How to split a 60-second AI video into shots that still read as one scene

A sixty-second AI video is never one prompt. Every generator on the market taps out somewhere between four and thirty seconds in a single generation, and the ones that let you ask for a longer render tend to drift — faces soften, hands multiply, the camera wanders off the subject it opened on. The way a sixty-second scene ships is a split: eight to twelve deliberate shots, each one in a model's clean range, stitched back together in the edit. That is what every ranking guide on this query says. What they do not say is how many credits the stitch costs on each engine, which engine lets you avoid the stitch entirely, and when the stitch is cheaper than the one-render.

I run a studio where all five of the current AI video engines sit on one shelf and bill from one wallet. Here is the shot plan for a sixty-second scene, written against the real numbers in lib/models.ts.

The length each engine renders in one generation

engine	one-render max	silent 720p / sec	voiced 720p / sec	audio toggle
Seedance 2.5	30s (hard cap)	70	70	always on
Kling O3 Pro	15s per shot (we join to 30s)	24	30	toggle
Seedance 2.0	15s	45	45	always on
Seedance 2.0 Fast	15s	25 (480p) / 50 (720p)	same	always on
Veo 3.1 Fast	8s (enum: 4, 6 or 8 — not a range)	20	30	toggle

Three numbers do most of the planning: 30 seconds on Seedance 2.5 in a single render, 15 on Kling and the two Seedance 2.0 tiers, 8 on Veo. Sixty seconds is at least two renders on 2.5, at least four on Kling or 2.0, and at least eight on Veo.

Option A — Seedance 2.5, two renders

Thirty seconds is the hard ceiling on Seedance 2.5. For a 60-second scene the shortest path is two renders at 30s each. The engine films inner cuts in a single pass — one generation can hold three or four separate shots with the camera moving between them — so the stitch is one hard cut in the editor at the 30-second mark, and the model does the rest.

part	length	credits (720p, voiced-always)
1 of 2 — act one, three inner cuts	30s	30 × 70 = 2,100
2 of 2 — act two, three inner cuts	30s	30 × 70 = 2,100
total	60s	4,200

At the Creator credit rate that is $31.50 for the finished 60-second video. At Studio annual it is $23.10. The ceiling here is quality of continuity between the two renders, which is what the thirty-reference catalogue on Seedance 2.5 is for — you load the same character sheet into both and the identity holds across the cut.

Option B — Kling O3 Pro, four 15-second shots

Kling caps one generation at 15 seconds. The studio's own joiner (lib/segments.ts) joins two 15-second shots into one 30-second render by rendering the second shot with the last frame of the first as a seed; four 15s renders cover 60. All four hold the same @Element1 through the cuts.

part	length	credits (720p, voiced)	credits (720p, silent)
1 of 4 — opener	15s	15 × 30 = 450	15 × 24 = 360
2 of 4	15s	450	360
3 of 4	15s	450	360
4 of 4 — close	15s	450	360
total	60s	1,800	1,440

Voiced 60s on Kling at 720p is 1,800 credits. That is $13.50 on Creator and $9.90 on Studio annual — a quarter of the Seedance 2.5 bill, and the character stability across cuts is what Kling is on the shelf for. The tradeoff is 720p only; there is no 1080p row on Kling here.

Option C — Veo 3.1 Fast, eight 8-second shots

Veo's enum is 4, 6 or 8 seconds — three discrete values, not a range. A sixty-second scene on Veo is seven renders at 8 seconds (56s) plus one at 4 (60s), or eight at 8 with a 4-second trim in the editor.

part	length	credits (720p, voiced)	credits (1080p, voiced — same price)
1 of 8	8s	8 × 30 = 240	240
2 of 8	8s	240	240
3 of 8	8s	240	240
4 of 8	8s	240	240
5 of 8	8s	240	240
6 of 8	8s	240	240
7 of 8	8s	240	240
8 of 8	4s	4 × 30 = 120	120
total	60s	1,800	1,800

Veo and Kling quote identically at the 60s mark on voiced 720p — 1,800 credits. The difference is Veo's native audio (the lip-sync lands in the generation, no pass needed after) and the 1080p row that comes at the same price as 720p. Veo's reference mode is pinned to 8 seconds, so a reference render is a reference render — the slider cannot ask for 6s with references on this engine.

