How to render long videos (over 200 frames) in ComfyUI?

How to render long videos in ComfyUI (over 200 frames)

Important tips for rendering videos over 200 frames

Rendering long videos (over 200 frames) in ComfyUI can be challenging. To ensure success, follow these guidelines:

1. Delete the "Video Merge" node. Use the "Save Image" node instead.

Reasons for deleting the "Video Merge" node

1. VHS video merging nodes often crash silently when processing hundreds of frames (200+), depending on the resolution. Therefore, it will be safer to delete it.

2. You cannot put the Save Image node and the Video Merge node at the same level because you may not be sure of the order of execution. The VHS node may run first and crash, preventing the Save Image node from executing.

3. To determine the execution order, please use ComfyUI-Manager:

1. Open the badge: #ID nickname.

2. A small number will appear at the top of each node in the workflow. If the nodes are at the same level, they are executed in ascending order based on these numbers.

2. Output path configuration

See details here:

How to save an image by path in ComfyUI?

Make sure you set the correct path or you won't be able to find the results in the file browser.

3. Use Meta Batch Manager

ComfyUI-VideoHelperSuite has a great node that can well divide video frames into batches, thereby reducing the stress on RAM and VRAM.

1. Add the Meta Batch Manager node to the workflow and connect it to the Load Video node and Video Combine node. 64 frames are a good starting value for the test batch size.

2. Best use cases for Meta Batch Manager

It works well with models that do not require context, such as upgrade models, facial detection, SAM2, etc.

But for models that require context, video to video conversion/restyle, etc., the effect is not that good. Of course, when you have to use batch processing because of limited resources, you should test it out to see if you need to do some splicing.

4. Limit the number of frames

Using a machine with sufficient resources:

1. Large machine: 24GB VRAM, 48GB RAM

2. Ultra-large machine: 48GB VRAM, 48GB RAM

No more than 200 frames per rendering:

1. Limit no more than 200 frames per rendering to avoid memory exhaustion.

So if your long video is more than 200 frames, you need to render multiple times.

1. Set "frame_load_cap" to 200, which means you will render 200 frames at a time.

2. Use "skip_first_frames" to decide which frame to start from. For example, if you want to start from frame 101, set "skip_first_frames" to 100.

5. Always start with shorter segments

Please use 10 to 20 frames of short clips to test the workflow, adjust the hyperparameters, etc. to ensure that the model handles your video/images well. After being satisfied with the test results, gradually increase the fragment duration, such as 1 second, 2 seconds, 5 seconds, 10 seconds...