Skip to main content

P-Image-Ideogram

P-Image-Ideogram is a fast text-to-image model from Pruna × Ideogram. OpenTryOn integrates the Pruna predictions API (Model: p-image-ideogram) — the same client as the rest of the P-Image family.

CLI modelMCP toolAdapterAuth
p-image-ideogramgenerate_p_image_ideogramPImageIdeogramAdapterPRUNA_API_KEY

No open weights. Studio Image → Generate picks it up from the live MCP catalog after an MCP restart.

This is not Ideogram 4.0 (generate --model ideogram / IDEOGRAM_API_KEY). Ideogram also publishes a first-party POST /v1/text-to-image/p-image-ideogram (four Quality levels, no Very High). We did not add a second hoster id; use Pruna for the five-level thinking surface.

Environment​

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

Key from the Pruna dashboard. Sent as the apikey header.

Thinking vs resolution​

Pruna thinking values (CLI uses hyphens; the adapter maps them to the API strings):

CLI / MCPAPI thinking1K price (list)Typical use
very-low"very low"$0.003Drafts, A/B, high volume
low"low"$0.0075Simple prompts at scale
medium"medium"$0.01Balanced production
high (default)"high"$0.015Complex prompts, everyday production
very-high"very high"$0.033Hero assets; slowest (~5.5s+ at 1K)

image_size is 1K (default) or 2K (ignored when --aspect-ratio custom). Typography: use high or very-high at 2K, keep copy short, and add an Exact visible text only: / CRITICAL TYPOGRAPHY: block. Difficult text does not work well at very-low / low.

Prompt upsampling is on by default. Pass --no-prompt-upsampling (MCP prompt_upsampling: false) for final JSON specs or exact on-image text.

JSON prompts follow the Ideogram 4.0 structured-caption schema (high_level_description + compositional_deconstruction).

CLI​

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

# Fast draft
opentryon generate --model p-image-ideogram \
  --prompt "editorial street style, no logos, no text" \
  --thinking very-low --no-prompt-upsampling

Python​

from tryon.api.pruna import PImageIdeogramAdapter

images = PImageIdeogramAdapter().generate_text_to_image(
    prompt='Lookbook cover. Exact visible text only: "ATELIER NOIR"',
    thinking="high",
    image_size="2K",
    aspect_ratio="3:4",
)
images[0].save("p-image-ideogram.jpg")

Notes​

  • Endpoint: POST https://api.pruna.ai/v1/predictions with headers Model: p-image-ideogram and Try-Sync: true (poll if needed). Rate limit: 500 rpm.
  • Defaults match Pruna: thinking=high, aspect_ratio=1:1, image_size=1K, prompt_upsampling=true, output_format=jpg.
  • Custom size: --aspect-ratio custom --width … --height … (0–2560, multiple of 16).
  • TryOn Studio: Connect → Pruna (PRUNA_API_KEY); Image generate model p-image-ideogram. Chat: name p-image-ideogram (otherwise generate defaults to nano-banana-pro). Saying only “ideogram” still pins Ideogram 4.0.
  • Provider docs: Pruna P-Image-Ideogram

See also​