Important Suggestions for Rendering Videos Over 200 Frames
Rendering long videos (above 200 frames) can be challenging in ComfyUI. To ensure success, follow these guidelines:
1. Remove the "Video Combine" Node. Instead, use the "Save Image" node.
Reasons for Removing the 'Video Combine' Node
- ⚠️ The VHS Video Combine node often crashes silently when processing hundreds of frames (200+), depending on the resolution. Therefore, removing it would be safer.
- You cannot place both the "Save Image" node and the "Video Combine" node at the same level, as the execution order might be uncertain to you. The VHS node might run first and crash, preventing the "Save Image" node from executing.
- To determine the execution order, use ComfyUI-Manager:
- Turn on the Badge: #ID Nickname.
- A small number will appear on top of each node in your workflow. Nodes execute in ascending order based on these numbers if they are at the same level.
2. Output Path Configuration
Please see the detailed information here:
⚠️ How to Save Image/Images by Path in ComfyUI?
⚠️ Make sure you set the correct path, or you will not find the result in the file browser.
3. Use Meta Batch Manager
ComfyUI-VideoHelperSuite has this nice node that works very well with splitting video frames into batches thus less stress on the RAM and VRAM.
- Add “Meta Batch Manager” Node into the workflow, connect it to Load Video node and Video Combine node. 64 frames is a good starting value to test out the batch size.
- Best use case for Meta Batch Manager
- It works well with models that do not require a context, for instance, upscaling models, face detection and SAM2, etc.
- But not so well for models that need context, video to video transformation/restyle, etc. Of course you should test it out and see if you need to do some stitching when you must use batching due to limited resources.
4. Limiting the Number of Frames
Use a machine with sufficient resources:
- Large Machine: 24GB VRAM, 48GB RAM
- Extra Large Machine: 48GB VRAM, 48GB RAM
⚠️ Render no more than 200 frames at a time:
- Limit each render to no more than 200 frames to avoid running out of memory.
So, If you have a long video with more than 200 frames, you need to render multiple times.
- Set the "frame_load_cap" to 200, meaning that you will render 200 frames at a time.
- Use "skip_first_frames" to decide which frame to start from. For example, if you want to start from the 101st frame, set "skip_first_frames" to 100.
5. Always start with shorter clips
Please test the workflow with short clips such as 10 to 20 frames, tweak the hyper parameters and such to make sure the models work well with your video/images. Once you are satisfied with the test result, gradually increase the clip duration, e.g. 1s, 2s, 5s, 10s…