Best AI image models

Image-generation models are tracked separately from text LLMs. This page is distinct from /best/vision, which ranks multimodal models that understand images rather than generate them. Compare image generation and editing models across GPT Image, Gemini Nano Banana, FLUX, Ideogram, and related creative providers. Distinct from /best/vision, which ranks multimodal image-understanding LLMs.

Last refreshed 2026-11-05. Next refresh: weekly.

How we rank

Image generation models are tracked separately from text LLMs. This page lists the active image generation models we track.

Eligibility — Active image generation models we track. Deprecated and pre-release models are left out. Distinct from /best/vision, which ranks multimodal LLMs that understand images.
Cross-links — A model's name links to its model page when we have one, with API pricing and provider routes where available. Models without a model page are listed by name only.

Verdict

Use Nano Banana 2.1 for image generation today.

Muse Image is the runner-up: 1325 vs 1269 on Arena score.

Researched 35d agoWhy this pick Methodology

1stTop pick

Researched 35d ago

Nano Banana 2.1

Arena score: 1325
Output (from): $7.50 / 1M

Try on provider Model detail

2ndShortlist

Researched 23d ago

Muse Image

Arena score: 1269
Output (from): —

Try on provider Model detail

3rdShortlist

Researched 33d ago

Qwen-Image-3.0-Pro

Arena score: 1254
Output (from): —

Try on provider Model detail

Ranked by LMArena text-to-image overall arena score (8)

Rank	Model	Lab	Arena score
1	Nano Banana 2.1	—	1325
2	Muse Image	—	1269
3	Qwen-Image-3.0-Pro	—	1254
4	Seedream 5.0 Pro	—	1250
5	Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image)	—	1249
6	Wan 2.6	—	1138
7	Qwen Image	—	1057
8	Recraft V3	—	1022

Scored but unranked (preview or retired) (8)

Model	Lab	Arena score	Status
MAI-Image-2.6	—	1329	Preview
MAI-Image-2.5	—	1253	Preview
Nano Banana Pro (Gemini 3 Pro Image Preview)	—	1232	Preview
MAI-Image-2	—	1183	Retired
Grok Imagine Image Pro	—	1161	Retired
GPT Image 1	—	1115	Retired
Gemini 2.0 Flash Image Generation	—	976	Retired
DALL-E 3	—	967	Retired

Scored but unranked (not eligible to rank) (29)

Model	Lab	Arena score
GPT Image 2.5 Sunburst	OpenAI	1427
GPT Image 2.5 Flare	OpenAI	1401
GPT Image 2	OpenAI	1381
Reve 2.1	—	1300
Reve 2.0	—	1268
Nano Banana Pro (Gemini 3 Pro Image)	—	1246
GPT Image 1.5	—	1236
Qwen-Image-2.1	—	1225
Ideogram 4.0	—	1200
Qwen-Image-2.0-Pro	—	1191
Grok Imagine Image	xAI	1169
Cosmos 3 Super Text2Image	—	1151
Hunyuan Image V3	—	1151
Seedream 4.5	—	1149
Imagen 4 Ultra	—	1146
Seedream 5.0 Lite	—	1140
Imagen 4	—	1128
Recraft V4	—	1115
GPT Image 1 Mini	—	1109
FLUX.1 Kontext [pro]	—	1059
Imagen 3	—	1057
Ideogram 3.0	—	1049
Luma Photon	—	1036
Runway Gen-4 Image	—	1024
FLUX.1.1 [pro]	—	1016
Lucid Origin	Leonardo	1013
Ideogram 2.0	—	1013
GLM Image	—	1012
Stable Diffusion 3.5 Large	—	938

Unranked (no LMArena score) (22)

Ordering and coverage: Listed alphabetically; these models have no score on this board and carry no rank.

