p-image-ideogram

Prerequisites

Install and load these skills before generating (skip if already in context via @pruna):

Or install the full suite once: npx skills add PrunaAI/pruna-skills@pruna -y

Follow each skill's Before generating / craft sections — do not restate guide content here.

Agent habit

Route by complexity: Use `p-image-ideogram` when photo generation needs more control — readable text in the image, structured JSON, hex/bbox placement, or high-detail photoreal shots. Use `p-image` for simple, quick photo generation. Use `p-image-edit` to change an existing photo; use `p-video-2-pro` / `p-video-2` (`p-video-animate` from a still) for motion.

In the first reply, name `p-image-ideogram` in backticks, confirm PRUNA_API_KEY is set (or stop with signup links from pruna-api), then ask for prompt / aspect ratio / any copy-on-surface (open intake → generation-diversity clarification intake). When drafting the prompt, follow Prompt craft below — do not paste skill examples.

Agent defaults (override API defaults): send thinking: "high", prompt_upsampling: true, and image_size: "1K" unless a profile in domain-configurations.md says otherwise. Use image_size: "2K" for dense in-image text, multi-panel layouts, and large output. Set prompt_upsampling: false when text is locked (JSON prompts, exact strings) or the user wants verbatim prompts only.

Premium path: when the user requests maximum quality or the composition is highly complex (many text elements, intricate multi-panel layouts, detailed structured scenes), send thinking: "very high" + image_size: "2K". Costs ~2× high ($0.033/1K, $0.066/2K) — confirm with the user before using unless they explicitly asked for top quality.

Speed path (same model): when the scene is simpler but you still want ideogram (or need a faster pass on this model), send thinking: "low", prompt_upsampling: false, and a nuanced, explicit prompt you fully draft — upsampling stays off because the prompt already carries the detail. Faster than the default high + upsampling path. For the simplest quick photo drafts, route to p-image instead.

When the job comes from a vertical-* workflow (or another multi-step production with spec copy or covers), pick thinking, image_size, and NL vs JSON from domain-configurations.md for that vertical and use-case # — do not use one global knob set for every industry.

vs p-image

Official parameters: P-Image-Ideogram

Prompt craft (dynamic + faithful)

Every input.prompt must be fresh and specific, and must keep the user's request. Diversity never overrides what the user asked for.

Fidelity check (before pay): if you remove the user’s named subject/product/setting from the prompt, the job is wrong — rewrite. Free axes only fill what the user did not specify.

Typography: list every string and surface; use image_size: "2K" and prompt_upsampling: false when legibility is critical (default thinking: "high"). Domain-specific profiles and vertical rows: domain-configurations.md. JSON captions (exact placement, brand hex, repeatable layout): ideogram-json-prompting.md.

Thinking & resolution

When NOT to use

Use a different skill instead:

HTTP (curl)

Create (async — recommended)

curl -X POST 'https://api.pruna.ai/v1/predictions' \
  -H 'Content-Type: application/json' \
  -H "apikey: ${PRUNA_API_KEY}" \
  -H 'Model: p-image-ideogram' \
  -d '{
    "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",
      "thinking": "high",
      "image_size": "1K",
      "prompt_upsampling": true,
      "aspect_ratio": "9:16"
    }
  }'

Poll and download: follow pruna-api.

Complete the random seed ritual from generation-diversity before writing prompts — do not pass the ritual string as API seed. Optional seed only when the user requests reproducibility.

Create (sync — quick test only)

curl -X POST 'https://api.pruna.ai/v1/predictions' \
  -H 'Content-Type: application/json' \
  -H "apikey: ${PRUNA_API_KEY}" \
  -H 'Model: p-image-ideogram' \
  -H 'Try-Sync: true' \
  -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","thinking":"high","image_size":"2K","prompt_upsampling":true,"aspect_ratio":"9:16"}}'

Generation flow

Follow generation-diversity still-image prompt flow every time:

1. Lock the request — subject, product, format, any text on the image.
2. Ritual seed — fresh string; derive free axes (camera, lighting, render_category_tag, aspect_ratio when unset).
3. Pick knobs — domain-configurations.md profile for the vertical/use case, else default thinking: high, image_size: 1K, prompt_upsampling: true; or speed path — thinking: low, prompt_upsampling: false, nuanced explicit prompt; or premium path — thinking: "very high", image_size: "2K" for maximum quality on complex compositions; raise image_size to 2K for dense in-image text or multi-panel layouts; set prompt_upsampling: false for locked text or JSON.
4. Draft explicit prompt — Prompt craft + image-prompting golden rules; fidelity check before pay.
5. Confirm — show prompt + knobs unless wording is locked.
6. POST — async curl below; poll via pruna-api; run p-image quality checklist in image-prompting before upscale/video.

Aspect ratio: pass aspect_ratio in input; if output dimensions do not match (e.g. asked 16:9, got portrait), retry once with explicit horizontal wide / vertical wording in the prompt.

Mood board / batch: new ritual per independent photo; different aspect_ratio per panel when format not locked.

Photo approved → edit: hand off to p-image-edit on the output URL for photo edits — do not run photo generation again for the same subject; do not use edit to fix dense multi-panel in-image text (regenerate instead).

Photo approved → video: use p-video (image-to-video) or p-video-animate when motion is next; upscale first with p-image-upscale if resolution is tight.

Required input

• prompt (string)

Common optional fields

• thinking: very low, low, medium, high, very high — default high; low + prompt_upsampling: false + explicit prompt for the speed path; very high for maximum quality on complex compositions
• image_size: 1K, 2K (ignored when aspect_ratio is custom)
• prompt_upsampling: boolean — default true with thinking: high; false on speed path, locked copy, JSON prompts, or verbatim wording
• aspect_ratio: 1:1, 16:9, 9:16, 4:3, 3:4, 3:2, 2:3, custom (with width / height up to 2560, multiples of 16)
• seed, output_format (jpg, png, webp), output_quality (0–100; ignored for png)

Typical next steps

Common follow-ons after this skill: