--- title: "Generate Images and Video with ComfyUI — Instructions" canonical: "https://build.nvidia.com/playbooks/visual-gen-ai/instructions.md" --- # Step 1. Install and launch ComfyUI Visit [comfy.org](https://comfy.org) to download and install ComfyUI for Windows, then launch ComfyUI on your hardware platform. > [!NOTE] > This playbook follows the desktop / portable ComfyUI path from [comfy.org](https://comfy.org). It does **not** invent host Python, Docker, or shell install commands. For a Linux host/container ComfyUI path on other hardware platforms, see [Generate Images and Videos with ComfyUI](https://build.nvidia.com/playbooks/comfyui). # Step 2. Create a first image with the starter template 1. Click **Templates**, then **Getting Started**, and choose **1.1 Starter - Text to Image**. 2. Connect the model node to the **Save Image** node so the graph forms a complete pipeline. 3. Press the blue **Run** button and watch the green node highlights as your hardware platform generates the first image. Change the prompt and run again to iterate. # Step 3. Generate images with FLUX.2-Dev 1. Open **Templates** → **All Templates** and search for **FLUX.2 Dev Text to Image**. 2. Select the template. ComfyUI loads the connected nodes (the workflow). 3. When prompted, download the model weights. Weight files (`.safetensors`) save into the correct ComfyUI folders automatically. FLUX.2 can exceed 30 GB depending on the version — allow time and disk space. 4. Save the workflow: open the top-left hamburger menu and choose **Save**. Press **W** to show or hide the Workflows list. If you close the download dialog before weights finish: 1. Press **W** to open **Workflows** 2. Select the workflow again so ComfyUI reloads it and re-prompts for any missing weights **Prompt tips for FLUX.2-Dev** - Start with clear subject, setting, style, and mood — for example: “Cinematic closeup of a vintage race car in the rain, neon reflections on wet asphalt, high contrast, 35mm photography.” Short-to-medium prompts are usually easier to control than long storylike prompts when learning. - Add constraints for framing (“wide shot” / “portrait”), detail (“high detail, sharp focus”), and realism (“photorealistic” / “stylized illustration”). - If results feel busy, remove adjectives instead of adding more. - Avoid negative prompting — describe what you want. Learn more in the [Black Forest Labs FLUX.2 prompting guide](https://docs.bfl.ai/guides/prompting_guide_flux2). **Save locations** Right-click the **Save Image** node to open the image in a browser or save it elsewhere. Default ComfyUI output folders are typically: - Windows (standalone / portable): under the ComfyUI install (for example an `output` folder beside the app) - Windows (desktop application): under the AppData directory - Linux: under the ComfyUI install location # Step 4. Generate video with LTX-2 Use the **LTX-2 Image to Video** template. Lightricks’ LTX-2 is an audio-video model for controllable, storyboard-style generation in ComfyUI. Unlike the starter and FLUX.2 text-to-image templates, this workflow combines an **image** and a **text prompt**. 1. Download the **LTX-2 Image to Video** template and its model weights when prompted. 2. Use an image from your FLUX.2-Dev run (or another still) as the image input. 3. Write the prompt like a short shot description, not a full movie script. A walkthrough video is available here: [LTX-2 in ComfyUI](https://youtu.be/ifxOXmL351I?si=xf2_j01BAQlUhfhY0). **Prompt tips for LTX-2** Write a single flowing paragraph in the present tense, or a simple script-style format with scene headings, action, character names, and dialogue. Aim for four to six sentences that: - Establish shot and scene (wide / medium / closeup, lighting, color, textures, atmosphere) - Describe action as a clear sequence, with visible character traits, body language, and camera moves - Add audio (ambient sound, music, dialogue in quotation marks) Match detail to shot scale — closeups need more precise character and texture detail than wide shots. Useful prompt ingredients include camera movement language, shot type, pacing, atmosphere, style, lighting, emotion, and voice/audio. **VRAM and quality** LTX-2 uses significant VRAM. Memory use rises with resolution, frame rate, length, and steps. ComfyUI can stream weights to system memory when GPU memory runs short — that can keep a run going at the cost of speed. Constrain resolution, length, and steps for reasonable generation times on your hardware platform. Learn more in the [Quick Start Guide for LTX-2 In ComfyUI](https://www.nvidia.com/en-us/geforce/news/rtx-ai-video-generation-guide/). # Step 5. Combine FLUX.2-Dev and LTX-2 in one workflow (optional) To avoid hopping between workflows when turning a still into video: 1. Open your saved **FLUX.2-Dev Text to Image** workflow. 2. Ctrl+left-click the FLUX.2-Dev Text to Image node to copy it. 3. In the **LTX-2 Image to Video** workflow, paste with Ctrl+V. 4. Drag from the FLUX.2-Dev node **IMAGE** output to the **Resize Image/Mask** input so a blue connector appears. 5. Save under a new name, then prompt for image and video in one graph. # Step 6. Cleanup (optional) Cleanup is optional rollback — not required to finish the playbook. 1. Quit ComfyUI 2. Delete saved workflows you no longer need from the Workflows list or on disk 3. Remove downloaded model weights you do not want to keep (large `.safetensors` files under your ComfyUI model folders) 4. Uninstall or delete the ComfyUI application folder if you want a full rollback # Step 7. Next steps 1. Explore additional ComfyUI templates for advanced image and video models 2. Follow the [NVIDIA Blueprint for 3D-guided generative AI](https://github.com/NVIDIA-AI-Blueprints/3d-guided-genai-rtx) for 3D-guided image and video pipelines 3. Share work and get help in the [Stable Diffusion subreddit](https://www.reddit.com/r/StableDiffusion/) and [ComfyUI Discord](https://discord.com/invite/comfyorg) 4. For Linux host/container ComfyUI setups on other supported hardware catalogs, see [Generate Images and Videos with ComfyUI](https://build.nvidia.com/playbooks/comfyui)