---
title: "Generate Images and Video with ComfyUI"
publisher: "nvidia"
type: "playbook"
updated: "2026-10-15T23:52:30.920Z"
description: "FLUX.2 image and LTX-2 video workflows you run locally with full creative control"
canonical: "https://build.nvidia.com/playbooks/visual-gen-ai.md"
---

# Basic idea

Visual generative AI lets you create photorealistic images and coherent video clips from text (and image) prompts — with creative control that cloud generators often limit. Running these workflows **locally on your hardware platform** keeps assets under your control, avoids per-token cloud costs, and shortens the iteration loop that real creative projects need.

**ComfyUI** is an open-source, node-based app for advanced image and video generation. You install it, load templates for models such as **FLUX.2** and **LTX-2**, download model weights on demand, and build or save workflows as reusable graphs.

- **Local creative control** — refine prompts and graphs at your own pace without cloud round-trips
- **Template-first start** — Getting Started and All Templates paths cover text-to-image and image-to-video
- **VRAM-aware model choices** — pick precision and settings that fit your GPU memory

# What you'll accomplish

You'll install ComfyUI on your hardware platform, run a starter text-to-image workflow, generate images with **FLUX.2-Dev**, generate video with **LTX-2**, and optionally combine both into one custom workflow.

# What to know before starting

**Required:**

- Comfort installing and launching desktop applications on your hardware platform
- Enough free disk space for large model weight downloads (tens of GB per model)

**Optional:**

- Familiarity with generative AI prompting (subject, framing, style, mood)
- Experience with node-based tools or graph editors

# Supported hardware platforms

Use the matrix below to confirm your hardware platform, recommended default local settings, and whether multi-node applies.

| Hardware platform | OS | Memory | Recommended default local settings | Multi-node capable hardware |
| :---- | :---- | :---- | :---- | :---- |
| **GeForce RTX** | Windows (ComfyUI desktop / portable); Linux paths noted for outputs | Dedicated VRAM (size varies by GPU) | ComfyUI from [comfy.org](https://comfy.org) · starter text-to-image template · FLUX.2-Dev · LTX-2 Image to Video · prefer lower-precision weights that fit your VRAM | — |
| **RTX PRO** | Windows (ComfyUI desktop / portable); Linux paths noted for outputs | Dedicated VRAM (size varies by GPU) | ComfyUI from [comfy.org](https://comfy.org) · starter text-to-image template · FLUX.2-Dev · LTX-2 Image to Video · prefer lower-precision weights that fit your VRAM | — |

# Prerequisites

**Hardware requirements**

- Supported hardware platform — see Supported hardware platforms matrix above
- Enough GPU memory for the model and resolution you choose (frontier video models need more VRAM as resolution, length, and steps increase)
- Enough storage for ComfyUI plus model weights (individual models can exceed 30 GB)

**Software requirements**

- Network access to download ComfyUI and model weights (for example from Hugging Face via ComfyUI’s download dialog)
- A web browser (ComfyUI UI and optional image preview)
- For the Windows install path in **Instructions**: use the ComfyUI download from [comfy.org](https://comfy.org)

# Time & risk

- **Estimated time:** 13 MIN for install and first starter image; longer on first run of FLUX.2 or LTX-2 while model weights download
- **Risk level:** Low to Medium
- Large model downloads can fail on unstable networks
- High resolution, frame count, or step counts can exhaust GPU memory or slow generation sharply
- **Rollback:** Quit ComfyUI and uninstall or delete the ComfyUI install folder; delete downloaded model weights and saved workflows you no longer need
- **Last Updated:** 08/03/2026
- Local ComfyUI workflow for FLUX.2 image and LTX-2 video generation on supported hardware platforms

## More

- [Instructions](/playbooks/visual-gen-ai/instructions.md)
- [Troubleshooting](/playbooks/visual-gen-ai/troubleshooting.md)