p-image-ideogram

Use when the user wants high-fidelity photoreal stills, editorial portraits, hero plates, legible text rendering, or poster typography—not the fastest draft stills or mood-board throughput.

p-image-ideogram (Pruna · Replicate deployment)

High-quality, fast text-to-image via the Pruna Ideogram preview deployment on Replicate. Runs on Replicate — not the Pruna P-API.

When to use

Goal	Use this
Photoreal hero plates, editorial portraits, product shots	Yes
Legible text rendering — posters, signage, ads, GTM graphics	Yes — mode: high or very high + image_size: 2K
Fastest possible draft stills, mood boards, bulk panels	No — use p-image (Pruna P-API; good quality, extremely fast)
Editing an existing image	No — use p-image-edit
Virtual try-on	No — use p-image-try-on

Mode & resolution

mode	Best for
very low	Basic photos, fast drafts, simple scenes
medium (default)	Candid realistic shots, documentary portraits, everyday photoreal
high · very high	Text rendering, advertising shoots, dense crowds / cast diversity — pair with image_size: 2K

Higher modes run stronger upsamplers (ideogram-sft, Gemini Flash, nano-banana-2-lite at very high). Prompt upsampling is built into mode — long specific prompts and named typography are OK when the brief needs readable text.

Dynamic persona & scenarios: realistic-persona-showcase.md · examples: example-prompt.md

Shared Replicate patterns: replicate-api.md

Environment

bash

export REPLICATE_API_TOKEN=r8_...

Model input (Replicate deployment)

Field	Notes
prompt	Required. Short text prompt for image generation.
mode	Quality/speed preset — see Mode & resolution. very low · low · medium (default) · high · very high.
aspect_ratio	1:1 (default) · standard ratios · custom (set width / height)
image_size	Output resolution budget: 1K (default) · 2K for text rendering, advertising shoots, and diversity-heavy scenes at high / very high. Ignored when aspect_ratio=custom.
width	When aspect_ratio=custom — rounded to nearest multiple of 16 (0–2560, default 1024).
height	When aspect_ratio=custom — rounded to nearest multiple of 16 (0–2560, default 1024).
seed	Optional integer for reproducibility.
output_format	jpg (default) · png
output_quality	0–100 (default 80); ignored for png.

Complete the random seed ritual (SSoT) before writing prompts — do not pass the ritual string as API seed. Set seed only when the user requests reproducibility.

Before generating

1. Generation diversity — ritual seed (SSoT) + explicit prompt structure. Pick mode per Mode & resolution: very low for basic photos, medium for candid realism, high / very high + 2K for text rendering, ads, and diversity. Name exact strings and placements for text briefs per text rules. Multi-example batches: different aspect_ratio per still.
2. Confirm prompt, mode, and aspect_ratio with the user. Run p-image-quality-checklist.md on outputs before downstream steps.

Production quality — photoreal personas

Same discipline as p-image: documentary portrait language, diverse cast, avatar-ready mouth visibility, try-on-ready body coverage. Full matrix: realistic-persona-showcase.md.

Lock hero plate URL at hero generation when the same identity continues to p-image-edit, p-image-try-on, or p-video-avatar.

Python (Replicate SDK)

python

import replicate

deployment = replicate.deployments.get("prunaai/p-image-ideogram-preview")
prediction = deployment.predictions.create(
    input={
        "prompt": "South Asian woman founder mid-30s, documentary portrait at cast-iron loft window, natural skin pores, mouth visible, hands away from mouth, golden hour side light, photoreal editorial",
        "aspect_ratio": "9:16",
        "mode": "medium",
    }
)
prediction.wait()
print(prediction.output)

HTTP (curl)

bash

curl -s -X POST \
  -H "Authorization: Bearer ${REPLICATE_API_TOKEN}" \
  -H "Content-Type: application/json" \
  -d '{
    "input": {
      "prompt": "Hong Kong neon alley at night, fearless grandmother in floral apron juggling dumplings, awning reads HAPPY HOUR 5-7, kiosk sign PRUNA AI, fish-eye lens, crisp legible typography",
      "aspect_ratio": "9:16",
      "mode": "high",
      "image_size": "2K"
    }
  }' \
  "https://api.replicate.com/v1/deployments/prunaai/p-image-ideogram-preview/predictions"

Poll urls.get until status is succeeded; download output (image URL).

Example: asynchronous (batch / multi-panel)

For N panels with no shared dependency, POST all jobs in parallel, then poll every urls.get. See parallel-execution.md.

Typical next steps

• Refine or composite: p-image-edit
• Virtual try-on: p-image-try-on
• Upscale output: p-image-upscale
• Animate still: p-video or p-video-avatar
• Full pipeline: pruna-generative-pipeline