P-Video-2-Pro-Reference

P-Video-2-Pro-Reference, based on MiniMax H3, is the reference-conditioned counterpart to P-Video-2-Pro. Generate from a prompt plus at least one image or video reference. Audio, first-frame, and last-frame stills are optional.

It offers reference-conditioned generation at 480p and 768p, 24 fps, with duration 5–15 seconds (default 5). Output includes generated audio; a prompt alone is rejected. Use mode: cost for the lowest list price (same output quality as speed), or mode: speed (default) for the faster recipe. There is no quality mode. prompt_upsampler is independent: off, turbo (default), or max; it does not change the price.

Note

When using P-Video-2-Pro-Reference, make sure to respect the copyright of the images, videos, and audio you use as input and the content you generate.

Pricing: billed from the finished run. Output, reference video, images, and audio each have a rate for mode; output and reference video also depend on resolution. prompt_upsampler and keep_audio do not change the price.

Output
Resolution	`mode: cost`	`mode: speed`
480p	$0.015 per second of output video	$0.02 per second of output video
768p	$0.025 per second of output video	$0.035 per second of output video

Reference video
Resolution	`mode: cost`	`mode: speed`
480p	$0.03 per second of reference video	$0.04 per second of reference video
768p	$0.06 per second of reference video	$0.06 per second of reference video

Reference images
`mode: cost`	`mode: speed`
$0.005 each	$0.01 each

Reference audio
`mode: cost`	`mode: speed`
$0.0015 per second	$0.0025 per second

The per-second output rates above are with no references. The cheapest request is $0.075 (cost, 480p, no references).

Speed: model execution time per second of output video.

Resolution	`mode: speed`	`mode: cost`
480p	~0.38 s per second of output video	~1.04 s per second of output video
768p	~0.80 s per second of output video	~2.12 s per second of output video

Tip

Log in to the API in the Pruna User Portal.

Prompt formula

Fast pass

Prompt plus one still or clip. Enough for first looks. Iterate in cost or speed mode.

[prompt] [image]

subject, action, scene, one image or video reference

Locked-in

Add camera, lighting, and style. Stack stills, clips, audio, and optional start / end frames.

[prompt] [image] [video] [audio]

subject, action, scene + camera, lighting, style, stills, clips, soft keyframes, optional reference audio or kept soundtrack

Fast pass: one image reference plus a short motion prompt.

Locked-in: an image reference for the character plus a video reference for the motion.

prompt: required. Name the subject, action, and scene, and say what to keep from the references. Add camera, lighting, and style when you need a repeatable final.
- Fast pass: “The person in the reference walks through a sunlit market.”
- Locked-in: “The reference image is the hero product. Keep that exact matte stainless kettle, gooseneck spout, and black handle. Photoreal product film, 8 seconds. Slow orbit on a light-gray sweep, a thin ribbon of steam. No logos, no captions. Audio: quiet studio, no voiceover.”
image / video: send at least one. Images are jpg, jpeg, png, or webp. Up to 9 images including start and end frames; up to 3 videos.
- Fast pass: one well-lit still.
- Locked-in: several stills or a short clip; generate locked compositions with P-Image-Ideogram when you need them.
audio / first_frame_image / last_frame_image: optional. Audio is a reference unless keep_audio is set. Start and end frames guide the open and close; they are not hard constraints.
- Fast pass: omit; write sound in the prompt.
- Locked-in: upload a track when timing must match, or pair start and end stills.

Slot	Fast pass (enough to run)	Locked-in (stronger control)
`prompt`	Subject, action, scene, and what to keep from the reference.	Subject + action + scene + camera, lighting, style; same wording every rerun.
`image` / `video`	One still.	Several stills or a short clip for identity, product, or motion.
`audio` / `first_frame_image` / `last_frame_image`	Empty; name sound in the prompt.	Uploaded track or opening and closing stills.

Tip

For comprehensive video prompting (motion, framing, atmosphere), see the Video Generation guide.

Choosing the right video model

Pruna ships performance video models that share the same prediction API, but each solves a different production problem. P-Video-2-Pro-Reference generates new footage from a prompt plus references (Model: p-video-2-pro-reference).