Option D — the hybrid

A sixty-second scene almost never wants one engine for every second. The decision rule the studio uses, per shot:

• Shot is a wide establishing frame or a place-first opener, 8s or under, native audio: Veo 3.1 Fast at 1080p.
• Shot is a two-shot or close-up on an identity through a camera move: Kling O3 Pro at 720p.
• Shot wants an inner cut (two camera angles inside one render, no editor stitch): Seedance 2.5.
• Shot is a product-only beat, no face, up to 15s: Seedance 2.0 Fast at 480p (25 cr/s).

A worked hybrid for a 60-second scene with two characters, one establishing wide, one monologue and three dialogue two-shots:

part	engine	length	720p voiced
0:00–0:08 establishing wide, native audio	Veo 3.1 Fast	8s	240
0:08–0:16 monologue close	Kling O3 Pro	8s	240
0:16–0:28 two-shot dialogue	Kling O3 Pro	12s	360
0:28–0:40 close reaction + whip	Kling O3 Pro	12s	360
0:40–0:52 two-shot payoff	Kling O3 Pro	12s	360
0:52–1:00 wide close, native audio	Veo 3.1 Fast	8s	240
finished 60s	—	60s	1,800

The hybrid lands at the same 1,800 credits as the Kling-only or Veo-only version, and it buys the one thing a single-engine stitch cannot: a native-audio opener and a native-audio close around four Kling shots that hold the face through every cut.

Why nobody else prices the stitch

The comparison every blog ranks on for this query is "which engine wins" or "which single-pass model can shoot a scene in one render". The question nobody is answering is "which engines for which seconds of this scene", because nobody is running all five on one credit currency. The second a shelf lets you quote a Veo opener and a Kling middle in the same wallet, the hybrid becomes the right answer, and the hybrid has a shape — 240 + 240 + 360 + 360 + 360 + 240 — that is not available on any single-engine credit sheet.

How to actually assemble this

The split-and-stitch workflow in the studio is one render card per shot. You pick the engine on the card, you lock the character set once in the catalogue, you drop each shot's seed frame into the next shot's reference where the join is tight, and you render in parallel — the account's concurrent-render limit is what decides how fast the eight-shot Veo plan finishes, not the model latency. The editor then concatenates in order, music over the top, captions beat-aligned, finished MP4 out at 9:16 or 16:9.

The three-step rule the studio runs under:

1. Plan the length first. Decide the scene's length before you touch an engine. Sixty seconds is eight 8s shots on Veo, four 15s shots on Kling, two 30s renders on Seedance 2.5. The plan is a line per shot before anything renders.
2. Lock the characters once. Reference sheet into the catalogue, every shot pulls the same tag. Not a new reference per shot, which is how faces drift.
3. Render in parallel, stitch in series. All shots render at once up to the account's concurrent limit; the editor timeline is the single source of truth for order, not render completion time.

A sixty-second AI video is a planning problem that costs $9 to $31 of credits once the plan is right.

FAQ

How do I make a 60-second AI video?

Split it into shots that each sit inside one engine's clean range, then stitch them in the edit. Sixty seconds is two 30-second renders on Seedance 2.5, four 15-second shots on Kling O3 Pro, or eight shots on Veo 3.1 Fast.

How long can each AI video engine render in one generation?

Seedance 2.5 renders up to 30 seconds, Kling O3 Pro, Seedance 2.0 and Seedance 2.0 Fast up to 15, and Veo 3.1 Fast 4, 6 or 8 seconds. Longer single renders tend to drift, which is why a long scene ships as a split.

How many credits does a 60-second AI video cost?

At 720p voiced, 60 seconds is 4,200 credits on Seedance 2.5, and 1,800 credits on Kling O3 Pro, on Veo 3.1 Fast, or on a Veo and Kling hybrid. On ReelDouble that is roughly $9 to $31 of credits depending on engine and plan.

How do I keep the same character across all the shots?

Lock the character once. Put the reference sheet in the catalogue and have every shot pull the same tag, rather than a new reference per shot, which is how faces drift. On Kling, all four 15-second shots hold the same @Element1 through the cuts.

Can I render all the shots at the same time?

Yes. In the ReelDouble studio each shot is its own render card, and all shots render in parallel up to the account's concurrent-render limit. The editor timeline, not render completion time, sets the order. Plan your shots.