A five-second Gen-4 clip on Runway costs 60 credits at 12 credits per second and a ten-second clip costs 120, Gen-4 Turbo cuts that to 25 or 50 for the same five- or ten-second duration caps, and Kling VIDEO 3.0 stretches continuous motion from three to fifteen seconds with pricing that jumps from 30 credits for a five-second 720p clip without native audio to 60 for five seconds of 1080p with native audio — yet most overnight reruns start because the hero still that fed I2V was still soft on face, wardrobe, or framing when the first motion credit burned.

Finish the AI video shot list before that burn, not after.

Runway’s Gen-4 help article is blunt about the box you are working inside: image-to-video requires an input image, durations are five or ten seconds only, Gen-4 Turbo bills at five credits per second (25 for five seconds, 50 for ten), and Gen-4 bills at twelve credits per second (60 for five seconds, 120 for ten). The same docs recommend testing in Turbo then moving to Gen-4, cap prompts at 1,000 characters, and cite common output shapes such as 16:9 at 1280×720 and 9:16 at 720×1280 at 24 fps (https://help.runwayml.com/hc/en-us/articles/37327109429011-Creating-with-Gen-4-Video). Kling’s VIDEO 3.0 guide sets a different envelope: flexible continuous clips from three to fifteen seconds, with I2V driven by a Start Frame plus optional Element Reference, and Multi-Shot or Custom Multi-Shot available on 3.0 (https://kling.ai/quickstart/klingai-video-3-model-user-guide). Kling’s per-second credit table is equally concrete: Native Audio at 1080p is twelve credits per second and at 720p is nine; without native audio, 1080p is eight and 720p is six; Voice Control adds two credits per second. That is why a five-second Native Audio 1080p pass lands at sixty credits and a five-second No Native Audio 720p pass lands at thirty.

Those numbers are the unit cost of every unfinished still you promote into motion. A soft face on the Start Frame does not cost “a little more prompt work.” It costs another full duration at the rate above, plus render time, while you hope prompt adherence holds an identity you never locked.

I2V does not cheaply repair identity. It animates whatever you handed it. If the hero still drifts on eye shape, wardrobe color, or crop, the motion model inherits that drift and burns a full five-, ten-, or fifteen-second credit block to show you the same mistake moving. CapCut or any NLE can assemble takes afterward, but neither is the system of record for the hero face or the shot budget. Assembly sits after approval. Credit burn sits before it.

On Kling, Elements are the production path for subject stability: supply two to four reference images, or a subject extracted from video, then pair them with a Start Frame so I2V has both pose and identity to follow. Runway’s Gen-4 path still starts from a required input image, so the still you approve is the continuity contract. Build character refs in the still stage first. Match wardrobe, palette, and framing across the Start Frames. Only then hop into motion. Persistent character and universe locks belong upstream of the first I2V credit, not inside failed motion retries.

An AI video shot list is not a moodboard caption list. It is a duration budget mapped to the caps of the model you will actually run. For Runway Gen-4 or Turbo, every row resolves to five or ten seconds. For Kling VIDEO 3.0, every row resolves somewhere between three and fifteen continuous seconds, and you decide whether that row needs Multi-Shot coverage or a single directed beat. Write columns production can enforce: shot ID, Start Frame status, duration target, model and mode, credit estimate at the rates above, Element or character-ref status, and audio path. When the sheet totals more credits than the week’s budget, cut shots before you burn motion. When a row still says “still TBD,” it does not enter the I2V batch.

Batching is the opposite of exploring identity inside the video model. Approve the stills, lock the refs, fill the duration budget, then send a coherent I2V queue: same character set, same framing rules, same prompt-adherence checklist, duration chosen on purpose. On Runway, that often means a Turbo pass for motion and timing at five credits per second, then Gen-4 on the winners at twelve. On Kling, that means Start Frame plus Elements for identity-critical shots, and a conscious choice of 720p versus 1080p and native audio versus silent based on whether the cut needs dialogue in-camera. Render time becomes predictable when the queue is not also doing identity R&D.

Havincy’s production loop is Idea, Create, Produce, Publish, Measure, Improve. Persistent characters and universes live in Create so the same face and world survive across stills before any motion hop. Image generation and editing lock the hero frames. Video and Director sit in Produce once those frames are approved, so I2V consumes a finished shot list instead of inventing one mid-render. Social publishing and analytics close the loop after the cut ships, where you learn whether the duration budget matched the platform. CapCut remains assembly. The system of record for hero face and shot budget stays upstream.

No. Identity iteration belongs in stills and character refs. Every I2V retry on a soft face pays the full per-second rate for motion you will discard. Lock the Start Frame, attach Elements when the subject must hold, then animate. If the face fails in motion, fix the still and re-enter the batch; do not treat the video model as a cheaper face editor.

Multi-Shot on Kling VIDEO 3.0 earns its keep when one continuous three-to-fifteen-second pass needs coverage that would otherwise mean several separate I2V burns and a harder assemble. If the beat is a single directed move on one locked Start Frame, a shorter single-shot clip usually burns fewer credits and keeps prompt adherence tighter. Budget the longer Multi-Shot clip when the cut needs that coverage inside one continuous duration, not because the shot list was never written.

Yes, when you are on Runway. The Gen-4 help docs recommend testing in Turbo at five credits per second, then moving approved motion to Gen-4 at twelve. Use Turbo to validate camera path, timing, and whether the locked still holds under motion. Spend the Gen-4 premium on the takes that already cleared the shot list, not on hunting for a hero face you should have finished in Create.