Scene Ripper includes 20 sequencer algorithms that arrange your clips into a sequence. Each algorithm uses a different creative logic to determine the order.
To use a sequencer, go to the Sequence tab, select an algorithm from the dropdown, and click Generate. Some algorithms require that you run specific analysis on your clips first (in the Analyze tab). Others open a dialog where you configure additional options before generating.
Some algorithms (like Storyteller and Exquisite Corpus) require a cloud API key. See the API Keys Guide for setup instructions.
A project holds any number of named sequences — run 20 different algorithms and keep every result side by side for comparison. Every algorithm run creates a new sequence instead of overwriting the previous one.
The leftmost dropdown in the Sequence tab header (and in the card view) shows all sequences in the project. Click a name to switch. The timeline, algorithm dropdown, and chromatic bar checkbox all update to reflect the selected sequence.
Click New Sequence to create a blank "Untitled Sequence" and switch to it. When the current sequence is empty, running an algorithm reuses it (rename + populate) instead of creating a second empty entry — avoiding orphan sequences.
If you manually drag or remove clips on the timeline and then switch sequences, Scene Ripper shows a Save / Discard / Cancel dialog so you don't lose edits. Algorithm runs don't trigger this prompt — they're always clean by design.
Changing the algorithm or direction dropdown on a populated sequence shows a Replace / Create New / Cancel dialog:
| Choice | Effect |
|---|---|
| Replace | The current sequence is overwritten with the new algorithm result |
| Create New | The new algorithm run adds a new sequence alongside the current one |
| Cancel | No change |
Right-click the sequence dropdown to open a context menu with Rename… and Delete options. Deleting an empty sequence is immediate; deleting a populated one asks for confirmation. You can't delete the last sequence — Scene Ripper always keeps at least one (a fresh empty "Untitled Sequence" is auto-created if you try).
Duplicate names are allowed; auto-naming uses a monotonic counter per algorithm (Chromatics, Chromatics #2, Chromatics #3, …).
These algorithms arrange clips based on a measurable property. Most support a direction option that controls the sort order.
Arrange clips along a color gradient or cycle through the spectrum.
Required analysis: Colors
Direction options:
| Direction | Description |
|---|---|
| Rainbow | Cycle through the full color spectrum (red, orange, yellow, green, blue, violet) |
| Warm to Cool | Start with warm tones (reds, oranges) and end with cool tones (blues, greens) |
| Cool to Warm | The reverse: cool tones first, warm tones last |
| Complementary | Alternate between complementary colors for maximum contrast |
Clips without color data can be appended to the end, excluded, or sorted inline depending on your settings.
Order clips from shortest to longest (or reverse).
Required analysis: None
Direction options:
| Direction | Description |
|---|---|
| Shortest First | Start with the quickest cuts and build to longer takes |
| Longest First | Start with longer takes and accelerate toward shorter clips |
Arrange clips from light to shadow, or shadow to light.
Required analysis: Brightness
If clips haven't been analyzed for brightness yet, this algorithm will automatically compute it when you generate. This may take a moment for large clip sets.
Direction options:
| Direction | Description |
|---|---|
| Bright to Dark | Start with the brightest clips and descend into darkness |
| Dark to Bright | Emerge from shadow into light |
Build from silence to thunder, or thunder to silence.
Required analysis: Volume
If clips haven't been analyzed for volume yet, this algorithm will automatically compute it when you generate. This uses FFmpeg to measure audio levels and may take a moment.
Direction options:
| Direction | Description |
|---|---|
| Quiet to Loud | Start with the quietest clips and build to the loudest |
| Loud to Quiet | Start loud and fade to quiet |
Arrange clips by camera shot scale, from wide establishing shots to tight close-ups (or reverse).
Required analysis: Shots (shot type classification)
Glide from distant vistas to intimate close-ups.
Required analysis: Shots (shot type classification)
Direction options:
| Direction | Description |
|---|---|
| Far to Close | Start with wide/establishing shots and move to close-ups |
| Close to Far | Start with close-ups and pull back to wide shots |
Arrange clips by where subjects are looking, from left to right or up to down.
