Introduction
P-Video 2 Pro Replace from Pruna AI takes existing footage and one or two reference photos to change its cast. Start with an MP4, MOV or WebM clip lasting 1 to 120 seconds and a photo for each replacement. Obtain permission for every likeness and confirm your rights to the photos before submitting. Do not use celebrities, public figures, acquaintances or images you lack rights to. The model page asks you to accept a consent notice before your first run.
Pruna says the edit preserves camera movement, framing, background, other people, timing and motion while the new person follows the original performance. The source determines duration and aspect ratio; neither has a separate setting. In our tests, 5.04-second and 2.00-second inputs returned those same durations as MP4 files at 24 fps. The reference can change clothing and hair along with the face.
Key features & highlights
- One or two replacements: A single photo targets the main person. Two photos map from left to right unless your prompt assigns them differently.
- Source footage up to 120 seconds: Videos may arrive at up to 8K. Anything after the 120-second mark is cut.
- Five output choices: 480p and the default 768p are generated directly. Pruna upscales the 768p result for 1080p, 2K and 4K, adding processing time rather than new model detail.
- Separate sound controls:
keep_original_audioretains the soundtrack by default.ignore_source_audiois off by default so the model can use sound to guide lip movement, as Pruna describes. - Drawn reference tested: In our tests, a comic heroine’s hair, jacket and cargo trousers carried into the cycling footage.
- Measured processing examples: In our tests, a 2-second source took 47 seconds at 480p and 160 seconds at 4K. A 5-second source at 768p took 60–90 seconds. These are observations rather than completion guarantees.
Request parameters
| Parameter | Required | Description |
|---|---|---|
video_url | Yes | Required source in MP4, MOV or WebM, 1–120 seconds and up to 8K. A longer clip is cut at 120 seconds. at least 1 characters |
reference_image_urls | Yes | Required: one or two photos containing one person each. One photo targets the main person; two map from left to right unless the prompt assigns them. at least 1 items · at most 2 items |
prompt | No | Optional directions for selecting the source person, assigning each photo and naming what should remain. at most 2000 characters |
resolution | No | Choose 480p or 768p (default) for direct generation. 1080p, 2K and 4K are upscaled from the 768p result and take longer. |
keep_original_audio | No | On by default to retain the source recording. In our tests, switching it off produced a result without an audio track. Default: |
ignore_source_audio | No | Off by default, allowing the model to listen to the source. Pruna says this guides the replacement’s lips. Default: |
seed | No | Optional integer to support comparisons while changing one input. Reusing it does not guarantee that every frame will match a previous run. |
Pricing
Charges follow the source clip: round its duration up to whole seconds, with a 1-second minimum and a 120-second cap. Your resolution choice determines the rate. Photo count, prompt text and either audio switch leave the rate unchanged. Unsuccessful tasks incur no charge.
| Specification | Price |
|---|---|
| $0.06 / second |
| $0.12 / second |
| $0.144 / second |
| $0.168 / second |
| $0.216 / second |
Use cases
- Preview an approved cast change: Pair an actor’s permitted photo with an existing scene to review a different appearance against the recorded performance.
- Assign a cycling pair: Use two photos and identify each rider by shirt color, then compare the assignments while checking the street and tram against the source.
- Try a character in a dance routine: Supply one character reference for a solo dancer and review the changed hair and outfit through the studio performance.
Prompt guide & pro tips
- Keep a solo request simple: For one person on screen, leave the prompt empty; the working instruction is “replace the main person.” Our solo dance example needed no prompt.
- Map both references clearly: Identify source clothing and photo order, such as “blue-shirt rider: first photo; red-shirt rider: second photo.” This overrides automatic left-to-right matching.
- State what should remain: Add a short instruction such as “keep the other people and street unchanged.” Pruna describes the source scene as the basis for the edit.
- Choose audio behavior separately: For “keep the dialogue and follow its timing,” leave
keep_original_audioon andignore_source_audiooff. For retained sound without audio-guided lips, turnignore_source_audioon. - Compare one change per run: Use an integer seed while testing a direction such as “second photo replaces the rider in red.” A seed supports controlled comparisons without promising identical frames.
- Review before choosing an upscale: Check the default 768p result for a direction such as “preserve the studio setting.” Choose 1080p, 2K or 4K when you want an upscaled version and can allow more processing time.
API quick start
# Generate the idempotency key once; reuse it when retrying after a lost response.
export SPICY_IDEMPOTENCY_KEY="${SPICY_IDEMPOTENCY_KEY:-$(uuidgen)}"
curl -X POST https://api.spicyapi.ai/api/v1/jobs/createTask \
-H "Authorization: Bearer $SPICY_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: $SPICY_IDEMPOTENCY_KEY" \
-d '{
"model": "pruna-ai/p-video-2-pro-replace/video-edit",
"input": {
"reference_image_urls": [
"https://example.com/reference.png"
],
"resolution": "768p",
"video_url": "https://example.com/reference.mp4"
}
}'
# Copy data.taskId from the response into TASK_ID, then poll the status.
TASK_ID="TASK_ID_FROM_RESPONSE"
curl "https://api.spicyapi.ai/api/v1/jobs/recordInfo?taskId=$TASK_ID" \
-H "Authorization: Bearer $SPICY_API_KEY"
See the API reference for authentication, the complete schema, polling, and error handling.