Model	Lab	Notes
Dreamshaper 8 LCM	Lykon	Fine-tuned Stable Diffusion 1.5 model optimized for improved photorealism. Uses LCM (Latent Consistency Model) for faster inference.
Flux	Black Forest Labs	High-quality image generation with fast inference speed; available in open-source and professional versions.
FLUX.1 Schnell	Black Forest Labs	Fast text-to-image generation model using rectified flow transformer architecture. Optimized for quick inference while maintaining high-quality outputs.
FLUX.2 Dev	Black Forest Labs	Advanced FLUX model generating highly realistic and detailed images with multi-reference support. Designed for professional-quality image synthesis.
FLUX.2 Klein 4B	Black Forest Labs	Ultra-fast distilled rectified flow transformer for rapid image generation and editing. Optimized for real-time applications with minimal latency.
FLUX.2 Klein 9B	Black Forest Labs	Ultra-fast distilled rectified flow transformer for text-to-image generation and editing. Delivers state-of-the-art quality for interactive workflows, real-time previews, and latency-critical applications.
GPT Image 2.6 Flare	OpenAI	OpenAI's fast, interactive GPT Image 2.6 model for image generation and editing. Accepts text and image inputs with an adjustable complexity setting, markup-guided edits, SVG output and animated GIF generation.
GPT Image 2.6 Sunburst	OpenAI	OpenAI's detail-oriented GPT Image 2.6 model for image generation and editing. Accepts text and image inputs with an adjustable complexity setting, markup-guided edits, SVG output and animated GIF generation.
Kling IMAGE 3.0	Kling	Kling IMAGE 3.0 available via Runware API. Image generation - $0.028 per generation (1K & 2K).
Kling IMAGE O3	Kling	Kling IMAGE O3 available via Runware API. Image generation - $0.028 per generation (1K & 2K).
LLaDA-Image	InclusionAI	InclusionAI's unified open image generation and editing base model. Apache 2.0 weights on Hugging Face (inclusionAI/LLaDA-Image). Distinct from the LLaDA2.2 agentic diffusion language-model family.
LLaDA-Image Turbo	InclusionAI	Faster distilled variant of LLaDA-Image for unified open image generation and editing. Apache 2.0 weights on Hugging Face (inclusionAI/LLaDA-Image-Turbo).
MAI-Image-2.6-Flash	Microsoft AI	Faster, lower-cost variant of MAI-Image-2.6 for high-throughput image generation and editing on Microsoft Foundry.
Midjourney V6	Midjourney	High-quality AI image generation with strong artistic control and style customization through Discord bot interface and web platform.
MiniMax-M2.5	MiniMax	MiniMax-M2.5 available via Runware API. Image generation - $0.27 per generation (token-based).
SD v1.5 Img2Img	Runwayml	Image-to-image variant of Stable Diffusion v1.5. Generates photo-realistic images from input images and text prompts with customizable parameters and mask support.
SD v1.5 Inpainting	Runwayml	Inpainting variant of Stable Diffusion v1.5. Fills masked image regions based on text prompts. Specialized for content-aware image editing.
SDXL Lightning	ByteDance	Lightning-fast generation of high-quality 1024px images in just a few steps. Open-source variant of SDXL optimized for rapid inference.
Silenz 1 (Leonardo)	Leonardo	Leonardo's advanced image generation model with strong prompt adherence and versatile style capabilities.
Stable Diffusion 3	Stability AI	Open-source image generation model with improved prompt adherence, multi-subject handling, and text rendering capabilities.
Stable Diffusion XL Base 1.0	Stability AI	Open-source diffusion-based text-to-image generative model supporting high-resolution image generation and modification based on text prompts.
Z-Image	Z Ai	Z-Image available via Runware API. Image generation - $0.0045 per generation (1024x1024).

Best multimodal / vision LLMs - image understanding and document QA.
Editorial image picks - qualitative tiers and Editor's Choice for creatives.
OpenAI API, Google AI Studio, Ideogram API, fal.ai

Best AI image models

How we rank

Use Nano Banana 2.1 for image generation today.

Nano Banana 2.1

Muse Image

Qwen-Image-3.0-Pro

Ranked by LMArena text-to-image overall arena score (8)

Scored but unranked (preview or retired) (8)

Scored but unranked (not eligible to rank) (29)

Unranked (no LMArena score) (22)

Related pages