Ideogram 4.5: repeated-edit tests, pricing and image examples

👁 154 reads

Ideogram 4.5, released on September 30, 2026, emphasizes precise image editing across repeated changes. In the source author’s eight-model test, regions meant to remain untouched scored SSIM 0.95 after ten edits, compared with 0.43 for GPT Image 2.5 Sunburst and 0.54 for Nano Banana Pro. SSIM is a similarity metric, not the percentage of pixels preserved. The source also found weaknesses in small Polish lettering. This guide collects the dated tests, USD prices and 57 gallery outputs, without presenting them as a new English-language benchmark.

The source author ran these tests on October 1, 2026 through fal.ai, giving the models the same tasks with one attempt rather than selecting a best result. Scripts calculated the reported metrics. This adaptation does not claim that the images were regenerated or that the language tests were repeated in English.

What is Ideogram AI?

Ideogram is an image-generation and editing company based in Toronto, founded in 2022 by former Google Brain researchers according to the source profile. Its early focus was legible lettering for logos, posters, covers and advertisements. Access includes ideogram.ai, supported mobile experiences and APIs; the available models and features depend on the route.

• Ideogram 4.5 - released September 30, 2026, with an emphasis on precise editing. A text-to-image endpoint also exists, so it is not accurate to describe 4.5 as incapable of generating an image from scratch.
• Ideogram 4.0 - the source’s June 2026 generation model with JSON prompting and released weights.
• Ideogram 3.0 - the earlier generation with Turbo, Default and Quality modes, listed as still available through APIs in the source snapshot.

The interface and the source test prompts use English. The lettering test deliberately requested Polish text, so its errors should not be transferred to English typography without another test. The broader product assessment appears in the Ideogram review.

Five points about Ideogram 4.5

1. Precise editing is its main focus. The source describes a mode intended to restore unchanged image regions. In fal.ai’s schema, edit_precision=high selects Precise Edit, while quality is a separate setting. Choosing high quality alone is not the same configuration.
2. Repeated edits with less image drift. That is the central manufacturer claim tested here. In the source sequence, the wall, bakery window and pavement remained much closer to their original appearance than in the other tested outputs.
3. Editing larger source images. The source describes inputs up to 4016 × 6016 pixels, approximately 24 megapixels, and crop-based edits that can be merged back into a larger image. It reports roughly one-megapixel text-to-image outputs through fal.ai, including 1024 × 1024, 864 × 1152 and 1280 × 720, despite a manufacturer native-2K claim. Endpoint limits and source preservation differ; check the current schema rather than assuming every route accepts or returns the same size.
4. Reference images and masks. The fal.ai schema checked October 12 allows up to four references alongside the primary image, or three references when a mask is also supplied. That distinction matters when assembling a request.
5. An announced weight release is not an available download. The source records an open-weights announcement without a date, license or hardware requirements for 4.5. It identifies 4.0 as the earlier release with weights on Hugging Face. Check the current official repository before planning local deployment.

Source tests and galleries

The original coverage comprised two comparisons and five prompt galleries. Each displayed example is described as the first result from its prompt, with errors documented rather than hidden.

01Ideogram 4.5 vs 7 Models: 10 Edits in a Row Without DriftWe tested whether Ideogram 4.5 can handle 10 consecutive edits without degrading the image. SSIM and ΔE measurements for 8 models show the results.Read →02Ideogram 4.5 Image Editing: 10 Before & After Examples10 real edits in Ideogram 4.5: label changes, banner translation, menu prices, time of day, text removal. Precise mode vs regular mode tested.Read →03Ideogram 4.5: 15 Graphic Styles Tested with PromptsWe generated 15 images in 15 graphic styles with Ideogram 4.5, from Polish poster school to vaporwave. Typography and flat color styles scored best, with correct diacritics in non-English text.Read →04Ideogram 4.5 Photography Test: 12 Shots with PromptsWe generated 12 photographic shots in Ideogram 4.5: portrait, macro, drone, packshot, flat lay, street, and more. The model handles technique well but struggles with local details.Read →05Ideogram 4.5 Brand Rebranding: 10 Concepts with PromptsWe created 10 brand identity concepts in Ideogram 4.5: coffee roaster, dairy, bakery, pharmacy, railway, cosmetics, brewery, festival, fintech and shelter.Read →06Ideogram 4.5 for Social Media & Print: 10 Projects with PromptsWe created 10 social media and print designs in Ideogram 4.5: YouTube thumbnail, Instagram carousel, wedding invite, business card, and more. Prompts and ratings included.Read →07Ideogram 4.5: Multi-Edit Model That Doesn’t Break PhotosIdeogram 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.Read →

Test 1: ten consecutive edits

The author began with a Krakow bakery photograph and asked eight models to make the same ten changes in sequence. Each model edited its own previous output. The requested changes included:

• repainting the door;
• adding a Polish opening-hours sticker showing 7-22;
• replacing a bicycle with a scooter;
• removing a car;
• changing the sign to a specified Polish bakery name;
• placing a cat on the windowsill;
• and changing the price on a board.

The analysis measured change in regions that none of the prompts was meant to alter, including the wall, the shop-window interior and the pavement.

