Video Post-Processing in ComfyUI

From Raw Generation to Finished Output

A raw video generation from Wan 2.2 or HunyuanVideo is already high quality, but post-processing in ComfyUI can take it further: frame-by-frame upscaling for larger output sizes, adding film grain for a cinematic feel, noise reduction for smoother renders, and proper MP4 export settings for sharing or further editing.

All of these happen inside ComfyUI using the same nodes you use for image post-processing — applied to the IMAGE batch representing your video frames.

Frame-by-Frame Upscaling

The same upscaling pipeline from the image generation track (Track 49, Lesson 7) applies to video. An IMAGE batch is just a set of images — you can pass it through any image-based node, including upscalers:

1. After WanVideo Decode, you have an IMAGE batch at the generation resolution (e.g., 480p)
2. Add Load Upscale Model → select 4x-UltraSharp
3. Add Upscale Image With Model → connect the IMAGE batch; the node processes every frame in the batch automatically
4. The output IMAGE batch is now at 4x the original resolution (e.g., 1920×1080)

Frame upscaling is slow — a 5-second clip at 24 FPS contains 120 frames, each processed individually through the 4x upscale model. The process runs sequentially. Estimate upscale time as 3–10 seconds per frame depending on GPU speed, or 6–20 minutes for a 5-second clip. Plan batch processing accordingly (queue and leave it running).

Adding Film Grain for Cinematic Feel

Film grain adds organic noise that makes AI-generated video less artificial-looking. In ComfyUI, you can add noise via the ImageCompositeMasked or custom grain nodes available through the community. A simple approach uses the Image Filter Sharpen or Add Noise nodes from the default ComfyUI toolkit.

For film grain specifically, look for grain nodes in the community:

• Search Manager for "Film Grain" or "noise overlay" custom nodes
• Grain should be fine (2–3 pixel scale) and at low opacity (5–15%) to feel authentic rather than distracting
• Apply grain after upscaling so it is at full resolution and appears at the correct perceived scale

The grain should flicker naturally between frames — do not use a fixed grain mask applied identically to every frame, as that creates a stationary noise pattern that looks wrong in motion. Most grain nodes regenerate the noise pattern per frame automatically.

Noise Reduction for Smoother Output

Some video generations, particularly at lower VRAM settings or with quantized models, produce slightly grainy or flickery output. A subtle blur or denoising pass before upscaling can smooth this out. Use the Image Blur node (Gaussian blur at sigma 0.5–1.0) or, if available, the Video Denoising custom node.

Apply denoising conservatively — too much blurring removes fine detail that the upscaler then cannot recover. The correct order is: denoise (very subtle) → upscale 4x → add film grain. This sequence removes the model's noise before upscaling, so the upscaler works on clean content, then adds intentional grain at the correct scale.

Exporting to MP4 with Video Combine

The Video Combine node from VHS handles final MP4 export. Key settings:

• frame_rate: match your target FPS. 24 for cinematic, 30 for web content, 60 if you used RIFE interpolation
• format: video/h264-mp4 for universal compatibility. H.264 plays everywhere and is the default for web sharing and video editors.
• crf: quality setting for H.264 (lower = better quality, larger file). 18 for near-lossless, 23 for good quality, 28 for smaller file size
• save_output: set True to save to disk rather than just preview
• filename_prefix: name prefix for the output file

Output files go to the ComfyUI output/ directory. For clips that need audio, import the exported MP4 into a video editor (DaVinci Resolve, CapCut, Premiere) and add audio there. LTX-Video's audio generation outputs to the same Video Combine node if that model is being used.