Extending and Stitching Clips

Beyond the Single Clip Limit

Most local video models produce clips of 5–20 seconds at comfortable VRAM usage. For longer-form videos, you need to generate multiple clips and join them into a coherent sequence. Two challenges arise: maintaining visual and temporal continuity between clips, and avoiding the hard cut that reveals each clip was generated separately.

This lesson covers the techniques for generating clip sequences, increasing frame rate through interpolation, and assembling the final video in ComfyUI.

Clip Extension: First-Frame Continuation

The most reliable way to maintain continuity between clips is to use the last frame of clip N as the first frame of clip N+1. This is the same image-to-video technique from Lesson 4, applied sequentially:

1. Generate clip 1 using text-to-video
2. Save the output and extract the last frame using VHS Load Video → set frame_start_index to the last frame number → Load single frame
3. Use that frame as the starting image for Wan 2.2 I2V (or HunyuanVideo I2V) to generate clip 2
4. Repeat for as many clips as needed

The I2V model generates clip 2 starting from the exact visual state that clip 1 ended on. Paired with a consistent text prompt describing the ongoing scene, this produces a narrative continuation that flows naturally. Vary the motion prompt slightly for each clip to keep the video dynamic rather than repeating the same motion pattern.

Overlapping Frames for Smoother Transitions

Even with first-frame continuation, the cut between clips can look slightly jarring because of color or lighting drift. Reduce this by generating clips that overlap by a few frames at the start. When stitching, discard the overlapping frames from clip 2 and blend the transition:

• End of clip 1: frames 45–48
• Start of clip 2: frames 1–4 (generated from clip 1's last frame — slightly similar to the overlap frames)
• In the final stitch: use frames 1–45 from clip 1, discard clip 2's first 2 frames, use clip 2 frames 3 onwards

This overlap-and-trim approach softens most visible seams between clips.

Frame Rate Interpolation with RIFE

Video models typically generate at 16–24 FPS. For smooth slow-motion or cinematic 60 FPS output, RIFE (Real-time Intermediate Frame Estimation) interpolates new frames between existing ones. Install via ComfyUI-Manager (search "ComfyUI-RIFE"). The RIFE node takes your IMAGE batch, multiplies the frame rate by 2x or 4x, and outputs a higher frame count batch.

Typical RIFE workflow:

• WanVideo Decode → IMAGE batch (24 FPS, 81 frames = 3.4 seconds)
• RIFE Interpolation node: multiplier 2 → 48 FPS, 162 frames (same duration, smoother)
• Video Combine: frame_rate 48 → smooth final output

Use RIFE multiplier 2 for most purposes. Multiplier 4 can introduce motion artifacts when the original motion is large or complex. RIFE performs best on smooth, low-motion content; fast action benefits less from interpolation.

Assembling Multiple Clips with VHS

To stitch multiple video files into one, use VHS's batch loading and combining capabilities:

1. Load each clip with a separate VHS Load Video node
2. Use the Merge Image Batches node to concatenate the frame sequences: clip 1 frames + clip 2 frames + clip 3 frames → single IMAGE batch
3. Pass the combined batch to Video Combine with the target FPS and codec settings
4. Save the final combined video

Ensure all clips are encoded at the same resolution and FPS before merging — mismatched dimensions cause errors in Merge Image Batches. If clips are at different resolutions, add an Image Resize node after each VHS Load Video to normalize them to the target resolution before merging.