	P-Video-2-Pro-Reference	P-Video-2-Pro	P-Video-2	P-Video	P-Video-Avatar	P-Video-Animate	P-Video-Replace	P-Video-Edit
One-line job	Generate cinematic footage from prompts + references	Generate cinematic footage from prompts	Generate premium footage from prompts	Generate fast / affordable footage	Speak from one still (script or audio)	Retarget one still with clip motion	Swap characters in existing footage	Rewrite content in existing footage
You start with	Prompt + at least one image or video (+ optional audio, start / end frames)	Text prompt (+ optional first / last frame)	Text prompt (+ optional image / audio)	Text prompt (+ optional image / audio)	Portrait still + `voice_script` or `audio`	Source video + one still	Source video + identity stills	Source video + text prompt
You keep from the source	Identity, motion, and canvas from the uploaded references	N/A (new scene); canvas follows reference images when set	N/A (new scene); strong I2V consistency when an image is set	N/A (new scene)	Aspect ratio of the still	Motion, timing, camera from the driver	Camera, timing, blocking, background	Camera, timing, subject performance
Typical ask	“Continue this clip / match these stills for 8 s at 768p.”	“Make an 8 s cinematic product or documentary clip with generated audio.”	“Make a 10 s product ad with music and lip-sync.”	“Draft a 5 s social clip cheaply and fast.”	“This spokesperson says this line in French.”	“Animate this catalog still using our winning ad take.”	“Put our creator in this UGC b-roll.”	“Turn this silver SUV red and keep the camera move.”
Generation time (5 s video)	4.0 s (768p `speed`, 1 image) · 10.6 s (768p `cost`, 1 image)	4.3 s (768p `speed`) · 9.1 s (768p `quality`)	4.5 s (draft: 2.0 s)	9.3 s	—	26.2 s (720p)	~17.9 s (720p)	—
Cost (5 s video)	$0.130 (768p `cost` + 1 image) · $0.185 (768p `speed` + 1 image)	$0.125 (768p `cost`) · $0.175 (768p `speed`) · $0.375 (768p `quality`)	$0.13 (draft: $0.08)	$0.10 (draft: $0.03)	$0.125 (720p) · $0.225 (1080p)	$0.15 (720p) · $0.30 (1080p)	$0.15 (720p) · $0.30 (1080p)	$0.225 (draft: $0.125)
Draft mode	No	No	Yes	Yes	No	No	No	Yes

Quick decision guide

Need multiple image, video, or audio references, or soft start / end frames on MiniMax H3 → P-Video-2-Pro-Reference. Use mode: cost when price is the priority (same output quality as speed).
Need cinematic generation from text or a single first / last frame, including ``quality`` mode → P-Video-2-Pro.
Need 1080p, imported audio muxed onto a new clip, or draft previews → P-Video-2.
Need the fastest / cheapest generation iteration → P-Video.
Single-speaker talking head, voice only, no music bed → P-Video-Avatar.
Footage exists and the hero still should move like the driver → P-Video-Animate.
Footage exists and you need different people in the same shot → P-Video-Replace.
Footage exists and you need a color, product, environment, object, or text edit → P-Video-Edit.

Key features

Marketing and ads: Animate brand stills, continue a winning take, or match a bed you already have. Prefer P-Video-Avatar only for voice-only talking heads.
Media and entertainment: Short-form narrative and two-shot dialogue that stay locked to uploaded faces, blocking, or audio.
Retail and e-commerce: Product loops from packshot stills. Keep motion subtle and the product readable.
Corporate and education: Training, explainers, and comms that must start and end on specific frames. Voice-only single-presenter scripts → P-Video-Avatar.
Gaming: Trailers, world reveals, and character moments that keep identity from concept stills or previous shots.
Sports: Documentary and broadcast-style coverage that continues a sideline clip. Not a fashion ad.
Generated audio: Output includes generated audio. Write sound or dialogue in the prompt, or set keep_audio with one uploaded file to keep that track.
References + soft keyframes: Animate from up to 9 images, 3 videos, and 3 audio files. Optional first_frame_image and last_frame_image guide the open and close; they are not hard constraints.
Cost and speed: mode: cost is the lowest-price recipe and matches speed output quality. mode: speed is the default faster recipe. There is no quality mode. prompt_upsampler (off / turbo / max) is independent of mode and does not change the price.

Practical constraints

Send at least one image or video with the prompt. A prompt alone is rejected.
Duration is 5–15 seconds (default 5).
Audio is optional. Set keep_audio only with exactly one audio file.
Input images: jpg, jpeg, png, webp.
Resolution is 480p or 768p. Output is 24 fps.
There is no draft mode and no quality mode. mode: cost is the lowest-price recipe (same output quality as speed); mode: speed is the default faster recipe.
This model generates new footage. Use P-Video-Edit to modify existing clips.
Rate limit: 250 requests per minute.

