SVI 2.0 PRO: Infinite-Length Videos in ComfyUI (Uncensored)

In digital content creation, generating long image-to-video sequences is becoming increasingly important. With SVI 2.0 Pro (Stable Video Infinity) creators can produce extended videos that maintain smooth motion and visual coherence. In this tutorial, we will guide you through using SVI 2.0 Pro together with the Wan2.2 Remix video model within ComfyUI. This combination allows for the creation of uncensored, long-duration videos suitable for a wide range of creative projects. By the end of this guide, you will have the knowledge to harness these tools and enhance your video generation capabilities. Let’s get started.

⚠️ Important: Use this tool responsibly. Ensure all uncensored and long video generation is ethical and complies with applicable laws.

Before starting, make sure your system meets the hardware and software requirements to run SVI 2.0 PRO with the Wan2.2 Remix video model in ComfyUI. For best results and to experiment with higher resolutions, we recommend an RTX 5090 GPU, or using RunPod for cloud-based GPU access.

Requirement 1: ComfyUI Installed

You’ll need ComfyUI installed either locally or on a cloud service. If installing locally on Windows and have enough VRAM, follow this guide:

👉 How to Install ComfyUI Locally on Windows

If you don’t have a powerful GPU, consider running ComfyUI on RunPod with a persistent network volume. For a step-by-step walkthrough, check out the dedicated article and video:

👉 How to Run ComfyUI on RunPod with Network Volume

Requirement 2: Download Models for SVI 2.0 PRO + Wan2.2 Remix

Download the necessary model files and place them in the correct directories within ComfyUI. Only the text encoder, diffusion models, VAE, and the SVI 2.0 PRO HIGH and LOW LoRAs are required. Optional Lightning LoRAs are not needed for this setup.

Verify Folder Structure

Confirm that your folders and files look like this:

ts
1📁 ComfyUI/
2└── 📁 models/
3    ├── 📁 text_encoders/
4    │   └── nsfw_wan_umt5-xxl_fp8_scaled.safetensors
5    ├── 📁 diffusion_models/
6    │   ├── Wan2.2_Remix_NSFW_i2v_14b_high_lighting_v2.0.safetensors
7    │   └── Wan2.2_Remix_NSFW_i2v_14b_low_lighting_v2.0.safetensors
8    ├── 📁 vae/
9    │   └── wan_2.1_vae.safetensors
10    └── 📁 loras/
11        ├── SVI_v2_PRO_Wan2.2-I2V-A14B_HIGH_lora_rank_128_fp16.safetensors
12        └── SVI_v2_PRO_Wan2.2-I2V-A14B_LOW_lora_rank_128_fp16.safetensors

Once all files are correctly placed and organized, you’re ready to load the SVI 2.0 PRO + Wan2.2 Remix workflow in ComfyUI and start generating long image-to-video sequences. Next, we’ll cover downloading and loading the workflow.

With your environment and model files in place, the next step is to download and load the SVI 2.0 PRO + Wan2.2 Remix JSON workflow into ComfyUI. This workflow contains all the nodes, model references, and configurations required to run the model properly for long image-to-video generation.

Step 1: Download the Workflow File

Begin by downloading the workflow file specifically designed for SVI 2.0 PRO with Wan2.2 Remix. This file ensures that all nodes and settings are correctly configured for smooth image to video generations. You can find the download link below:

👉 Download Stable Video Infinity (SVI) 2.0 Pro Workflow JSON

Step 2: Load the Workflow in ComfyUI

After downloading the workflow file, open ComfyUI and drag and drop the .json file onto the canvas. This will load the full workflow, allowing you to start generating long image-to-video sequences immediately.

Once the workflow is loaded, you’re ready to move on. In the next section, we’ll go through configuring the workflow for optimal long image-to-video generation with SVI 2.0 PRO.

SVI 2.0 PRO allows you to generate long-form videos from images and prompts. To get consistent results, it’s crucial to load the correct models, maintain image proportions, and set frame and sampling parameters carefully. The workflow below focuses on the settings that have the biggest impact on your output.

Must-Have Settings for SVI 2.0 PRO Long Video Generation

When generating long videos, you can extend your original clip using multiple extended groups. In this example, we’ll start with an initial 5-second clip generated from the uploaded image and extend it with 3 additional groups, resulting in a total of 20 seconds (original 5s + 3 × 5s extensions). You can adjust the number of extended groups to make the video as long as you like.

1. Models & LoRAs

Before generating videos, it’s critical to select the correct models and LoRAs. Loading the right files ensures proper image-to-video conversion, accurate lighting, and smooth motion.

2. Initial Image & Width/Height

• Upload your starting image.

• Set width and height in the configuration.

• Preferably keep the same ratio as the original image to avoid distortion.

3. Sampling Steps & Motion Control

• Steps: 4

• Split Sigma Steps: 2

• ModelSamplingSD3 Node: 8 (Higher values can produce more motion, but can also increase distortion. Test to find the best balance)

4. Frames & FPS

• Frames per segment: 85 (≈ 5 seconds per group)

• Frame rate: 16 FPS (Keep the frame rate consistent across all extended groups for smooth motion)

5. Prompts & Seeds for Extended Groups

• For each extended group:

  1. Provide a detailed prompt to guide the extension:
     Prompts should clearly describe what the subject is doing next, how the movement evolves, and how the camera follows the action. Including movement direction, speed, posture changes, and camera tracking helps create fluid, dynamic motion across extended groups and maintains visual consistency throughout the video.

  2. Choose a unique random seed, or reuse the previous seed if the previous rendering was good.

  3. Maintain a consistent frame rate across groups.

Now that we have all the settings configured correctly, you can click RUN to start generating your video. Keep in mind that the first run may take longer because the models need to be loaded into memory, but subsequent runs will be relatively fast.

💡 Note: You can add as many extended groups as needed to continue the video—enabling effectively infinite-length generation, limited only by your available GPU resources.

In the next section, we will showcase a longer generated video to demonstrate how the extended groups and settings work in practice.

The example below was rendered on an RTX 5090 at a resolution of 1920×1080, using 85 frames per extended group with a random seed. The total render time was approximately 1,100 seconds, producing a final video length of 20 seconds. Despite the longer duration, the video maintains strong temporal consistency and smooth motion across the entire sequence. Keep in mind that higher width and height values significantly increase render time, while lowering the resolution will reduce rendering duration accordingly.

Below, we present a sensational example of a woman mastering her gym routine 💪🔥—I’m not even sure what she’s training for, but I’m definitely not complaining.

By following this guide, you now have the knowledge to generate long, image-to-video sequences in ComfyUI using SVI 2.0 PRO and the Wan2.2 Remix video model (uncensored). From loading the workflow, selecting the correct models and LoRAs, to configuring extended groups, frames, and motion settings, you’ve learned how to create smooth, dynamic, and high-quality video outputs.

With this workflow, you can experiment with "infinite" length videos if you have the resources and GPU capabilties, multiple prompts, and unique seeds, giving you full creative control over your video generation projects. Remember to always maintain ethical use and respect copyright and personal boundaries when generating content.