Edit Video Pro

Replace a span of an existing video with newly generated content — a Seedance-2-family reference bridge, seamlessly stitched back into the source. Cloud edition only.

Overview

Edit Video Pro takes an existing video and regenerates a chosen time window — the replace span — while keeping the footage before and after it untouched. It’s built for surgical, longer-form replacements on the Seedance 2 family: reshoot a scene, swap out an action beat, or restyle a moment, all while the rest of the clip stays exactly as it was.

Where Video Retake targets short (down to 2s), audio-or-video-only fixes on LTX 2.3 Pro, Edit Video Pro always regenerates both audio and video together, supports much longer spans via the same multi-segment chaining Generate Video Pro uses for long single-shot clips, and requires a minimum 4-second span.

Cloud edition only. Edit Video Pro requires a Cloud subscription.

Input handles

Handle	Direction	Accepts	Notes
`video`	target	Video producers	Required — the source clip to edit. The node reads the clip’s duration client-side to size the span slider; the server independently re-probes the source when reserving credits and never trusts that client-side hint for money.
`prompt`	target	Text producers + visual pickers	Describes what should replace the selected span
`imageReferences`	target	Image producers (ordered, multi)	Reference images carried into the replace-span generation
`video`	source	n/a	Output — the full clip with the span replaced

Configuration

Field	Type	Default	Notes
Provider	Select	`seedance-2`	`seedance-2` (full), `seedance-2-fast`, `seedance-2-mini` — Seedance-2-family only, same set Generate Video Pro uses
Prompt	Text	—	Describes what should replace the selected span; also settable via the `prompt` handle
Replace Span (From / To)	Number pair (seconds)	`0` – `8`	The window to replace. Minimum span length 4s. Maximum span length is the configured cap (120s by default) — see Span rules
Generate Audio	Checkbox	on

Edit Video Pro has no Resolution or Aspect Ratio fields — both are derived from the source video (see Providers).

Providers

Edit Video Pro is scoped to the Seedance 2 family:

Provider	Label	Resolutions
`seedance-2`	Seedance 2.0	480p / 720p / 1080p / 4K
`seedance-2-fast`	Seedance 2.0 Fast	480p / 720p only
`seedance-2-mini`	Seedance 2.0 Mini	480p / 720p only

The bridge always renders at the largest resolution tier the selected provider supports that’s still at or below the source video’s shorter dimension — there’s no manual resolution picker. A 4K source paired with seedance-2 bridges at 4K; the same source paired with seedance-2-fast or seedance-2-mini bridges at 720p, since neither offers a higher tier.

How replace mode works

Reference the cut points. If there’s at least ~2 seconds of original footage before the span, that footage is used as a “head” reference so the new content picks up naturally from where the kept footage left off. Likewise, if there’s at least ~2 seconds of footage after the span, it’s used as a “tail” reference so the new content blends back into what comes next. A span starting at 0:00 has no head reference; a span ending at (or within a fraction of a second of) the source’s own end has no tail reference.
Generate the bridge. The replacement content is generated by the selected Seedance 2 model in reference-conditioned mode, anchored by whichever of the head/tail references exist. When the requested span (plus a small per-edge stitch buffer) is longer than a single generation call can produce, it’s automatically split into multiple segments — each one continuing from the final frames of the one before it, the same chaining mechanism Generate Video Pro uses for its own long single-shot clips.
Stitch it back in. The kept footage before and after the span is cut and re-encoded, then spliced together with the freshly generated bridge into one continuous output — the same total structure as the source, with only the requested window changed.

Span rules

Minimum span: 4 seconds (spanEnd − spanStart).
Maximum span: 120 seconds by default. Self-hosted deployments can raise or lower it via the EDIT_VIDEO_PRO_MAX_SPAN environment variable — GET /v1/nodes reports the active cap for this node.
The span must fall inside the video. Edit Video Pro replaces a portion of a clip, not the whole thing — leave some original footage on at least one side. A span that covers the entire clip start-to-end gets none of the reference-bridge benefit (there’s no existing footage to anchor to on either side) and is better served by Generate Video Pro or Generate Video, which generate the whole clip directly.
A requested span that reaches past the end of the probed source video is rejected before anything is reserved or generated.