Examples

Reference · image Reference image

Output

Metro poster · image reference · 16:9 · 768p · 12s · mode: speed · prompt_upsampler: max

Example prompt

The reference image is the hero frame. Keep that exact Paris metro wall, that metal frame, and that cream poster. The poster text stays exactly DO YOU KNOW THAT THIS COULD BECOME REAL? with the small word Happn under it. The letters never change. The camera stays wide and still. Photoreal, 12 seconds, one continuous shot. The platform is busy. Many commuters walk through the frame the whole time, some left to right, some right to left, just passing, looking ahead, not at the poster. Among them, two people walk faster than everyone else, toward each other. Their chins stay down. Each is focused on a phone held in one hand. They do not look up during the walk. From the left, a woman with a short black bob and a rust wool coat. From the right, a man with short dark hair and an olive coat. Their mouths stay closed. Each visible hand has five fingers. Because they are on their phones, they walk straight into each other near the center, early in the shot. At the moment they collide, both look up at the poster. The poster stays on the wall, unchanged. No extra words, no captions, no other logos.

The reference image is the hero frame. Keep that exact Paris metro wall, that metal frame, and that cream poster. The poster text stays exactly DO YOU KNOW THAT THIS COULD BECOME REAL? with the small word Happn under it. The letters never change. The camera stays wide and still. Photoreal, 12 seconds, one continuous shot. The platform is busy. Many commuters walk through the frame the whole time, some left to right, some right to left, just passing, looking ahead, not at the poster. Among them, two people walk faster than everyone else, toward each other. Their chins stay down. Each is focused on a phone held in one hand. They do not look up during the walk. From the left, a woman with a short black bob and a rust wool coat. From the right, a man with short dark hair and an olive coat. Their mouths stay closed. Each visible hand has five fingers. Because they are on their phones, they walk straight into each other near the center, early in the shot. At the moment they collide, both look up at the poster. The poster stays on the wall, unchanged. No extra words, no captions, no other logos.

Reference · video

Output

One minute · video reference · 16:9 · 768p · 2 × 10s · mode: speed · prompt_upsampler: off

Example prompt

Shot 1 — the phone Use the reference video only to keep that man and that woman, in those same coats: he wears the dark herringbone overcoat, she wears the beige trench, her hair in a bun. Ignore the rooftop, the rail, and the wine. Faces and mouths stay out of frame the whole time. We see only their backs. New scene. Contemporary mini-drama, 10 seconds, one continuous shot, no cuts. A wet city street at night. The camera follows from behind at shoulder height. For the first two seconds they walk, her a step ahead. Then she raises a dark phone in her right hand, screen toward him and away from the camera, and holds it there. Then he reaches with his right hand, one thumb and four fingers, and takes the phone from her. He keeps the phone. They keep walking away. The phone screen shows no readable text. Audio: rain and footsteps only. No music. No spoken words. No voices. No captions. Shot 2 — the door Use the reference video only to keep that man and that woman, in those same coats: he wears the dark herringbone overcoat, she wears the beige trench, her hair in a bun. Ignore the rooftop, the rail, and the wine. Faces and mouths stay out of frame the whole time. We see only their backs. New scene, the next moment. Contemporary mini-drama, 10 seconds, one continuous shot, no cuts. They arrive at a dark front door on a quiet night street. The camera stays behind them the whole time. They stop at the door, backs to us. She is closest to the door. After a few seconds she steps inside. The door closes. He stays outside, back to us, holding a dark phone at his side. Audio: light rain, then quiet rain. No music. No spoken words. No voices. No captions.

Shot 1 — the phone
Use the reference video only to keep that man and that woman, in those same coats: he wears the dark herringbone overcoat, she wears the beige trench, her hair in a bun. Ignore the rooftop, the rail, and the wine. Faces and mouths stay out of frame the whole time. We see only their backs. New scene. Contemporary mini-drama, 10 seconds, one continuous shot, no cuts. A wet city street at night. The camera follows from behind at shoulder height. For the first two seconds they walk, her a step ahead. Then she raises a dark phone in her right hand, screen toward him and away from the camera, and holds it there. Then he reaches with his right hand, one thumb and four fingers, and takes the phone from her. He keeps the phone. They keep walking away. The phone screen shows no readable text. Audio: rain and footsteps only. No music. No spoken words. No voices. No captions.