Ideogram preserved those regions best in this sequence, though its 0.95 score is not perfect identity. The source describes a growing pink cast in the Nano Banana outputs; Grok and FLUX.2 Max changed the window frame in the first step when only the door was targeted. GPT Image 2.5 completed nine of ten instructions, placing the cat behind the glass instead of on the sill, while texture and lighting drifted across successive edits. The complete 80-image sequence, step-by-step metrics and method are in the full editing test.

Test 2: Polish lettering

The source gave nine models four tasks involving Polish diacritics: a sale poster, a traditional cafeteria menu, a book cover and a dumpling package. Across the tasks, 13 text strings had to be reproduced exactly. These were not English-language lettering prompts.

Ideogram did not lead this test. Large headings were strong, but a small date line was corrupted and one initial received the wrong diacritic. The source’s GPT Image, Nano Banana and Grok entries reproduced more of the tested Polish strings correctly. This is evidence for those examples, not a blanket claim about English copy. All 36 images are in the Polish-lettering comparison.

Example galleries with prompts

The source used Ideogram 4.5’s text-to-image route at high quality for the generation galleries, producing 47 images alongside ten editing examples. Each includes a prompt and notes about the result.

• 15 graphic styles - including Polish poster art, risograph, Bauhaus, linocut and historical neon design. The gallery retains the actual original examples.
• 12 photographic approaches - an 85 mm portrait, macro, drone view, product shot, double exposure and tilt-shift.
• Ten brand-identity concepts - a cafe, brewery, pharmacy, railway, fintech service, packaging and signs.
• Ten social and print layouts - a YouTube thumbnail, Instagram carousel, invitation, menu and wine label.
• Ten photo-editing examples - product-label changes, seasons, brand colors, banner localization and inserting a product into a scene.

Ideogram 4.5 pricing

The source fal.ai price snapshot bills by image and quality tier. Quality and edit precision are separate controls: the price table below does not by itself select Precise Edit. Figures are US dollars before any applicable taxes; confirm the selected endpoint’s current rate before generating.

The source reports that its repeated-edit test, lettering comparison, 47 gallery images and ten editing examples cost a low double-digit dollar total, without an exact consolidated invoice in this article. High quality costs more per image than many of the cited alternatives. Whether it saves money overall depends on how many retries or manual corrections your own task needs.

The source lists these ideogram.ai subscription prices in US dollars before tax. Annual figures are monthly equivalents, not the total annual invoice:

• Free - weekly slow-credit allowance, with public images in the source plan description.
• Plus - $20/month, or $15/month billed annually, with 1000 priority credits and private images in the source snapshot.
• Pro - $60/month, or $42/month billed annually, with 3500 priority credits.
• Team - $30 per person per month, with a two-person minimum in the source snapshot.

The source describes AI editing and reference-image features in the application as consuming priority credits. Check current plan entitlements and renewal terms before subscribing.

Where to use Ideogram 4.5

• The ideogram.ai application - a browser workflow with text-layer editing, depending on the current interface and plan.
• Ideogram APIs - the source distinguishes precise editing from generation-and-editing routes. Output sizing and preservation depend on the endpoint and settings.
• Launch partners and integrations - the source lists fal, Picsart, Krea, Runway, Pika, Gamma, Luma and ComfyUI. This list does not imply identical features or local downloadable weights in every integration.

Which task suits which workflow?

The broader comparison is in the AI image generator ranking. Separate coverage: GPT Image 2.5.

Ideogram 4.5 compared with 4.0

The source AA snapshots place Ideogram 4.5 below the leading overall image entries. A single-output preference ranking and a ten-step preservation test evaluate different things. In the source’s repeated-edit scene, Ideogram drifted least among the tested models; that result does not establish superiority on every image task.

Frequently asked questions

Is Ideogram 4.5 free?

The source describes weekly slow credits on free application accounts, public outputs and priority-credit requirements for AI editing. Its API price range is $0.008-$0.22 per image by quality tier. Confirm current account eligibility and endpoint pricing rather than assuming every 4.5 feature is free.

How were the prompts and lettering tested?

The source author wrote the main instructions in English and placed the desired Polish text in quotation marks. That setup tests following English instructions while rendering another language. It does not provide a separate English-lettering accuracy score; inspect small print at full size.

What is the difference between high quality and high edit precision?

They are separate API parameters. The checked fal.ai schema uses edit_precision=high for Precise Edit with unchanged-region restoration; regular uses the generation route. The quality parameter independently offers very_low, low, medium and high. Set both deliberately rather than assuming quality=high enables precise editing.

Can I run Ideogram 4.5 locally?

The source records only an announcement of future 4.5 weights, without a release date or license. It identifies 4.0 as having downloadable weights. Check official repositories for the current state; a ComfyUI integration alone does not prove local inference.

Is Ideogram 4.5 better than GPT Image 2.5?

It preserved the untouched regions better in the source ten-edit sequence. GPT Image performed better on the source’s Polish lettering and overall AA snapshot. Choose based on the task and review method rather than treating either result as a universal ranking.

API setting reference: fal.ai Ideogram V4.5 Edit schema.

// read moreIdeogram 4.5 Image Editing: 10 Before & After ExamplesAI ToolsKrea 2 vs Ideogram 4.5: Which AI Image Generator to ChooseBenchmarksIdeogram 4.5 vs 7 Models: 10 Edits in a Row Without DriftBenchmarks

Kinetyra