Ideogram 4.5

​

Overview

• No files → text-to-image (fal:ideogram/v4.5)
• With files → editing (fal:ideogram/v4.5/edit). The first file is the image to edit, files 2–5 are references.

​

Text-to-image

import { createVarg, generateImage } from "vargai/ai"

const varg = createVarg({ apiKey: process.env.VARG_API_KEY! })

const { image } = await generateImage({
  model: varg.imageModel("ideogram_v4_5"),
  prompt: "minimalist concert poster with bold text reading VARG LIVE 2026, orange and black",
  aspectRatio: "3:4",
  providerOptions: { fal: { quality: "medium" } },
})

curl -X POST https://api.varg.ai/v2/image \
  -H "Authorization: Bearer $VARG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "ideogram_v4_5",
    "prompt": "minimalist concert poster with bold text reading VARG LIVE 2026, orange and black",
    "aspect_ratio": "3:4",
    "provider_options": { "fal": { "quality": "medium" } }
  }'

​

Parameters

​

Editing

curl -X POST https://api.varg.ai/v2/image \
  -H "Authorization: Bearer $VARG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "ideogram_v4_5",
    "prompt": "Restyle the poster using the colors and mood of the reference",
    "files": [
      { "url": "https://example.com/poster.png" },
      { "url": "https://example.com/reference.jpg" }
    ],
    "provider_options": { "fal": { "quality": "low" } }
  }'

​

Masks and high precision

• Mask — mask_url must match the source size. Black areas are edited, white areas are kept. With a mask, at most 3 references.
• High precision — edit_precision: "high" restores every pixel you did not ask to change. Same price as regular.

{
  "model": "ideogram_v4_5",
  "prompt": "Replace the year with the word TOUR in the same bold font",
  "files": [{ "url": "https://example.com/poster.png" }],
  "provider_options": {
    "fal": {
      "mask_url": "https://example.com/mask.png",
      "edit_precision": "high",
      "quality": "medium"
    }
  }
}

​

Parameters

​

Pinning a route

​

Tips

• Start at low to iterate on layout and wording, then re-run the winner at medium or high.
• Turn off prompt expansion (enable_prompt_expansion: false) when you need the exact wording and nothing else.
• Fix one word with a mask instead of regenerating the whole design.

​

Related models