Credit pricing

credits = 10 + ceil(refRate(resolution) × (bridgeSeconds + referenceSeconds))

10 — flat fee covering the probe/bridge/stitch overhead, charged once per run.
refRate(resolution) — the per-second reference rate for the source-derived resolution tier (the same “-ref” rate Generate Video Pro’s continuation segments use). At 720p this is 6.25 credits/sec; see Generate Video Pro’s pricing section for the 480p / 1080p / 4K rates.
bridgeSeconds — the actual generated length: the replace span, plus 0.3 seconds per kept edge (head and/or tail) to cover the stitch overlap, rounded to whole seconds via the same segment-split rule Generate Video Pro uses (single generation call up to 15s; longer bridges split into multiple chained segments).
referenceSeconds — 2 seconds per bracketing reference clip actually used (head and/or tail, per How replace mode works) plus 2 seconds per internal chain join, for bridges that split into multiple segments. (Two seconds is the minimum reference length the Seedance 2 family accepts.)

The resolution tier is derived from the source video, never requested directly — the server reads the source’s actual dimensions when reserving credits (see Providers). If that read fails for any reason, the reservation worst-cases at the top tier the selected provider supports, so the reservation can only ever be too high, never too low — any difference between a worst-case reservation and the real cost is refunded once the run completes. When the read succeeds, which is the common case, the reservation already matches the final charge exactly.

Worked examples (720p, `seedance-2`)

Scenario	Span	Source length	Bridge length	Segments	Reserved credits
Minimum span, mid-video	4s	20s	5s	1	67
Mid-video span	10s	20s	11s	1	104
Span starting at 0:00	10s	20s	10s	1	85
Span ending at the video’s end	10s	20s	10s	1	85
Longer mid-video span	20s	40s	22s	2	185

The two 10-second spans on the same 20-second source land on different totals depending on position: a mid-video span keeps both a head and a tail reference (104 credits), while a span flush against either end of the source loses one of those two references and reserves less (85 credits either way — starting at 0:00 or ending at the source’s own end cost the same). The 20-second span is long enough that its bridge, plus the stitch buffer, exceeds a single generation call’s cap, so it chains across 2 segments — the extra chain-join seconds are part of why it isn’t simply double the 10-second mid-video row.

Honesty

Output frame rate is 24 fps.
Kept footage is re-encoded, not stream-copied. The untouched footage before and after the replaced span is cut with a high-quality encode (crf 18) and joined to the generated bridge with the same splice encode the platform’s other stitch operations use — the delivered file is a single consistently-encoded video, not a byte-identical copy of the original spliced with new content.
Free-tier output carries the standard watermark, the same as every other video-output node.
Any failure or cancellation is a full refund, including the flat fee. Unlike Generate Video Pro’s multi-segment runs (which can keep and bill for whatever completed before an interruption), Edit Video Pro either delivers the fully stitched result or refunds the entire reservation — there’s no partial delivery of a half-replaced span.
Runs survive worker restarts. Progress checkpoints after every bridge segment; if the processing worker restarts mid-run (for example during a platform deploy), the run resumes automatically from the checkpoint instead of failing — already-generated bridge segments are never re-generated. Only a run that stalls again after its automatic resume is failed and refunded.

Best practices

Leave at least ~2 seconds of original footage on each side of the span you want the bridge to reference — a span that starts at 0:00 or ends at the source’s own end (or leaves less than ~2 seconds on a side) loses that side’s continuity anchor.
Keep the prompt focused on what happens inside the replaced window; the kept footage on either side is untouched and doesn’t need describing.
For a fix shorter than 4 seconds, or one that only needs to touch audio or video (not both), use Video Retake instead — it supports windows as short as 2 seconds and can replace audio-only, video-only, or both, on a different provider (LTX 2.3 Pro).
For a full-clip regeneration rather than a partial replace, use Generate Video Pro (long-form) or Generate Video (single-shot).