Skip to content
trending

News

Ideogram 4.5: Multi-Edit Model That Doesn’t Break Photos

Ideogram 4.5 launched Sept 30, 2026 with drift-free multi-step editing, 24 Mpx support, and $0.008-0.22 per image. We tested the producer's claim on 8 models.

By 3 min read
Ideogram 4.5: Multi-Edit Model That Doesn’t Break Photos

👁 121 reads

Ideogram released Ideogram 4.5 on September 30, 2026. This isn’t another pretty-picture generator - it’s a model for step-by-step photo editing without color and texture drift. It works in the app, via API, and from day one with partners including fal.ai. It costs $0.008 to $0.22 per image. Open weights are promised but without a date. The next day we verified the producer’s claim on 8 models and it held up.

Ideogram has been known for one thing: text on graphics. Version 4.5 shifts the focus to editing. The producer describes a problem familiar to anyone who’s edited a photo in ChatGPT or Gemini: after the third or fourth correction, the image starts living its own life. Colors change, faces shift, wall plaster and material textures transform, even though you only asked for a different door color.

Ideogram 4.5 Hub - everything about the model: tests, pricing, comparisons and prompts.

What Ideogram Actually Showed

  • Multi-step editing. The same image can be corrected many times without pixel shifts, color changes, and artifacts.
  • Precise Edit. A mode that returns an image in input dimensions and restores from the original the pixels the edit didn’t concern.
  • Large photos. Source up to 4016 × 6016 px. The model can correct a crop of a large photo and paste it back without downsizing the whole.
  • References and mask. Up to four reference images, optional mask for the area to change.
  • Four quality tiers. From $0.008 to $0.22 per image, price doesn’t depend on resolution.
  • Open weights “soon”. Without date, license, or hardware requirements.

Where It Works From Launch Day

  • ideogram.ai app and Ideogram API (two endpoints: precise editing and generation with editing),
  • fal.ai - where we ran our tests,
  • Picsart, Krea, Runway, Pika, Gamma, Luma,
  • ComfyUI.

Bold Producer Claim

Ideogram showed a comparison in which GPT Image 2.5 Sunburst, Nano Banana Pro, and Nano Banana 2 “become unusable after a few edits,” while 4.5 stays clean. That’s a producer comparison, and those always cherry-pick examples in their favor. So we ran our own.

We Tested It: 10 Edits, 8 Models

One bakery photo, 10 commands in sequence, each model correcting its own previous result. At the end we measured how much remained of the fragments it had no right to touch.

Model Similarity After 10 Steps (SSIM) What Happened to the Image
Ideogram 4.5 0.95 changes only what I asked for
Seedream 5.0 Pro 0.58 growing pink tint, different door
Nano Banana 2 0.57 pink wall at the end
Nano Banana Pro 0.54 pink wall, repainted storefront frame
Qwen Image 3 0.48 faded colors
GPT Image 2.5 Sunburst 0.43 9 out of 10 commands, plaster and lighting repainted from scratch
FLUX.2 Max 0.32 new pavement, added text on storefront
Grok Imagine Image 2.0 0.22 from first step different wall and frame
Bakery after 10 edits in Ideogram 4.5
Ideogram 4.5, step 10
Bakery after 10 edits in Nano Banana 2 with pink tint
Nano Banana 2, step 10

All steps from all models can be reviewed in the full editing test.

Where Ideogram 4.5 Doesn’t Win

  • Small text (we tested in Polish). In our text test it wrote correctly 11 out of 13 labels. GPT Image 2.5, Nano Banana Pro, Nano Banana 2, and Grok did 13 out of 13.
  • Voting arenas. In Artificial Analysis Ideogram 4.5 has 1014 points in text-to-image (34th place out of 167), barely more than Ideogram 4.0. In editing it has 1064 points (23rd place). Arenas judge a single image, though, not the tenth correction.
  • Price in high tier. $0.22 per image is more than most rivals. Medium tier costs $0.06.
  • Generation resolution. Through fal.ai in text-to-image mode we got images around 1 Mpx, not 2K.
Google · Your sourcesQuantslant higher in your Google - one clickAdd to preferred sources →

What This Means

For designers, online stores, and agencies this is the most interesting image release in months. Correcting a product photo, localizing a banner for another market, or a series of changes to a client photo are exactly the tasks where other models quietly break the image. See what Ideogram 4.5 can do in practice:

Kinetyra

// read next

More on this from QUANTSLANT

Piotr Olszewski

Editor and publisher · Quantslant

I edit and publish Quantslant. AI tools help with research, drafts and translation; choosing topics, running the tests, checking facts and the decision to publish are mine, and so is the responsibility for what appears here.

read nextPerplexity Deep Research: Complete Guide to AI Analysis in 3 MinguidesPerplexity Deep Research: Complete Guide to AI Analysis in 3 MinPerplexity Deep Research delivers comprehensive reports in under 3 minutes, scoring 67.15% on DRACO benchmark. Learn how to use it effectively.

Most read

  1. 1Best AI Video Generator Apps for Your Phone (2026)
  2. 2Claude Opus 5.5 Motion Graphics: 60s Showreel from One Prompt
  3. 3Wan 2.2 on RTX 4090: timings, VRAM and four test clips
// site map

This page in other languages1

Quantslant in numbers 1,423posts in archive22news from the last 7 days18video models in the observatory
🎬 AI Video Benchmarks: Higgsfield DoP vs OpenAI Sora • Higgsfield vs Runway Gen-3 • AI Video Credit Calculator
× ‹ enlarge ›