Skip to content
English
Esc
↑↓navigate↵open⌘Jpreview
On this page

Genjutsu Object Swap

Replace objects in a source video using reference images

Request model Endpoint
higgsfield/genjutsu-1.0-object-swap Video generation

Input modes

Mode References Reference roles
reference Reference media 2 to 11 reference_image 1 to 10
reference_video 1

Text is optional in these input modes on explicitly supported channels: reference

Prompt limit: 5,000 characters; longer prompts are rejected with 400.

Parameters

Parameter Description Values Default Required
duration Duration: Output video duration in seconds. 4 to 30 5 No
resolution Resolution: Output resolution. 480p, 720p, 1080p 720p No
aspect_ratio Aspect ratio: Output aspect ratio. auto auto No
generate_audio Generate audio: Generate native audio with the video. false false No
n Outputs per request: Number of outputs from one API request (one Job), not the application batch size. For multiple Jobs, submit separate requests; n=1 does not prevent application-managed batching. 1 1 No

Examples

reference input:

{
  "model": "higgsfield/genjutsu-1.0-object-swap",
  "input": {
    "type": "reference",
    "prompt": "A paper boat drifting down a rainy city street, slow tracking shot",
    "references": [
      {
        "role": "reference_image",
        "file_id": "file_..."
      },
      {
        "role": "reference_video",
        "file_id": "file_..."
      }
    ]
  },
  "parameters": {
    "duration": 5,
    "resolution": "720p",
    "aspect_ratio": "auto",
    "generate_audio": false,
    "n": 1
  }
}

The authoritative contract is GET /v1/models/higgsfield/genjutsu-1.0-object-swap?view=contract. A valid parameter does not mean every channel can run it, so quote the complete request. Prices and channels are listed in Model Market.

Source video and billing

Use input.type: "reference" with one 4–30 second MP4 as reference_video and 1–10 reference_image inputs. In your prompt, use @video_1 and @image_1 to identify the source and replacement.

To have the person in the reference photos perform the source video’s movements, use Genjutsu Motion Transfer. See the Genjutsu request examples.

Yir uses NanoAPI for this model. The output length follows the source clip. The request’s duration does not trim the clip, and the model has no audio switch. For a numeric quote, upload the source through Files. Estimates round up the longer measured audio or video track to whole seconds. Final billing uses the channel’s actual task charge.

Compatible request names

Submit and quote requests also accept genjutsu-1.0-object-swap.