Shot 2 — the door
Use the reference video only to keep that man and that woman, in those same coats: he wears the dark herringbone overcoat, she wears the beige trench, her hair in a bun. Ignore the rooftop, the rail, and the wine. Faces and mouths stay out of frame the whole time. We see only their backs. New scene, the next moment. Contemporary mini-drama, 10 seconds, one continuous shot, no cuts. They arrive at a dark front door on a quiet night street. The camera stays behind them the whole time. They stop at the door, backs to us. She is closest to the door. After a few seconds she steps inside. The door closes. He stays outside, back to us, holding a dark phone at his side. Audio: light rain, then quiet rain. No music. No spoken words. No voices. No captions.

Reference · image Reference image

Output

Stainless kettle catalog · image reference · 16:9 · 768p · 2 × 8s · mode: speed · prompt_upsampler: turbo

Example prompt

Shot 1 — packshot The reference image is the hero kettle. Keep that exact kettle with the lid fully closed. Photoreal product film, 8 seconds. Slow orbit on the light-gray sweep. The lid stays shut. No steam, no pour, no hand. No captions. Shot 2 — kitchen serve The reference image is the hero kettle. Keep that exact matte stainless gooseneck kettle, lid fully closed, black lid knob, black handle. Start in the kitchen: the cream ceramic cup is already on the oak counter and stays ceramic. Morning window light. A right hand enters from the right, grasps the black handle, lifts this same kettle, and pours a thin steady stream from the tip of the spout into that cup. The stream is pale amber tea, and the liquid in the cup is that same pale amber tea, from the first drop to the end. The cup stays on the counter the whole time. The lid never opens. The face stays out of frame. Exactly one thumb and four fingers. No glass carafe. No second vessel. No logos, no text.

Shot 1 — packshot
The reference image is the hero kettle. Keep that exact kettle with the lid fully closed. Photoreal product film, 8 seconds. Slow orbit on the light-gray sweep. The lid stays shut. No steam, no pour, no hand. No captions.

Shot 2 — kitchen serve
The reference image is the hero kettle. Keep that exact matte stainless gooseneck kettle, lid fully closed, black lid knob, black handle. Start in the kitchen: the cream ceramic cup is already on the oak counter and stays ceramic. Morning window light. A right hand enters from the right, grasps the black handle, lifts this same kettle, and pours a thin steady stream from the tip of the spout into that cup. The stream is pale amber tea, and the liquid in the cup is that same pale amber tea, from the first drop to the end. The cup stays on the counter the whole time. The lid never opens. The face stays out of frame. Exactly one thumb and four fingers. No glass carafe. No second vessel. No logos, no text.

Reference · image Reference image

Output

Faceted perfume bottle, then a spray on the nape · image reference · 16:9 · 768p · 2 × 8s · mode: speed · prompt_upsampler: turbo / max

Example prompt

Shot 1 — bottle hero The reference image is the hero bottle. Keep that exact faceted glass bottle, silver collar in place, pale gold liquid at the same level. Photoreal beauty product film, 8 seconds. Slow orbit on white marble. The collar stays on. No spray, no mist, no hand, no drip. No people, no captions. Shot 2 — spray on the nape The reference image is the hero bottle. Keep that exact faceted glass perfume bottle, silver atomizer, pale gold liquid at the same level. Photoreal beauty film, 8 seconds, one continuous shot. Open already on a woman seen from behind, hair in a bun, bare nape, in a quiet beige bathroom. She holds that bottle in both hands at the nape of her neck. She presses the atomizer. A bright, clearly visible cloud of fine mist leaves the nozzle and lands on the skin of her nape. The mist hangs in the air for a moment, then fades. We never see her face. No logos, no writing, no captions, no second person.

Shot 1 — bottle hero
The reference image is the hero bottle. Keep that exact faceted glass bottle, silver collar in place, pale gold liquid at the same level. Photoreal beauty product film, 8 seconds. Slow orbit on white marble. The collar stays on. No spray, no mist, no hand, no drip. No people, no captions.

