Magnific generates multi-shot videos from start and end frames

• A 15-second fish-market sequence jumps from a hidden tentacle to a gravity-flipped flooded boulevard in one generation, after the team set endpoint frames, according to Magnific's workflow post.
• The shared prompt scripts five three-second cuts, including framing, action, camera behavior, sound, and a warm-to-cool grade shift, in Magnific's five-shot prompt.
• The key constraint is to keep one element stable while increasing only one variable, then save the reveal for the final two seconds, as Magnific's shot-planning guide puts it.

Magnific's Video Generator documentation calls the two-reference option “Start and end image”: the system generates the transition between them. The fish-market example turns that control into a tiny prewritten edit, with cuts, escalation, and a final impossible wide shot all specified before generation.

The images in Magnific's reference-image post establish radically different bookends: an overhead fish stall at night and a flooded European boulevard where cars, boats, a tram, and birds hang upside down.

The official docs describe the same setup as a start image, an enabled end-image toggle, then a prompt describing the motion between them. That makes the endpoint images part of the direction, rather than loose inspiration for a text-only prompt.

Magnific's full prompt divides the 15 seconds into this shot list.

1. 0-3s: A gloved vendor slices fish while a tentacle steals a lemon and knocks over crates.
2. 3-6s: A red-haired woman walks past the market, unaware of the disturbance.
3. 6-9s: An octopus erupts from a storm drain and destroys her sandwich.
4. 9-12s: First-person view drops through the drain into open water.
5. 12-15s: An aerial pullback reveals the flooded street turned upside down, ending on a freeze frame for a logo card.

The prompt also locks a photoreal handheld-documentary look, hard cuts at impact moments, and a color progression from golden dawn to deep underwater blue. Each beat has one dominant action and a stated camera instruction.

The workflow's compact rule set comes from Magnific's shot-planning guide.

• Number shots and timestamp them.
• Hold one element constant across the sequence.
• Escalate one variable, not five.
• Place the reveal in the last two seconds.

Here, the market is the continuity anchor. The escalation travels from a barely noticed tentacle, to a public attack, to a submerged POV, then to a city-scale reversal. The prompt operates as a compact storyboard.

Magnific's Video Generator docs separate three approaches that can look similar in the final clip:

• Start image: the system animates outward from one reference frame.
• Start and end image: it generates the transition between two references.
• Multi-shot mode: it accepts separate prompts and durations for up to six shots in one generation.

The same documentation says multi-shot trades some transition control for speed, while separately generated clips offer more control in the Video Editor. Its model guide also notes that availability varies by model, resolution, and duration, so the endpoint-frame approach and the shot list are related controls, not a universal setting across every video model.