Compare models
vs

Ideogram 3.0 vs Recraft V3

The two models pitched at design teams, head to head. Both land below their reputation on this suite (3.59 and 3.16), and Recraft is weakest exactly where design work needs consistency. What the data says before you standardize on either.

Same prompts, parameters and seeds across both models. The green check marks each row’s winner. Quality criteria (adherence, text, transparency, composition) are scored by an automated Claude vision tier.
Compare

Head to head

Pilot dataset (pilot-0). Criterion scores are 0–5 means over all runs.
  Ideogram 3.0 Recraft V3
Overall 2.97 / 5 3.29 / 5 ✓
Cost per image $0.060 $0.040 ✓
p50 latency 17.9s 7.4s ✓
Text Accuracy 4.7 / 5 ✓ 4.0 / 5
Color Accuracy 2.8 / 5 ✓ 1.9 / 5
Transparency 0.0 / 5 2.0 / 5 ✓
Resolution 3.0 / 5 ✓ 3.0 / 5 ✓
Prompt Adherence 3.9 / 5 ✓ 3.3 / 5
Composition 4.0 / 5 ✓ 3.0 / 5
Scores

Category winners

Hover a benchmark to see what it tests, or open its full results in a new tab.
  Ideogram 3.0 Recraft V3
Typography & Text Rendering 3.3 / 5 ✓ 3.2 / 5
Logos, Icons & Vector-Style 2.8 / 5 3.3 / 5 ✓
Brand-Color Fidelity 2.9 / 5 3.0 / 5 ✓
Transparency & Cutouts 2.2 / 5 3.1 / 5 ✓
Composition & Negative Space 3.1 / 5 3.3 / 5 ✓
Spatial & Compositional Adherence 3.1 / 5 3.6 / 5 ✓
Human Subjects: Faces & Hands 2.9 / 5 3.4 / 5 ✓
Product & E-Commerce Staging 3.2 / 5 3.7 / 5 ✓
Consistency & Repeatability 2.9 / 5 3.1 / 5 ✓
Print Detail & Resolution 3.0 / 5 3.4 / 5 ✓
Style Adherence & Art Direction 3.0 / 5 ✓ 3.0 / 5
Print-Production Graphics 3.1 / 5 3.6 / 5 ✓
UI & Design Mockups 3.3 / 5 3.5 / 5 ✓
Diagrams & Data Viz 3.0 / 5 3.2 / 5 ✓
Structured Spec Adherence 2.8 / 5 3.1 / 5 ✓
Benchmarks

Every benchmark, side by side

One representative generation per model (first seed). Click an image to enlarge, or the benchmark name for all runs and scores.

Typography & Text Rendering

Wordmark

Compare all models

Poster Headline

Compare all models

Product Label

Compare all models

CJK Neon Sign

Compare all models

Logos, Icons & Vector-Style

Mascot Logo

Compare all models

Icon Set

Compare all models

Monochrome Emblem

Compare all models

Brand-Color Fidelity

Two-Color Illustration

Compare all models

Duotone Portrait

Compare all models

ColorChecker Chart

Compare all models

Smooth Gradient

Compare all models

Transparency & Cutouts

Die-Cut Sticker

Compare all models

T-Shirt Graphic

Compare all models

Isolated Cutout

Compare all models

Composition & Negative Space

Banner with Negative Space

Compare all models

Vertical Story Background

Compare all models

Centered Product Margin

Compare all models

Spatial & Compositional Adherence

Object Counting & Placement

Compare all models

Scene Relations

Compare all models

Negation: Clean Background

Compare all models

Human Subjects: Faces & Hands

Hand Holding a Sphere

Compare all models

Product & E-Commerce Staging

Cosmetics Hero

Compare all models

Floating Sneaker

Compare all models

Consistency & Repeatability

Character Series: Greenhouse

Compare all models

Character Series: Desk

Compare all models

Character Series: Market

Compare all models

Print Detail & Resolution

Fine-Line Mandala

Compare all models

Illustrated Map

Compare all models

Style Adherence & Art Direction

Mid-Century Travel Poster

Compare all models

Isometric Office

Compare all models

Print-Production Graphics

Greeting Card

Compare all models

Seamless Pattern Tile

Compare all models

Stationery Mockup

Compare all models

Three-Panel Comic

Compare all models

UI & Design Mockups

E-Commerce Product Page

Compare all models

Diagrams & Data Viz

Labeled Bar Chart

Compare all models

Structured Spec Adherence

Structured JSON Scene

Compare all models
AI Editor

Route both. Finish in the AI Editor.

Whichever model wins this matchup for your jobs, neither delivers exact brand colors, editable text or guaranteed transparency. The IMG.LY AI Editor is the human-in-the-loop layer that does, giving your users the control they need to be productive with generative AI.

Explore the AI Editor Open Demo
IMG.LY AI Editor: generative image tools next to manual editing controls on a canvas

Frequently asked questions

Neither wins every category. This page runs the models on identical prompts and marks each row’s winner in the head-to-head table, so the right choice depends on the criteria your product needs: cost, text accuracy, transparency, brand color and so on.

The head-to-head table lists the measured cost per image for each model and marks the lower-cost one on the Cost row; latency is shown the same way. Verify against the provider before committing volume.

The green check marks the best value in each row: highest score for quality criteria, lowest for cost and latency. When every model scores zero on a row (transparency, for most of the field) no winner is marked, because least-bad is not best.

Each image is scored 0 to 5 per criterion. Measured criteria (resolution, latency, cost, transparency, color accuracy) are computed automatically; quality criteria (text accuracy, prompt adherence, composition) are judged by an automated Claude vision tier against each prompt’s checklist. Page scores are unweighted means over all of a model’s runs in that scope. Blind expert-panel review has not run yet; the dataset is pilot-0.

Yes. You can route generation to either model per job and refine the output in an editable canvas. The IMG.LY AI Editor pairs any model’s generation with background removal, brand kits and editable text, so your users take whichever model’s result to production quality.