Pruna AI

OpenTryOn integrates Pruna's unified predictions API (POST /v1/predictions with a Model header) through a shared client in tryon.api.pruna.client.PrunaClient.

Model	CLI	Adapter	Role
`p-image`	`generate --model p-image`	`PImageAdapter`	Ultra-fast text-to-image
`p-image-ideogram`	`generate --model p-image-ideogram`	`PImageIdeogramAdapter`	Pruna × Ideogram T2I (thinking levels)
`p-image-edit`	`edit --model p-image-edit`	`PImageEditAdapter`	Edit / compose 1–5 images
`p-image-upscale`	`edit --model p-image-upscale`	`PImageUpscaleAdapter`	Upscale to 1–128 MP
`p-image-try-on`	`vton --model p-image-tryon`	`PImageTryOnAdapter`	Multi-garment virtual try-on
`p-video`	`video-generate --model p-video`	`PVideoAdapter`	T2V / I2V (+ optional audio)
`p-video-2-pro`	`video-generate --model p-video-2-pro`	`PVideo2ProAdapter`	Fast MiniMax H3-based T2V / I2V (always-on generated audio)
`p-video-replace`	`video-generate --model p-video-replace`	`PVideoReplaceAdapter`	Identity swap in a source clip
`p-video-avatar`	`video-generate --model p-video-avatar`	`PVideoAvatarAdapter`	Talking-head from portrait + script/audio
`p-video-animate`	`video-generate --model p-video-animate`	`PVideoAnimateAdapter`	Animate a subject with source motion

Auth: PRUNA_API_KEY (optional PRUNA_BASE_URL). Key is sent as the apikey header.

Docs: Pruna model guides · P-Image-Ideogram · P-Video-2-Pro

Ideogram 4.0 via Pruna is skipped — use opentryon generate --model ideogram (IDEOGRAM_API_KEY). P-Image-Ideogram is a different model: opentryon generate --model p-image-ideogram (PRUNA_API_KEY). Full notes: P-Image-Ideogram.

Shared client

All adapters upload local files via /v1/files, create predictions with Try-Sync: true by default, and poll /v1/predictions/status/{id} when needed. Try-on lives under tryon.api.vton for historical reasons but reuses the same client; image/video models live in tryon.api.pruna.

Authentication

export PRUNA_API_KEY="your_api_key"
# optional:
# export PRUNA_BASE_URL="https://api.pruna.ai"

CLI examples

# Text-to-image
opentryon generate --model p-image \
  --prompt "editorial fashion still, soft window light" \
  --aspect-ratio 3:4

opentryon generate --model p-image-ideogram \
  --prompt 'Lookbook cover. Exact visible text only: "ATELIER NOIR"' \
  --thinking high --image-size 2K --aspect-ratio 3:4

# Multi-image edit
opentryon edit --model p-image-edit \
  --images person.jpg garment.jpg \
  --prompt "Dress the person in the garment, studio lighting"

# Upscale
opentryon edit --model p-image-upscale \
  --image result.jpg --target 8 --enhance-details

# Multi-garment try-on
opentryon vton --model p-image-tryon \
  --person-image person.jpg \
  --garment-image top.jpg --garment-image bottoms.jpg

# Text / image to video
opentryon video-generate --model p-video \
  --prompt "model walks toward camera, soft breeze" \
  --duration 5 --resolution 720p

opentryon video-generate --model p-video \
  --prompt "gentle head turn and smile" \
  --image still.jpg --duration 5

# Faster MiniMax H3-based sibling (always-on generated audio, no audio input)
opentryon video-generate --model p-video-2-pro \
  --prompt "model walks toward camera, soft breeze" \
  --duration 8 --resolution 768p --mode speed

# Identity replace in video
opentryon video-generate --model p-video-replace \
  --video source.mp4 \
  --images identity.jpg \
  --instruction-prompt "Place the reference person into the video"

# Talking-head avatar (script or audio)
opentryon video-generate --model p-video-avatar \
  --image portrait.jpg \
  --voice-script "Welcome to our spring collection." \
  --voice "Zephyr (Female)"

opentryon video-generate --model p-video-avatar \
  --image portrait.jpg \
  --audio speech.mp3

# Animate subject with source motion
opentryon video-generate --model p-video-animate \
  --video driver.mp4 \
  --image subject.jpg \
  --instruction-prompt "Keep the subject’s outfit and lighting"

