P-Image-Ideogram

P-Image-Ideogram is Pruna AI’s text-to-image model, built in collaboration with Ideogram. It offers five thinking levels that trade off speed, price, and quality: Very Low, Low, Medium, High, and Very High.

P-Image-Ideogram beauty shade campaign (Medium) P-Image-Ideogram fashion lookbook JSON (Medium)

Use it for photoreal campaigns, product and people-led visuals, precise typography, and structured JSON prompts with hex colors and bounding boxes. Pair High (the default) for everyday production, Very Low or Low for drafts and high-volume exploration, and Very High at 2K or above for hero assets where final quality matters more than latency or cost.

At a glance

Metric

P-Image-Ideogram

Speed (1K, prompt upsampling on)

from 0.4 s (Very Low) to 5.5 s (Very High), depending on thinking, image_size, and prompt_upsampling

Price

$0.003 (Very Low, 1K) to $0.066 (Very High, 2K) per image

Thinking levels

"very low", "low", "medium", "high" (default), "very high"

Text rendering

Strong; use High or Very High at 2K+ for difficult typography

API model header

Model: p-image-ideogram

Thinking modes

Latencies below are minimum times at 1K with prompt_upsampling: true. Prices are per image for image_size 1K or 2K. Higher thinking values generally work best with more detailed prompts. On vague prompts, High can add unrequested detail. Low is often better for simple, non-specific prompts. Very High spends extra compute at generation time to push final quality further, so it is the slowest and most expensive level: reach for it when quality outweighs latency and cost, not as an everyday default.

thinking

1K

2K

Latency

Best for

"very low"

$0.003

$0.006

from 0.4 s

High-volume generation, drafts, exploration, A/B testing

"low"

$0.0075

$0.015

from 0.8 s

Cost-efficient production at scale; simple prompts

"medium"

$0.01

$0.02

from 1.3 s

Everyday production: balanced quality and cost

"high" (default)

$0.015

$0.03

from 1.3 s

Complex prompts, fine detail, brand-critical assets, difficult text at 2K+

"very high"

$0.033

$0.066

from 5.5 s

Maximum quality: hero assets, difficult prompts, and quality-critical work at 2K

Strengths and limits

Strengths

Details

Photorealism

Very strong performance on people, products, and lifestyle scenes

Text rendering

Reliable on-image typography when you use "high" or "very high" at 2K+

Flexible prompting

Natural language and structured JSON; hex colors and bounding boxes supported

Speed and cost

Fast and affordable from Very Low through High; Very High trades both for quality

Limits

Impact

Creative styles

Non-photoreal and stylized looks are still improving

Low-mode artifacts

"low" can show more artifacts than "medium" or "high"

Text at low thinking

Difficult text rendering does not work well in "very low" or "low". Use "high" or "very high" at 2K

Very High cost and latency

"very high" costs about 2.2x "high" and starts at 5.5 s at 1K. Use it for final hero assets, not for iteration

Mode choice

Higher thinking is not always better: match the mode to the prompt. "low" for simple prompts, "high" for complex ones. On vague prompts, "high" may add unwanted detail

Feature highlights

Four capabilities that define how to use P-Image-Ideogram in production. Open Show prompt under each example to copy the text.

Five generation levels via thinking: different quality / latency / cost.

Note

Higher modes work best with detailed prompts. On a vague prompt, "high" can invent unrequested detail. Prefer "low" or "medium" for simple exploration, and keep "very high" for prompts detailed enough to reward the extra compute.

Street-style editorial

fashion_street_style Very Low
Very Low
fashion_street_style Low
Low
fashion_street_style Medium
Medium
fashion_street_style High
High
fashion_street_style Very High
Very High
Show prompt
Editorial fashion photograph of a fictional young creative professional walking through a rainy residential neighborhood in Copenhagen, oversized neutral tailoring, silver accessories, natural walking movement, soft overcast light, wet pavement reflections, authentic street-style photography, realistic face and hands, no visible logos, no text.

Domain use cases

Useful for lookbook moods, street-style editorials, show collateral, and colorway lifestyle strips before a shoot.

