V2V
Lenscowboy · Documentation
Video to Video — vendor reference
2D tab operation
← All docs
Video to Video (V2V) is one of the three-tier operations in the 2D tab. It takes a source video plus optional reference media and produces a new video that follows the source's motion, the references' look, or both — depending on the vendor you pick. This page covers the four V2V vendors wired into the cascade, what each is good at, and the reference and resolution caps you'll bump into.

Overview

Video to Video is the 2D-tab operation for transforming an existing video clip into a new one — restyle, recompose against references, change the look, retime motion, or build a composite. Unlike Image to Video (which generates motion from a still) or Text to Video (which generates from prompt alone), V2V always anchors on a real source clip you supply.

Each vendor handles V2V differently — some preserve motion strictly, some treat the source as a composition anchor, some accept a single reference and others accept several. The cascade exposes all four so you can pick the right one for the shot.

The cascade

Video to Video uses the four-step Operation → Platform → Vendor → Model cascade. You pick the operation (Video to Video), then narrow by platform, then vendor, then the specific model variant. The dropdowns filter each other so you only see valid combinations. 

Operation:  Video to Video
Platform:   fal | piapi | runway
Vendor:     Bytedance | Lightricks | Runway
Model:      <variants per vendor>

The same cascade shape is used for the adjacent Motion Control operation and the new Text to Video operation — once you learn it for V2V, you can move between them without re-learning the UI.

Vendors at a glance

Four V2V routes are wired today. The summary table covers reference caps, resolution support, and output-duration range — the things that most often decide which one fits a given shot.

Vendor Style Video refs Image refs Audio refs Resolution
fal · Seedance 2.0
Bytedance, standard + fast		 Motion-preserving V2V up to 3
(1 source + 2 extra)		 up to 9 up to 3 480p · 720p
fal · LTX 2.3 Ref
Lightricks		 Single-source restyle 1
(source only)		 — — LTX native
piapi · Seedance Omni Reference
Bytedance, composition mode		 Multi-reference composition up to 3
(1 source + 2 extra)		 up to 9 up to 3 480p · 720p · 1080p
runway · Gen-4 Aleph
Runway		 Runway V2V 1
(source)		 — — Runway native
Output duration

All four vendors accept a duration in the 4–15 second range, set via the 2D tab's Duration column. Each vendor has its own per-clip cap on reference material that's enforced automatically — very long reference clips are trimmed before upload so you never need to pre-trim them yourself.

fal · Seedance 2.0 (and Fast)

Motion-preserving V2V

Seedance 2.0 V2V on fal is a true motion-preserving video-to-video model — the source clip's motion is the dominant anchor, and the prompt + extra references shape the look. Two variants: Standard for quality, Fast for cost and turnaround.

Accepts 1 source video + up to 2 extra reference videos, up to 9 reference images, and up to 3 reference audio refs. Caps out at 720p — if you need 1080p, see Resolution & upscaling.

Pick this when: you want the source clip's motion to come through faithfully and only the look to change. Examples: rotoscope-style stylisation, plate-to-plate restyle, on-model character reskin.

fal · LTX 2.3 Ref V2V

Single-source restyle

Lightricks' LTX 2.3 Reference V2V is a single-input model — one source video, no separate image or audio reference slots. It exposes a video_strength knob (via the Extra Params column) so you can dial how strongly the source's motion and composition carry through versus the prompt.

Pick this when: the shot is simple, you don't have extra reference media, and you want LTX's distinctive aesthetic. It's also fast and cheap, which makes it a good first pass before committing to a heavier vendor.

piapi · Seedance 2.0 Omni Reference

Multi-reference composition

Same underlying model family as fal's Seedance, but served through piapi in composition mode. Up to 3 videos (1 source + 2 extra), up to 9 images, up to 3 audio refs. Native 1080p support — the one capability where piapi beats fal's Seedance route.

Composition mode is not the same as motion-preserving V2V. Seedance Omni Reference treats every reference — videos included — as compositional input, with the prompt deciding what to keep, what to mix, and what to anchor. If you give it a plate and a character, expect the model to compose them rather than to track the plate's motion strictly.

References are addressable in the prompt as @video1, @video2, ..., @image1, @image2, ... — this is how you direct which reference plays which role. If you don't tag, the system adds a default anchor directive so all references contribute.

Pick this when: you want to combine multiple references into one shot, you need 1080p direct, or you want the most expressive prompt-driven control over which reference becomes the subject vs the environment.

runway · Gen-4 Aleph

Runway V2V

Runway's Gen-4 Aleph V2V model — single source video, no separate reference slots. Strong for cinematic restyle and stylised motion. Resolution and duration follow Runway's published Aleph spec.

Pick this when: you're already in a Runway-driven workflow, you want Aleph's particular look, or you're chaining V2V with other Runway models (Gen-4, Gen-4 Turbo) in the same shot.

How to choose

A few rules of thumb:

Reference type matters

When you drop reference files into the 2D tab's reference column, files are sorted by extension — video files (.mp4, .mov, .webm) become video references, image files (.png, .jpg, .webp, ...) become image references, audio files (.mp3, .wav, .m4a, ...) become audio references. You don't need to put them in separate columns.

Resolution & upscaling

Only piapi · Seedance Omni Reference outputs 1080p directly today. For the other vendors, the canonical pattern is to chain a Video Upscale pass after the V2V row:

  1. Row N: Video to Video at the vendor's native resolution (e.g. 720p with fal Seedance).
  2. Row N+1: Video Upscale → Topaz, picking a model appropriate to the source motion (Proteus, Artemis, Nyx, Gaia, or Starlight). The 2D pipeline's job-chaining picks up Row N's output as Row N+1's input automatically.

Topaz upscale runs as a separate operation in the same 2D tab and is documented under the Pipeline guide.