Shot 2 — spray on the nape
The reference image is the hero bottle. Keep that exact faceted glass perfume bottle, silver atomizer, pale gold liquid at the same level. Photoreal beauty film, 8 seconds, one continuous shot. Open already on a woman seen from behind, hair in a bun, bare nape, in a quiet beige bathroom. She holds that bottle in both hands at the nape of her neck. She presses the atomizer. A bright, clearly visible cloud of fine mist leaves the nozzle and lands on the skin of her nape. The mist hangs in the air for a moment, then fades. We never see her face. No logos, no writing, no captions, no second person.

Reference · image Reference image

Output

Camel wool overcoat · image reference · 16:9 · 768p · 8s + 12s · mode: speed · prompt_upsampler: max

Example prompt

Shot 1 — off the rail Match the reference image exactly. The coat is a long straight camel wool overcoat on a light wooden hanger: small flat collar, exactly four black buttons in one vertical line from the chest to the hem, two narrow vertical slit pockets, long straight sleeves, hem at mid-calf. Keep that same coat, that length, those four buttons, and those slit pockets in every frame. Photoreal fashion film, 8 seconds, one continuous shot. A woman with dark shoulder-length hair, a cream knit, and black trousers stands at a pale oak rail. Only one hanger is on the rail, and it holds that exact coat, fully buttoned. She lifts that one coat off the hanger with both hands and holds it up, still fully buttoned. Each hand has five fingers. Her mouth stays closed. No tags, no writing, no logos. Shot 2 — on, buttoned, walking The first image is the hero coat. The second image is that same coat held up. Copy that garment exactly: long straight camel wool overcoat, small flat collar, exactly four black buttons in one vertical line, two narrow vertical slit pockets, hem at mid-calf. Photoreal fashion film, 12 seconds, one continuous shot, camera stays wide so the whole coat from collar to hem stays in frame. The same woman, dark shoulder-length hair, cream knit, black trousers, bright pale hallway. She puts both arms into that coat, fastens the four black buttons from the top down, then walks toward the camera for several steps. The coat keeps four buttons, the slit pockets, and the mid-calf hem while she walks. Each hand has five fingers. Her mouth stays closed. No tags, no writing, no logos.

Shot 1 — off the rail
Match the reference image exactly. The coat is a long straight camel wool overcoat on a light wooden hanger: small flat collar, exactly four black buttons in one vertical line from the chest to the hem, two narrow vertical slit pockets, long straight sleeves, hem at mid-calf. Keep that same coat, that length, those four buttons, and those slit pockets in every frame. Photoreal fashion film, 8 seconds, one continuous shot. A woman with dark shoulder-length hair, a cream knit, and black trousers stands at a pale oak rail. Only one hanger is on the rail, and it holds that exact coat, fully buttoned. She lifts that one coat off the hanger with both hands and holds it up, still fully buttoned. Each hand has five fingers. Her mouth stays closed. No tags, no writing, no logos.

Shot 2 — on, buttoned, walking
The first image is the hero coat. The second image is that same coat held up. Copy that garment exactly: long straight camel wool overcoat, small flat collar, exactly four black buttons in one vertical line, two narrow vertical slit pockets, hem at mid-calf. Photoreal fashion film, 12 seconds, one continuous shot, camera stays wide so the whole coat from collar to hem stays in frame. The same woman, dark shoulder-length hair, cream knit, black trousers, bright pale hallway. She puts both arms into that coat, fastens the four black buttons from the top down, then walks toward the camera for several steps. The coat keeps four buttons, the slit pockets, and the mid-calf hem while she walks. Each hand has five fingers. Her mouth stays closed. No tags, no writing, no logos.

Reference · image Reference image

Reference · audio

Output

Café opening · image + audio references · 16:9 · 768p · 12s · mode: speed · prompt_upsampler: max

Example prompt

The reference image is the hero cafe. Keep that exact cream storefront, wide window, oak tables, two black pendant lamps, and blank panel. The customer entrance is the glass door on the right, the door that shows the kitchen. Its handle sits on the left edge of that door, so the hinges are on the right and the door swings inward, into the shop. The tall dark door on the left stays closed the whole time. Photoreal commercial, 12 seconds, one continuous shot. A person seen only from behind, dark coat, short dark hair, walks in from the street straight to that right door. They reach it by the fourth second. They push the handle with one hand and the right door swings inward. They step through that right doorway and go inside the cafe. We keep seeing their back. Each hand has five fingers. No face. The left door never moves. No letters, no logos, no captions.

