Background seasons hypercut — lock the subject, swap the background
A true hyperlapse drifts. The camera moves a little, the framing shifts a little, and across four cuts the subject ends up in four slightly different places. A locked-subject background swap removes the drift entirely. The background is the variable. The subject is the constant.
What the technique is
Start with one locked source shot. Camera does not move. Subject does not move. Run N copies of that shot through a background-replacement pass — one pass per target state. The subject is masked and held pixel-for-pixel; the background outside the mask gets rewritten to the new condition. Cut between the copies on a fast rhythm and the result reads as time passing on a single anchor.
The other win is creative control. A true hyperlapse delivers whatever the weather did. The swap delivers exactly the states the script needs.
When to use it
Any time the subject is the point and the environment is meant to read as variable. Time-passing transitions in a montage. Anchored before-and-after pairs. A character or product in the same pose across different worlds, on demand.
How Hybrig encodes it
The Background seasons hypercut workflow template in /studio drops the source plate, a foreground mask, and a list of background-replacement prompts onto the canvas. One Remotion composition stitches the swapped copies on a rhythmic cut. Pair the cuts to a music sting if you have one — 8 to 24 frames per beat depending on how many states you are cycling.
Background replacement on a per-frame video pass is metered cloud work today. Per [[project_cloud_polish_by_beat_type]] you finish the local edit first, identify the one or two beats per spot where a seasons hypercut actually earns its money, then send only those beats through the cloud pass. Do not run this on every shot in the timeline. Pick the beat where time-passing IS the story.
Per [[feedback_no_batch_paid_jobs]] run the cloud passes one at a time. The fal queue can eat the budget on stuck jobs faster than you can say batch.
The render-process workflow
Five steps run in order. The same source plate enters the cloud edit pass N times with N different prompts; the N results assemble into one rhythmic hypercut.
Step 1 — Source capture
Node: video-upload for a real locked tripod plate, or flux-lora-still → wan-i2v for an AI-generated source plate. In: a recorded MP4 (2-4 seconds is plenty) or a prompt pair. Out: a single locked source clip where the foreground subject is sharp and the background reads as a single environmental state. Why: every downstream pass operates on this plate. The lock is non-negotiable — if the camera or subject drift, the resulting hypercut drifts with it and the technique loses its read.
Step 2 — Mask the foreground subject
Node: not yet a dedicated palette node — see open questions. Today the subject is keyed by a single mask frame the cloud edit holds for the clip duration. In: the source clip plus a one-frame silhouette of the subject. Out: a held alpha mask passed to every downstream edit job, identifying which pixels are foreground (hold) and which are background (rewrite). Why: the cloud edit pass needs to know what to hold. Without an explicit mask the model is free to rewrite the subject too, and the pixel-registration that makes the hypercut work breaks.
Step 3 — N background-replacement passes, one at a time
Node: not yet a dedicated palette node — see open questions. Today each background swap is a manual cloud handoff per target state. In: the source clip, the foreground mask, and one prompt per target state. Out: N background-swapped clips, each identical in foreground pixels and different in background pixels. Why: the subject pixel-registration across all N clips is what makes the hypercut read as time passing on a single anchor rather than as N loosely related shots. The job runs serially, never batched, per [[feedback_no_batch_paid_jobs]] — metered cloud compute gets billed by the second and a single bad batch can erase a budget.
Step 4 — Assemble the rhythmic cut
Node: remotion-timeline. In: the N background-swapped clips plus a target cut rhythm (8-24 frames per beat depending on how many states the sequence cycles). Out: a single assembled hypercut clip where each background state holds for the chosen beat length and cuts to the next. Why: Remotion owns the timeline assembly. The composition can pair the cut rhythm to a music sting frame-accurately, which is what makes a two-second hypercut feel composed rather than chopped.
Step 5 — Encode and emit
Node: video-output. In: the assembled hypercut from step 4. Out: an MP4 on disk tracked in /operations. Why: terminal node. Every Hybrig graph ends here so the file is registered with the operations dashboard.
Studio workflow shortcut
The seasons hypercut template is not yet a one-click entry in the /studio palette — the mask + N background-edit stage is held together today by manual cloud handoffs, not by a single drag-and-drop node. Once the mask and background-replacement nodes land (see open questions), the template will live at /studio?template=background-seasons-hypercut and drop the full chain onto the canvas in one click. Until then, assemble it by hand using the node types named in section 5.
Example use cases
- Real estate listings. Same property, four seasons, one locked exterior shot. The house holds; the trees, sky, and ground state cycle through spring, summer, fall, winter.
- E-commerce product reveals. Same product on the same pedestal, four lifestyle backgrounds. Studio white, kitchen counter, beach towel, gym floor — the product reads as the constant and the context as the menu.
- Automotive content. Same car parked in the same pose, four different terrains. City street, mountain switchback, desert highway, snow pass. Paint, badge, and wheels stay pixel-identical across cuts.
- Roofing and home services. Roof-aging story beats. Same roof from the same angle, four weather states hammering it across compressed time.
Open questions
- No dedicated cloud-edit background-replacement node is in the Studio palette today. The workflow template wires source video + mask reference + prompt list using generic primitives and notes the cloud-edit step in the description. When the backing node lands, the template swaps to it.
- Auto-masking the foreground subject across the full clip is a future pass. Today the mask is a single keyed frame the cloud edit holds for the duration.