IdeogramIdeogram·🖼️ Image Generation·New

Ideogram V4.5

anonymized
Try on Venice.ai ↗
Quick reference
Ideogram V4.5 — TLDR
  • 🆕 Ideogram's latest text-to-image model, released September 2026
  • 🏢 Built by Ideogram, which emphasizes multilingual in-image text rendering
  • 🎯 Iteration on Ideogram 4.0, offered alongside it on Ideogram's platform
  • 📏 The 4.0 generation introduced native 2K-resolution output
  • 🔧 Structured JSON prompts with bounding boxes and color palettes
  • 🌐 Multilingual in-image typography is a core family focus
  • 💬 Optional "magic prompt" expansion turns plain text into structured captions
  • 🧠 Paired with a matching inpainting/editing variant for revisions
💰 Pricing
$0.260
per image
📅 On Venice since
Sep 30, 2026
2 days ago
Provider

Ideogram is a text-to-image AI company founded in 2022 by Mohammad Norouzi, William Chan, Chitwan Saharia, and Jonathan Ho. The team set out to build a generative image model with a particular strength that many rivals struggled with: rendering legible,…

Read full profile →
3 models on Venice
2 image · 1 inpaint
Since Jun 3, 2026

About this model

Ideogram V4.5 is the current text-to-image model from Ideogram, a lab whose models emphasize multilingual in-image text rendering. It arrives as a point-release refinement of Ideogram V4, and Ideogram's own site lists both 4.0 and 4.5 as the models behind its apps, Studio, and API.

The 4.x generation marked a significant architectural reset for the family. According to Ideogram's model card and technical blog, Ideogram 4 is a foundation model trained entirely from scratch rather than a fine-tune or distillation, and it introduced a structured JSON prompting interface with explicit bounding-box layout control, per-element color palettes, strong multilingual text rendering, and native 2K image output. The text encoder is a vision-language model (Qwen3-VL-8B-Instruct), with the diffusion transformer consuming hidden states from 13 of its intermediate layers, and training used JSON captions exclusively, so the inference-time format matches what the model saw during training. Sampler presets trade steps for speed, from a short turbo configuration up to a 48-step quality preset.

V4.5 continues that interface and design-oriented focus, with a companion editing model, Ideogram V4.5 (edit), for inpainting and targeted revisions to existing images. Ideogram has not published a separate technical card for the 4.5 point release, so no documented quality deltas over 4.0 are available. The family is typically applied to typography-heavy layouts such as posters, packaging and UI mockups.

This About section is AI-generated from public sources (Claude Opus 5), with no human editing. It may contain inaccuracies — verify critical details against the sources listed above.

Data sources: Venice API · HuggingFace · Wikipedia — enrichment updated 1d ago