Colorway lifestyle strip

fashion_colorway Very Low
Very Low
fashion_colorway Low
Low
fashion_colorway Medium
Medium
fashion_colorway High
High
fashion_colorway Very High
Very High
Show prompt
Wide ecommerce product lifestyle strip for a fictional sneaker colorway launch.

A single generic white sneaker on a sunlit concrete step, photoreal materials, no logos on the shoe. Clean negative space on the right.

Exact visible text only:
Product line exactly: "AERO RUN"
Colorway exactly: "Salt / #E8E2D6"
No other text, no logos, no watermark.

Runway invitation

fashion_runway_invitation Very Low
Very Low
fashion_runway_invitation Low
Low
fashion_runway_invitation Medium
Medium
fashion_runway_invitation High
High
Show prompt
Vertical fashion-week invitation card on warm ivory paper with a minimal architectural photograph of a concrete gallery doorway in the upper half.

Exact visible text only:
Title exactly: "ATELIER NOIR"
Line 2 exactly: "FALL 26 SHOW"
Line 3 exactly: "19 SEPT - PARIS"
No logos, no QR codes, no additional text, no watermark.

Integration

Endpoint

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

Rate limit: 500 requests per minute

Full API reference: P-Image-Ideogram

Authentication

-H 'apikey: YOUR_API_KEY'

Asynchronous (default)

curl -X POST 'https://api.pruna.ai/v1/predictions' \
-H 'Content-Type: application/json' \
-H 'apikey: YOUR_API_KEY' \
-H 'Model: p-image-ideogram' \
-d '{
    "input": {
        "prompt": "Vertical beauty campaign visual for a fictional fragrance launch...",
        "thinking": "medium",
        "image_size": "2K",
        "aspect_ratio": "3:4",
        "prompt_upsampling": true
    }
}'

Synchronous (Try-Sync)

curl -X POST 'https://api.pruna.ai/v1/predictions' \
-H 'Content-Type: application/json' \
-H 'apikey: YOUR_API_KEY' \
-H 'Model: p-image-ideogram' \
-H 'Try-Sync: true' \
-d '{
    "input": {
        "prompt": "Editorial fashion photograph of a fictional model in Copenhagen rain...",
        "thinking": "low"
    }
}'

Configuration

Parameters below match the live P-Image-Ideogram API reference and follow the same request shape as p-image.

Required parameters

Parameter

Type

Description

prompt

string

Short text prompt for image generation. Supports natural language and JSON structured prompts.

Optional parameters

Parameter

Type

Default

Description

thinking

string

"high"

Quality level: "very low", "low", "medium", "high", "very high". Higher thinking → higher quality; pair "high" or "very high" with detailed prompts and 2K+ for text.

prompt_upsampling

boolean

true

Enhance the prompt before generation. Set false to pass the raw prompt directly to the model.

aspect_ratio

string

1:1

1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, or custom

image_size

string

1K

Output resolution budget: 1K or 2K (ignored when aspect_ratio=custom). Use 2K with "high" or "very high" for best text rendering.

width

integer

1024

Custom width (0–2560, multiple of 16). Only when aspect_ratio=custom.

height

integer

1024

Custom height (0–2560, multiple of 16). Only when aspect_ratio=custom.

seed

integer

random

Random seed for reproducible generation

output_format

string

jpg

Output image format

output_quality

integer

80

Output quality 0–100 (ignored for PNG)

Example: text rendering at High + 2K

{
    "input": {
        "prompt": "Vertical beauty campaign... Exact visible text only: Product name exactly \"ROUGE ATELIER\"...",
        "thinking": "high",
        "image_size": "2K",
        "aspect_ratio": "3:4",
        "prompt_upsampling": true,
        "seed": 42
    }
}

Example: fast draft at Very Low

{
    "input": {
        "prompt": "Editorial fashion photograph of a fictional model walking in Copenhagen rain, no logos, no text.",
        "thinking": "very low",
        "prompt_upsampling": false
    }
}

Example: raw JSON prompt at Very High without upsampling

{
    "input": {
        "prompt": "{\"aspect_ratio\": \"3:4\", \"high_level_description\": \"Luxury fashion lookbook cover...\", \"compositional_deconstruction\": {...}}",
        "thinking": "very high",
        "prompt_upsampling": false
    }
}