Video to Video
Transform existing video using AI with a text prompt.
Overview
The Video to Video node applies AI-powered transformations to an existing video based on a text prompt. Use it for style transfer, content modification, or creative re-interpretation of video footage.
Configuration
| Field | Type | Default | Description |
|---|---|---|---|
| Provider | Select | wan | AI model for transformation |
| Prompt | Textarea | — | Description of the desired transformation |
promptPrefix / promptSuffix |
text | – | Optional pre/post text wrapped around the prompt at run time (settings panel → Pre & post text; hidden from app users; captured by presets). See Prompt pre & post text. |
Seedance 2.5 Edit — additional fields
| Field | Type | Default | Description |
|---|---|---|---|
Resolution (v2vResolution) |
Select | 480p | 480p / 720p / 1080p output |
Generate Audio (generateAudio) |
Checkbox | on | Native audio on the generated clip |
Seed (seed) |
Number | — | Seed for reproducible results |
Seedance 2.5 deliberately exposes no Duration, Aspect Ratio, Audio Setting, Multi-shot, Expand-prompt or Negative Prompt field on this node — see Seedance 2.5 Edit below for why.
Wan 2.7 VideoEdit — additional fields
| Field | Type | Default | Description |
|---|---|---|---|
| Duration | Select | Auto | Auto / 5s / 10s — target output length |
| Resolution | Select | 1080p | 720p or 1080p output |
| Audio | Select | Auto | Auto = AI-generated audio; Origin = keep source audio |
| Expand prompt | Checkbox | off | AI enhances the prompt before sending to KIE |
| Negative Prompt | Textarea | — | What to avoid in the output |
| Seed | Number | — | Seed for reproducible results |
Inputs & Outputs
Inputs:
- Video (required) — source video to transform
- Cinematography (optional) — parameter pickers whose hints join the prompt
- Text Prompt (optional) — from upstream node
- Negative (optional) — text-only; ignored on the Seedance 2.5 Edit lane
- Image Refs (optional) — Seedance 2.5 Edit only; up to 30 reference images
- Audio Refs (optional) — Seedance 2.5 Edit only; up to 10 reference audio clips
The two reference handles are always visible but dimmed on every other model — those endpoints take at most one reference image and no reference audio, so the handle reports “not supported by [Model]” and wiring one has no effect.
Outputs:
- Transformed video URL
Seedance 2.5 Edit
Seedance has no video-to-video endpoint. It edits a video it is given as a reference when the prompt reads as an edit instruction. So this model’s lane on the node is dispatched as a text-to-video job in edit shape:
- The source clip is sent as reference video 1.
- Your prompt is prefixed with
edit @video_1 as follows:and a newline, so the model reads the whole sentence as an instruction about that clip. If your prompt already opens with that instruction, it is not added twice. - Aspect ratio is
adaptiveand duration isAuto— the result keeps the source clip’s own length and aspect ratio. That is why the node offers no Duration and no Aspect Ratio field for this model.
The source clip must be 4–30 seconds. Shorter or longer clips are rejected by the provider.
Reference images and {image:N}. Wire images into Image Refs and cite
them positionally in the prompt — {image:1}, {image:2}, … in the order the
edges were made. The editor resolves them to @image_1, @image_2 on the wire,
the same binding Generate Video uses (see
Generate Video → Referencing wired assets in the prompt).
Reference audio works the same way with {audio:N}.
make it black and white, heavy film grain
the woman wears {image:1} and stands in {image:2}
Negative prompts are not sent on this lane — the edit dispatch carries no
negative field, and Seedance has no native negative_prompt. Put what you want
avoided into the instruction itself.
Supported Providers
| Provider | Credits | Notes |
|---|---|---|
Seedance 2.5 Edit (seedance-2-5) |
dynamic — see below | Whole-clip prompt-driven edit; keeps source length + ratio; 30 image / 10 audio references; native audio |
Wan 2.6 (wan) |
18 cr | General-purpose V2V, reliable results |
Wan 2.6 Flash (wan-flash) |
13 cr | Fast V2V with optional audio & multi-shot |
Wan 2.7 VideoEdit (wan-videoedit) |
320 cr | Guided editing with reference image, audio control, prompt extend |
Luma Modify (luma-modify) |
320 cr | Strong at style transfer and artistic modifications |
Runway Aleph (runway-aleph) |
350 cr | High-quality transformations, flexible aspect ratio |
HappyHorse Edit (happyhorse-edit) |
350 cr | Up to 60s input, 720p (default) or 1080p output |
Seedance 2.5 Edit — credits
Because the lane is a Seedance reference-video run, it is billed exactly like
one on Generate Video: ceil(perSec × (clip_seconds + delivered_seconds)) on
the -ref ladder for the chosen resolution.
Duration is Auto, so the run is reserved at the model’s longest clip (30 s)
and refunded down to the length actually delivered — which, for an edit, is
the source clip’s own length. Your balance must cover the reservation for the
run to start.
The formulas, the worked examples and what happens when a clip cannot be measured are all one place — see Generate Video → Auto duration and Reference videos bill input + output duration.
Best Practices
- Keep source video short (5-10s) for best results
- Describe the transformation clearly: “change to anime style” rather than describing the entire scene
- For Seedance 2.5 Edit: write the prompt as an instruction about the clip (“make it black and white”, “the woman wears {image:1}”) — it is read as one. Start at 480p while iterating; the reservation scales with resolution
- For Wan 2.7 VideoEdit: connect a reference image to guide the visual style of the output
- Use Expand prompt when your prompt is very brief — the AI fills in cinematic details
- Test with the cheapest provider (Wan Flash) before upgrading to Runway or Luma
Common Use Cases
- Apply artistic styles to recorded footage (painterly, anime, cinematic)
- Change scene characteristics (day to night, summer to winter)
- Create variations of existing video content
- Transform live-action footage into animated styles
Tips
- V2V works best when the transformation is stylistic rather than structural — changing colors and textures works better than adding or removing objects
- For structural changes, consider using Image-to-Image on keyframes then Image-to-Video instead
- Use consistent prompts when processing multiple clips for a project