Typography and layout
Legible text, hierarchy, and composition
Gradually Image Model Index
Compare AI image models, model by model
28 image generation models, 378 head-to-head matchups, and separate arenas for text-to-image generation and image editing. Compare pricing, model access, open weights, specifications, and documented sample images.
- image models
- 28
- head-to-head matchups
- 378
- arena results
- 213
- separate arenas
- 12
Current leaders
Which AI image model is the best?
GPT Image 2 ranks first for text-to-image generation in the stored Artificial Analysis snapshot. Reve 2.1 ranks first for image editing.
That is not a universal verdict. Editing confidence intervals can overlap, and typography, portraits, product shots, open weights, price, and API access can change the right choice.
Scope
Image model or AI image generator?
An image model is the underlying generation system, such as GPT Image 2, Nano Banana 2, or FLUX.2 Max. A generator is the app around one or more models, including its interface, subscription, editing workflow, storage, and usage rights.
Use this page to compare model performance. Use the generator guide to choose a complete product.
Decision criteria
How to choose the right image model
- Task
- Keep text-to-image generation and image editing separate. A model can be strong at one and average at the other.
- Visual quality
- Check overall arena results as well as relevant categories such as typography, portraits, realism, and commercial design.
- Sample images
- Use identical prompts as evidence, but account for provider-specific resolution, quality tiers, and randomness.
- Price
- Compare the same task, resolution, and quality setting. Subscription prices are not the same as API prices per image.
- Access
- Check API availability, web-app access, regional restrictions, lifecycle status, and whether local execution is possible.
- Rights and weights
- Open weights, commercial terms, output rights, and training-data policies answer different legal and technical questions.
Popular model matchups
The image model comparisons people search for
This database complements our AI image generator guide. That article compares apps and subscriptions. This page compares the underlying image models and benchmark configurations.
Model selection
28 important models and configurations
Quality tiers such as Max, Pro, and High remain separate when an arena reports distinct scores and prices. This avoids misleading comparisons between different runtime tiers.
OpenAI
GPT Image 2GPT Image 1.5
Reve
Reve 2.1
Nano Banana 2Nano Banana ProNano Banana 2 Lite
Microsoft AI
MAI-Image-2.5MAI-Image-2.5 Flash
ByteDance
Seedream 5.0 ProSeedream 4.0
xAI
Grok Imagine Image Quality
Black Forest Labs
FLUX.2 MaxFLUX.2 FlexFLUX.2 ProFLUX.2 Klein 9B
Alibaba
Qwen Image 2.0 ProWan 2.6 Text to Image
HiDream
HiDream-O1-Image-1.5HiDream-O1-Image
Krea
Krea 2 Medium TurboKrea 2 Large
Luma Labs
Luma UNI 1 Max
Recraft
Recraft V4.1 Utility
Ideogram
Ideogram 4.0
Tencent
HunyuanImage 3.0 Instruct
Adobe
Firefly Image Model 5
Midjourney
Midjourney V8.1
Stability AI
Stable Diffusion 3.5 Large
Original sample images
Four fixed prompts under one documented protocol
Every run uses the same archived prompt, one request, and the first returned image. We archive aspect ratio, display size, source, model identifier, and generation time. Provider resolution, quality tier, randomness, and prompt processing can still differ. API runs also include a hash of the complete request.
Product photography
Materials, reflections, and fine details
Character and detail
Anatomy, clothing, and spatial consistency
Infographic
Numbers, labels, and visual organization
Central database
Models, prices, arena scores, and sources are stored in one structured dataset.
Two separate arenas
Text-to-image generation and image editing remain separate tasks.
No cherry-picking
One request per prompt, always the first result, with no later quality selection.
Visible data gaps
Missing measurements are never replaced with estimated bars. Missing sample runs are identified as well.
Image model comparison methodology
Arena data retrieved on August 16, 2026. Arena prices refer to 1,000 images at 1,024 × 1,024 pixels using the documented standard settings.
FAQ
Questions about the image-model comparison
What does the comparison measure?
It combines separate arena results, task-specific prices, technical specifications, availability, and documented samples without creating one artificial overall score.
Are all sample images perfectly comparable?
No. Prompts and selection rules are standardized, but provider resolution, quality tiers, prompt processing, and randomness can still differ.
Why are model variants listed separately?
Max, Pro, High, and other tiers can have different scores, prices, and runtime settings. Combining them would hide meaningful differences.
How current are the benchmarks?
The stored arena snapshot was retrieved on August 16, 2026. Model facts and samples carry separate source dates.