Python quick start

from dotenv import load_dotenv
load_dotenv()

from tryon.api.pruna import (
    PImageAdapter,
    PImageEditAdapter,
    PImageIdeogramAdapter,
    PImageUpscaleAdapter,
    PVideoAdapter,
    PVideo2ProAdapter,
    PVideoAnimateAdapter,
    PVideoAvatarAdapter,
    PVideoReplaceAdapter,
)
from tryon.api.vton import PImageTryOnAdapter

# Generate
images = PImageAdapter().generate_text_to_image(
    prompt="luxury knitwear flatlay on marble",
    aspect_ratio="1:1",
)
images[0].save("out.png")

ideogram = PImageIdeogramAdapter().generate_text_to_image(
    prompt='Lookbook cover. Exact visible text only: "ATELIER NOIR"',
    thinking="high",
    image_size="2K",
)
ideogram[0].save("p-image-ideogram.jpg")

# Edit
edited = PImageEditAdapter().generate_image_edit(
    prompt="Replace the background with a clean studio",
    image=["photo.jpg"],
)

# Upscale
hires = PImageUpscaleAdapter().upscale(image="out.png", target=8)

# Try-on
tryon = PImageTryOnAdapter().generate_and_decode(
    person_image="person.jpg",
    garment_images=["top.jpg", "bottoms.jpg"],
)

# Video
mp4 = PVideoAdapter().generate_text_to_video(
    prompt="runway walk, cinematic tracking shot",
    duration=5,
    resolution="720p",
)
open("clip.mp4", "wb").write(mp4)

fast_mp4 = PVideo2ProAdapter().generate_text_to_video(
    prompt="runway walk, cinematic tracking shot",
    duration=8,
    resolution="768p",
    mode="speed",
)
open("fast_clip.mp4", "wb").write(fast_mp4)

replaced = PVideoReplaceAdapter().generate_video_replace(
    video="source.mp4",
    images=["identity.jpg"],
)

avatar = PVideoAvatarAdapter().generate_video_avatar(
    image="portrait.jpg",
    voice_script="Welcome to our spring collection.",
)

animated = PVideoAnimateAdapter().generate_video_animate(
    video="driver.mp4",
    image="subject.jpg",
)

Parameter notes

P-Image

Required: prompt
Optional: aspect_ratio (incl. custom + width/height), seed, prompt_upsampling, LoRA fields

P-Image-Ideogram

Required: prompt (natural language or Ideogram 4.0 JSON)
Optional: thinking (very-low … very-high, default high), image_size (1K/2K), aspect_ratio (default 1:1), prompt_upsampling (default on)
Typography: high or very-high at 2K; --no-prompt-upsampling for final JSON specs
See P-Image-Ideogram

P-Image-Edit

Required: prompt, 1–5 images
Optional: aspect_ratio (default match_input_image), turbo (default True), seed

P-Image-Upscale

Required: image
Optional: target MP (1–128, default 4), output_format, enhance_details, enhance_realism

P-Image-Try-On

Required: person image + ≥1 garment image (up to 11; 6 recommended)
Optional: prompt, turbo, reference_pose, output_format / output_quality
Pricing (per Pruna): $0.015 first garment + $0.008 each additional

P-Video

Required: prompt
Optional: image (I2V), audio (duration follows audio), duration 1–20s, resolution 720p/1080p, fps 24/48, draft, prompt_upsampling

P-Video-2-Pro

Required: prompt
Optional: image (I2V), last_frame_image, duration 5–15s (default 5), resolution 480p/768p (default 768p), aspect_ratio, mode speed/quality (default speed), prompt_upsampler off/turbo/max (default turbo), seed
Fixed 24fps; audio is always generated and cannot be supplied as input (unlike p-video)

P-Video-Replace

Required: source video, 1–3 identity images
Optional: instruction_prompt, resolution, target_fps, turbo, audio flags

P-Video-Avatar

Required: portrait image, plus voice_script and/or audio (audio wins if both)
Optional: voice, voice_language, resolution, video_prompt, voice_prompt, negative-prompt fields

P-Video-Animate

Required: source video, subject image
Optional: instruction_prompt, resolution, target_fps, turbo, audio flags

Shared client​

Authentication​

CLI examples​

Python quick start​

Parameter notes​

P-Image​

P-Image-Ideogram​

P-Image-Edit​

P-Image-Upscale​

P-Image-Try-On​

P-Video​

P-Video-2-Pro​

P-Video-Replace​

P-Video-Avatar​

P-Video-Animate​

See Also​