All In One NSFW Long Video Workflow

All in One Workflow for QWEN/WAN

This workflow allows you to create long videos by chaining videos together. The core strategy is to use QWEN to generate key frames then use a start image and end image or start video and end image to perpetually chain video segments. The key frames maintain consistency across video segments.

Models

This workflow uses two models that are capable of producing the nsfw content.

QWEN Rapid AIO: https://huggingface.co/Phr00t/Qwen-Image-Edit-Rapid-AIO/tree/main

Unfortunately this is a big model and I'm not sure if there's a gguf anywhere for it. It's also a very important model because it allows us to make the keyframes.

WAN 2.2 NSFW:

WAN (high) - /model/2053259?modelVersionId=2477552

WAN (low) - /model/2053259?modelVersionId=2477555

The WAN model does have ggufs available.

Nodes

This workflow does use a lot of custom nodes but I think most of them are fairly common. The one that is not too common is the painter long video node which can be found here.

https://github.com/princepainter/ComfyUI-PainterLongVideo

Steps

*Note: There is a group bypasser by each workflow. Only one should be enabled at a time.

1. Enter project name. All files are stored in subfolders of that project to keep everything together. You can edit the filenames on the individual sub flows.

2. Generate some starting characters (optional). ZIT is included in the workflow but you can use any method you want to get starting characters.

3. Generate a scene. Start by prompting qwen to adjust each individual character first in the scene generator. Once you have the characters you want then combine them in the scene generator.

4. Generate some key frames. You don't have to generate them all up front. Since the video generation works by chaining videos any number of times, you can start working on the video and then go back and generate for key frames later. Just be sure to always have some reference images generated that have faces clearly visible to help with consistency.

5. Generate initial video. Start by using a start image and end image and ensure mode 1 is selected on the video subgraph options. It uses an input selector so mode 1 uses the start image and mode 2 uses the start video.

6. Continue video. Select mode 2 in the subgraph and upload the video you want to extend. The workflow always outputs 2 videos, the first is the newly generated video prior to upscale. The second video is created by first upscaling the newly generated video and stitching it to the input video. Always use the final upscaled video when chaining.

7. Finally you can interpolate with rife if desired.

The video is always upscaled so the latest video is more or less complete. If you have memory issues or it's too slow for you then you can just remove all the upscalers in the wan subgraph. I prefer to have it upscaled because the intent of the workflow is to be able to come back and extend your video at any time.

Usage Tips

• When generating key frames make sure you're generating from an image that clearly shows the character faces.

• I like to generate an image that's bigger than what I'm generating the video at and then crop what I want for the shot in the video. This avoids the need to prompt WAN for camera action because it will just naturally pan and zoom based on the key frame (assuming you have enough information between the start and end image for it to recognize). I use GIMP to crop with aspect ratio maintained.

• QWEN is pretty poor at editing an image with more than two people. Often times it's easier to generate an image with two people and then insert more once you have it the way you want.

• The project name is basically the folder you want to store everything in. Each group also has a filename input so that you can further sub divide where everything is stored. It doesn't all have to be dropped in one folder.

One Last Thing

The reason there are two different Qwen workflows is because of some funkiness in the qwen edit plus node. Having the vae connected causes awful zooming and other issues. Use the first one (generate key frames) for more subtle changes when you are stitching together two scenes and need most aspects of the generated image to be the same as the first. Use the second one (Generate Scene) when the changes are more drastic or you are adding people to the scene. The keyframe generator does have an input for image size as its intended to make minor tweaks to the start image so it keeps the same size.