One image, one motion problem
Runway introduced Gen-4 on 31 March 2025. The company framed the release around fidelity, motion and controllability, and its later Gen-4 prompting documentation describes image-to-video clips of five or ten seconds made from an input image and a text prompt. The documentation’s practical split is revealing: the image carries subjects, composition, colour, lighting and style; the text should concentrate on motion.
That division is more useful to an editor than a promise of “consistency.” It asks a team to decide what is already locked before any movement is requested. If the wardrobe, lens feeling, spatial relationships and light direction are vague in the source image, a motion prompt cannot turn them into a dependable series bible. A handsome still is an input; a continuity plan is a separate piece of work.
Build the brief before the prompt
For each recurring character or location, make a small reference sheet: approved source image, framing, screen direction, wardrobe and props, light direction, palette, scale cues, and known exclusions. Add an editorial note that is not a prompt: what must remain recognisable from the preceding shot. This is where a production can say “the red bag stays in the left hand” or “the doorway remains behind the performer” without pretending that a tool has guaranteed either outcome.
Then write a motion instruction for one observable action. Runway’s guide recommends beginning simply and adding elements one at a time; it distinguishes subject, camera, scene and style motion. That provides a sensible logging structure. Record the input version, the motion request, the output selected and the rejection reason. A clip may fail because the character changes, but it can also fail because its movement answers the wrong dramatic question. Those are different notes, and treating them as different problems stops endless prompt decoration.
Use physical, visible language. “The subject turns toward the window; locked camera” gives an editor a testable intention. “A moment of liberating self-discovery” delegates too much interpretation. Clear direction also makes a review conversation less mystical: did the subject turn, did the camera remain locked, and did the shot earn its place beside the previous one?
Edit in shot-sized units
The five- and ten-second durations described in Runway’s documentation encourage a practical constraint: make each generation a shot with a single job. Do not ask a short clip to establish a place, introduce a character, change emotion, execute a complex camera move and resolve a plot beat. Choose its primary job, then cut out when that information has arrived. The edit can create the larger motion; the generation only has to survive its entry and exit.
Review adjacent clips as pairs, not as a gallery. Check the outgoing frame against the incoming frame for geography, gaze, object position, light and the direction of movement. Listen for an audio plan even if the generated material is silent: a sustained room tone or a motivated sound can make a cut legible, while a mismatch can expose it. Freeze frames are useful here. Mark the exact frame where a hand, doorway or horizon stops matching, then decide whether the cut can hide it or the shot must be replaced.
Runway’s materials are first-party product guidance, not a guarantee that every output will maintain identity or world state. They do support a durable craft lesson: decide continuity in an explicit brief, ask for limited movement, and judge the result in the edit rather than in isolation. That keeps the director and editor responsible for the film’s memory.
Sources & context
- Runway · Introducing Gen-4 (31 Mar 2025)
- Runway · Gen-4 Video Prompting Guide (accessed 19 Sep 2026)
Sources checked 19 September 2026. Release framing and product guidance are attributed to Runway. Screen Method did not use Gen-4 or independently verify output consistency, pricing, availability, rights implications or production results. This retrospective article was prepared in 2026.