Required analysis: Gaze direction
Direction options:
| Direction | Description |
|---|---|
| Left to Right | Start with subjects looking left and progress to subjects looking right |
| Right to Left | The reverse: right-looking clips first |
| Up to Down | Start with subjects looking up and progress to subjects looking down |
| Down to Up | The reverse: down-looking clips first |
Clips without gaze data (no face detected) are appended at the end of the sequence.
Group clips where subjects are looking in the same direction. All "looking left" clips are grouped together, then all "looking right", and so on — largest groups first.
Required analysis: Gaze direction
Within each group, clips are sorted by their actual gaze angle. Clips without gaze data are appended at the end.
These algorithms find connections between clips based on visual similarity.
Chain clips together by visual similarity. Each clip is placed next to the one it most closely resembles, creating a continuous visual flow.
Required analysis: Embeddings (visual feature extraction via DINOv2)
If clips haven't been analyzed for embeddings yet, this algorithm will automatically compute them when you generate. Embedding extraction can take a while for large clip sets.
Find hidden connections between clips at cut points. Analyzes the last frame of each clip and the first frame of the next to find the most visually similar transitions.
Required analysis: Boundary embeddings
If clips lack boundary embeddings, this algorithm will automatically compute them. This analyzes the first and last frames of each clip.
Randomly shuffle clips into a new order. Opens a dialog where you can optionally apply random transforms to each clip.
Required analysis: None
Dialog options:
- Random H-Flip: Randomly mirror some clips horizontally
- Random V-Flip: Randomly flip some clips vertically
- Random Reverse: Randomly play some clips in reverse
When transforms are enabled, the dialog pre-renders each affected clip via FFmpeg before assembling the sequence. A progress bar shows rendering status.
Keep clips in their original order. This is the simplest algorithm: clips appear in the sequence exactly as they were detected, preserving the source video's timeline.
Required analysis: None
Cut clips to the rhythm of a music track. Opens a dialog where you select an audio file, preview the waveform with beat markers, and generate a sequence where onset strength drives visual contrast — stronger beats trigger bigger visual jumps between consecutive clips.
Required analysis: Embeddings (DINOv2, auto-computed if missing)
Dialog workflow:
- Click Select Music File to choose an MP3, WAV, FLAC, M4A, AAC, or OGG file
- The audio is analyzed and a waveform is displayed with beat/onset markers overlaid
- Adjust the Sensitivity slider to control the number of cut points ("Fewer Cuts" to "More Cuts")
- Choose a Beat Strategy from the dropdown: Onsets (transients/hits), Beats (regular pulse), or Downbeats (strong beats only)
- Click Generate to match clips to beat intervals. Each clip is trimmed to fit its slot; clips shorter than their slot are looped. Clips can repeat when there are more beat slots than clips.
The algorithm uses DINOv2 visual embeddings to measure similarity between clips. At each cut point, it measures the onset strength and selects a clip whose visual distance from the previous clip matches that strength — hard hits get jarring visual jumps, soft transitions get visually similar clips.
These algorithms use language models or vision models to make creative decisions. They require a cloud API key — see the API Keys Guide.
Generate a poem from on-screen text. Opens a multi-step dialog.
Required analysis: Text extraction (OCR/VLM)
Dialog workflow:
- Enter a mood or vibe prompt (e.g., "melancholic and introspective") and select a poem length (Short, Medium, or Long)
- The dialog extracts on-screen text from your clips using OCR or a vision-language model, showing a progress bar
- An LLM arranges the extracted text into a poem, displayed as a drag-reorderable list where each line maps to a clip
- You can Regenerate for a different poem or manually reorder lines before clicking Create Sequence
Create a narrative from clip descriptions. Opens a multi-step dialog.
Required analysis: Describe (AI-generated clip descriptions)
All clips must have descriptions before using this algorithm. If some clips lack descriptions, the dialog will offer to exclude them or redirect you to the Analyze tab.