The reference image is the hero cafe. Keep that exact cream storefront, wide window, oak tables, two black pendant lamps, and blank panel. The customer entrance is the glass door on the right, the door that shows the kitchen. Its handle sits on the left edge of that door, so the hinges are on the right and the door swings inward, into the shop. The tall dark door on the left stays closed the whole time. Photoreal commercial, 12 seconds, one continuous shot. A person seen only from behind, dark coat, short dark hair, walks in from the street straight to that right door. They reach it by the fourth second. They push the handle with one hand and the right door swings inward. They step through that right doorway and go inside the cafe. We keep seeing their back. Each hand has five fingers. No face. The left door never moves. No letters, no logos, no captions.

Reference · image Reference image

Output

Teal bouclé sofa · image reference · 16:9 · 768p · 2 × 8s · mode: speed · prompt_upsampler: turbo

Example prompt

Shot 1 — studio catalog The reference image is the hero sofa. Keep that exact empty teal boucle sofa, oak legs, and even unpressed cushions. Photoreal furniture catalog film, 8 seconds. Slow orbit at seat height on a warm-white sweep. Nobody sits. No hands. Cushions stay even. No captions. Shot 2 — in-room The reference image is the hero sofa. Keep that exact empty sofa. Photoreal furniture film, 8 seconds. The same sofa is in a sunlit living room with a pale oak floor and a linen curtain. Slow dolly. The sofa stays empty, cushions unpressed, no people, no objects added. No captions.

Shot 1 — studio catalog
The reference image is the hero sofa. Keep that exact empty teal boucle sofa, oak legs, and even unpressed cushions. Photoreal furniture catalog film, 8 seconds. Slow orbit at seat height on a warm-white sweep. Nobody sits. No hands. Cushions stay even. No captions.

Shot 2 — in-room
The reference image is the hero sofa. Keep that exact empty sofa. Photoreal furniture film, 8 seconds. The same sofa is in a sunlit living room with a pale oak floor and a linen curtain. Slow dolly. The sofa stays empty, cushions unpressed, no people, no objects added. No captions.

Reference · video

Output

Deep-blue coupe · video reference · 16:9 · 768p · 2 × 8s · mode: speed · prompt_upsampler: off

Example prompt

Shot 1 — coastal road Use the reference video only to keep the same unmarked deep-blue coupe. Ignore the reference location, camera, and action. New scene, not a continuation. Photoreal automotive film, 8 seconds. That coupe drives on a wet coastal road at dusk, ocean and cliffs beside it. No salt flat, no empty horizon, no parked car. Both doors stay closed. Headlights are already on and stay on. Tracking camera at wheel height. No people, no captions. Shot 2 — night street Use the reference video only to keep the same deep-blue coupe. Ignore the reference location and camera. New scene. Photoreal film, 8 seconds. The coupe is parked on a rain-soaked city street at night. Neon is abstract color only, with no letters and no shop names. Wet cobblestones, not a salt flat. Doors stay closed. Headlights are already on and stay on. Slow three-quarter push. No plate, no people, no captions.

Shot 1 — coastal road
Use the reference video only to keep the same unmarked deep-blue coupe. Ignore the reference location, camera, and action. New scene, not a continuation. Photoreal automotive film, 8 seconds. That coupe drives on a wet coastal road at dusk, ocean and cliffs beside it. No salt flat, no empty horizon, no parked car. Both doors stay closed. Headlights are already on and stay on. Tracking camera at wheel height. No people, no captions.

Shot 2 — night street
Use the reference video only to keep the same deep-blue coupe. Ignore the reference location and camera. New scene. Photoreal film, 8 seconds. The coupe is parked on a rain-soaked city street at night. Neon is abstract color only, with no letters and no shop names. Wet cobblestones, not a salt flat. Doors stay closed. Headlights are already on and stay on. Slow three-quarter push. No plate, no people, no captions.

Integration

P-Video-2-Pro-Reference uses the same Pruna predictions API as other performance models. Upload each file first, then pass the returned URLs. Send at least one image or video reference with the prompt.

Tip

For more information on how to use the API, see the API Reference.

