginner-friendly pipeline for creating soft, craft-like animations—great for cozy title cards, character snippets, or looping social posts. Note: the audio VAE (LTX2_audio_vae_bf16.safetensors) is listed in the assets but is not wired into this canvas by default.

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.

3cb47a67b81a.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 this workflow (TypeScript)
import { Comfy } from "@comfyorg/sdk";

const client = new Comfy({ apiKey: "comfyui-..." });
const wf = await client.workflows.fromFile("workflow_api.json");
const job = await client.run(wf);
await job.getOutputs("<output-node-id>")[0].toFile("output.png");

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