Dialog workflow:
- Optionally enter a thematic focus and choose a narrative structure (Three-Act, Chronological, Thematic, or Auto)
- Set a target duration (10 minutes to 90 minutes, or use all clips)
- The LLM arranges clips into a narrative, assigning roles like "setup," "climax," and "resolution"
- Review the narrative as a drag-reorderable list and adjust the order before confirming
Match your clips to a reference video's structure. Opens a dialog.
Required analysis: Varies by selected dimensions
Dialog workflow:
- Select a source video as the "reference" — its clips become the template structure
- All clips from other sources become the matching pool
- Adjust seven dimension sliders to control matching weights: Color, Brightness, Shot Scale, Audio Energy, Visual Match (DINOv2), Movement, and Duration. Dimensions without analysis data are grayed out
- Optionally allow repeated clips with the "Allow Repeats" checkbox
- Click Generate to find the best-matching clip for each position in the reference
Interpret a drawing as an editing guide. Opens a large canvas-based dialog.
Required analysis: Colors
Dialog workflow:
- Draw on the canvas (or import an image). The Y-axis maps to pacing (spiky = fast cuts, smooth = slow) and color maps to color matching against your clips
- Set a target duration and FPS
- Choose between Parametric mode (pixel-level analysis with a granularity slider) or VLM mode (a vision model interprets the drawing's meaning)
- Click Generate to match your drawing to clips and assemble a timed sequence
Isolate clips featuring a specific person. Named after Joseph Cornell's 1936 found-footage film. Opens a dialog.
Required analysis: None (face detection runs in the dialog)
Dialog workflow:
- Upload 1-3 reference photos of the person you want to find. Each image is analyzed for faces, and a green bounding box highlights the detected face
- Configure matching sensitivity (Strict, Balanced, or Loose)
- Choose how to order matched clips: Original, Duration, Color, Brightness, Confidence, or Random
- Set a frame sample interval (how often to check for faces within each clip)
- Click Generate to scan all clips for the matching face and build a sequence from the results
Sequence clips based on where subjects are looking. Named after the Billy Idol song and the Georges Franju film. Opens a dialog with three modes.
Required analysis: Gaze direction
Dialog modes:
| Mode | What it does |
|---|---|
| Eyeline Match | Pair clips with complementary gaze directions for shot-reverse-shot patterns. If person A looks left, the next clip shows person B looking right. Tolerance slider controls how strict the matching is (5°-30°). |
| Filter | Keep only clips where subjects look in a specific direction (at camera, left, right, up, or down). Non-matching clips are appended at the end. |
| Rotation | Sweep through a range of gaze angles, creating a progressive rotation effect. Select the axis (horizontal or vertical), set a start and end angle, and choose ascending or descending direction. |
Dialog workflow:
- Select a mode from the dropdown at the top
- Configure the mode-specific parameters (tolerance, category, or angle range)
- Click Generate to build the sequence
- Results appear on the timeline
Clips without gaze data are always appended at the end of the sequence.
Build a sequence one clip at a time with an LLM collaborator. The algorithm opens a dialog where you interactively accept, reject, or swap each proposed next clip, with a rationale from the LLM explaining each transition.
Required analysis: Describe, Embeddings
Dialog workflow:
- Pick a starting clip (the LLM won't choose the first one)
- The dialog proposes the next clip with a short rationale (why it pairs well with the current tail of the sequence)
- Choose: Accept (add to sequence), Reject (ask for a different proposal), Swap (see alternatives), or Stop (finish)
- Repeat until you're satisfied or the pool is exhausted
Embeddings power a local candidate shortlist so the LLM only sees the most similar clips at each step — this keeps prompts small and responses fast. Without embeddings the algorithm falls back to random sampling, which degrades proposal quality. Each accepted transition is saved as a rationale on the resulting SequenceClip, visible in exported SRTs.
Requires a cloud LLM API key.