API endpoint: Base URL: https://api.pruna.ai/v1/predictions

Authentication

-H 'apikey: YOUR_API_KEY'
-H 'Model: p-video-2-pro-reference'

Image reference (asynchronous)

curl -X POST 'https://api.pruna.ai/v1/predictions' \
  -H 'Content-Type: application/json' \
  -H 'apikey: YOUR_API_KEY' \
  -H 'Model: p-video-2-pro-reference' \
  -d '{
    "input": {
      "prompt": "The person in the reference walks through a sunlit market",
      "image": ["https://api.pruna.ai/v1/files/fqadqq42xq"],
      "duration": 5,
      "resolution": "768p",
      "aspect_ratio": "adaptive"
    }
  }'

Cost mode (synchronous)

curl -X POST 'https://api.pruna.ai/v1/predictions' \
  -H 'Content-Type: application/json' \
  -H 'apikey: YOUR_API_KEY' \
  -H 'Model: p-video-2-pro-reference' \
  -H 'Try-Sync: true' \
  -d '{
    "input": {
      "prompt": "The person in the reference walks through a sunlit market",
      "image": ["https://api.pruna.ai/v1/files/fqadqq42xq"],
      "duration": 5,
      "resolution": "480p",
      "mode": "cost"
    }
  }'

Video, audio, and soft keyframes

Upload each file first, then pass the returned URLs:

curl -X POST "https://api.pruna.ai/v1/files" \
  -H "apikey: YOUR_API_KEY" \
  -F "content=@/path/to/your/file.jpg"

curl -X POST 'https://api.pruna.ai/v1/predictions' \
  -H 'Content-Type: application/json' \
  -H 'apikey: YOUR_API_KEY' \
  -H 'Model: p-video-2-pro-reference' \
  -d '{
    "input": {
      "prompt": "Continue the motion and match the reference audio",
      "video": ["https://api.pruna.ai/v1/files/video123"],
      "audio": ["https://api.pruna.ai/v1/files/audio123"],
      "first_frame_image": "https://api.pruna.ai/v1/files/start123",
      "last_frame_image": "https://api.pruna.ai/v1/files/end123",
      "duration": 5,
      "resolution": "768p",
      "mode": "speed"
    }
  }'

Configuration

Required parameters

Parameter	Type	Description
`prompt`	string	Text prompt for video generation

Send at least one image or video reference with the prompt. Audio is optional.

Optional parameters

Parameter	Type	Default	Description
`prompt_upsampler`	string	`"turbo"`	Rewrite the prompt before generation: `"turbo"` (fast default), `"max"`, or `"off"`. `"turbo"` reads the references, including audio. `"max"` falls back to `"turbo"` if it fails or takes longer than 90 s. Independent of `mode`; does not change the price. `"off"` still runs `"turbo"` when a start or end frame is set
`image`	array of URIs	—	Reference images
`video`	array of URIs	—	Reference videos
`audio`	array of URIs	—	Reference audio. Set `keep_audio` to use one file as the video’s audio
`keep_audio`	boolean	`false`	Keep the uploaded audio as the video’s audio. Needs exactly one audio file. Does not change the price
`first_frame_image`	string (URI)	—	Optional opening image. Soft keyframe that guides the open; not a hard constraint. Counts as a reference image
`last_frame_image`	string (URI)	—	Optional closing image. Soft keyframe that guides the close; not a hard constraint. Counts as a reference image
`duration`	integer	`5`	Video duration in seconds (5–15). With `keep_audio`, audio within 1 s of this sets the length
`resolution`	string	`"768p"`	`"480p"` or `"768p"`
`mode`	string	`"speed"`	Generation recipe: `"speed"` or `"cost"`. `"cost"` matches `"speed"` output quality at the lowest list price. There is no `"quality"` mode
`aspect_ratio`	string	`"16:9"`	`"adaptive"`, `"16:9"`, `"9:16"`, `"4:3"`, `"3:4"`, `"3:2"`, `"2:3"`, `"1:1"`. `adaptive` follows the references
`seed`	integer	random	Random seed for reproducible generation

Argument recommendations

Use these patterns for consistent quality:

prompt: follow subject / action / scene, then say what to keep from each reference. Add camera, lighting, and style when you need a repeatable final.
image / video / first_frame_image / last_frame_image: send at least one image or video; start and end frames guide the clip.
audio / keep_audio: upload only when timing must match; set keep_audio with one file to keep that track.
duration: 5–15 seconds (default 5).
mode: use cost for the lowest list price (same output quality as speed); use speed when latency matters.
prompt_upsampler: leave turbo for production; compare off vs turbo vs max on the same seed before scaling.
resolution: default 768p; use 480p for cheaper, faster drafts.