Image Model
Explore image models for fast generation, precise editing, and consistent visual results.
Nano Banana
Google's Nano Banana family focuses on fast image generation and editing, with strong prompt understanding and text rendering. The lineup ranges from the high-end 2.1 model to the lightweight 2 Lite.

Nano Banana 2.1 - Built on Gemini 3.6 Flash and positioned as a full replacement for Nano Banana 2. It adds mask-based editing, multi-turn consistency, up to 14 reference images, 1K/2K/4K output, adjustable thinking levels, Google Search grounding, and improved text and infographic layout.

Nano Banana 2 - Officially Gemini 3.1 Flash Image. A faster upgrade from Nano Banana Pro with high-fidelity output, deep prompt understanding, fast inpainting and outpainting, style and object replacement, hand and face repair, and upscaling from 512px to 4K.

Nano Banana 2 Lite - A fast, lightweight, cost-efficient version for quick generation, editing, testing, and comparison. It supports text-to-image, natural-language image editing, multi-turn refinement, and character and object consistency.
GPT Image
OpenAI's GPT Image family focuses on strong text rendering, realistic detail, and reliable editing across multi-turn workflows.

GPT Image 2 - Strong text rendering for labels, posters, menus, and signage; more realistic skin, materials, lighting, and depth; better UI, product, and mockup visuals; improved prompt understanding and consistency.

GPT Image 2.5 Sunburst - The quality-focused base variant of GPT Image 2.5. Offers sharper details, more natural lighting, richer textures, faster generation, precise local editing, high-fidelity reference images, consistent multi-turn editing, and better visual understanding.
Qwen Image 2.1

A Qwen-family image model that unifies text-to-image generation and editing while balancing quality, inference efficiency, and cost. Its visual generation component has 7B parameters.
Native support for generating and editing transparent images.
Accepts 1–10 reference images for multi-subject composition, outfit swapping, and interior layout changes.
Localized editing with colored circles, brush strokes, or masks.
Improved typography, portrait lighting, and fine details.
MOSS-Image

An open-source text-to-image model from MOSI.AI (OpenMOSS), built around accurate text rendering and controllable layouts. It uses a 5B single-stream DiT backbone pretrained from scratch.
Bilingual long-form text rendering.
Multi-element layout generation with prompt support up to 4K tokens.
Supports anime and photorealistic styles.
Krea 3

A production-focused image model for fast, reference-driven creation. It supports prompt-based generation, editing an existing visual, and combining multiple references while retaining key details.
Natural-language edits without complex manual masking.
Keeps products, characters, and text consistent.
Combines multiple references for subject, setting, aesthetic, material, and composition.
Targeted regional edits and native 1K/2K output.
Z Image Turbo

A fast image-generation model focused on realistic quality and efficiency. It uses 6B parameters and 8 NFEs and is designed for quick, affordable generation.
Realistic, high-aesthetic output with lifelike textures and details.
Bilingual Chinese and English prompt support.
Prompt enhancement and reasoning with world knowledge.
Sub-second inference latency.
Flux.3 Image Klein

A compact 5.6B-parameter model from Black Forest Labs that combines advanced generation and editing controls. It is available as a 16-step Distilled Klein for speed and a 50-step Base Klein for customization.
Output up to 4K in supported workflows.
Combine up to 10 reference images for one result.
Bounding-box layout control and pixel-preserving edits.
Typography and multilingual text support.
The Base version supports fine-tuning and LoRA training.
Seedream 5.0 Pro

ByteDance's advanced image generation and editing model, designed for controlled creation, precise editing, structured composition, and professional visual output.
Professional-grade image generation.
Precise local editing and layer-like image control.
Strong prompt understanding and improved typography and text rendering.
Multi-reference image creation.
SeaArt Film Edit 4.0

An image generation and editing model in the SeaArt Film family. It supports instruction-based editing, keeps subjects, settings, and branding consistent across edits, and accepts multiple images for combining, style transfer, swapping, and derivative edits.
Edit with text commands to modify, add, or reference elements.
Preserves details and style across multiple generations and modifications.
Accepts multiple images for combining, style transfer, and derivative edits.
Supports enhanced resolution up to 4K.
Last updated