AI Models

Google's Nano Banana 2.1 is the successor to Nano Banana 2 — Flash-level speed with sharper visual quality, cleaner typography and more consistent multi-turn edits. Outputs 1K, 2K or 4K in 14 aspect ratios including 1:4, 4:1, 1:8 and 8:1 panoramas, fuses up to 14 reference images, and can ground on Google web and image search for real-time subjects. Pick a reasoning level to trade speed for fidelity. Provide images to edit them, or leave them empty for text-to-image.

OpenAI's GPT Image 2.5 family for image generation and editing. Sunburst is the most capable model, built for edits where precision matters. Flare is the fastest for high-quality everyday images. Both accept text and reference images, support transparent backgrounds and custom sizes up to 4K, and offer five quality tiers from quick low-cost drafts to maximum detail.

Qwen-Image 3.0 generates dense, information-rich layouts in one pass from prompts up to 4.5k tokens, with crisp text rendering down to 10px across 12 languages and 20+ fonts — posters, web pages, and UI screens. The Pro variant adds images-within-images layouts and realistic simulation of web, game, and live-stream interfaces. Add up to 3 reference images to switch to editing.

xAI's Grok Imagine Image 2.0 is made for precise image generation and editing, with a designer's care for typography and layout. Dense, multi-part visuals hold together and small text comes out sharp. Start from a prompt or edit with up to 3 reference images, in 15 aspect ratios from 9:20 portrait to 5:2 wide, at 1K or 2K, with two quality tiers.

Meta's agentic image model that generates and edits images from text and reference images. Before it renders, it can search the web for visual references and current facts and compute structured layouts for infographics and charts, so real places, products, brands and data come out accurate.

Luma's Uni 1 image model from the Luma Agents API. Generates 2K-resolution images from text, or edits an existing image with natural-language instructions — background replacement, object modification, lighting changes, style transfer — while preserving the parts you don't mention. Up to 8 reference images guide style and content in either mode, web search grounds prompts that mention real-world subjects, and a manga style preset is included. Uni 1 Max is a drop-in higher-quality variant.

Ideogram 4.0 image generation and remixing model with best-in-class text and typography rendering. Generates from text or remixes a provided image guided by the prompt. Prompts are automatically enhanced with Magic Prompt, and images render at 2K resolution across 23 aspect ratios with Fast, Balanced, and Best quality tiers.

OpenAI's GPT Image 2 generates and edits images with close instruction following and dense, legible text in many languages. Start from a prompt, or edit with up to 16 reference images and an optional mask that marks the area to change. Choose square, portrait, landscape, 2K or 4K sizes, transparent backgrounds, three quality tiers, and PNG, JPEG or WebP output.

ByteDance's Seedream 5.0 image family for generation and editing. Lite is fast and affordable with up to 4K output. Pro is the flagship — deep-thinking prompt understanding, native text rendering in 14 languages, and region-precise edits, with up to 10 reference images. Provide images to edit them, or leave them empty for text-to-image.

FLUX.2 [max] is the highest-quality model in Black Forest Labs' FLUX.2 family, made for final assets and the hardest edits. It follows prompts closely and keeps faces, products and styles consistent across up to 8 reference images, at any width and height up to 2048×2048. Provide reference images to edit or combine them, or leave them empty for text-to-image.

OpenAI's GPT-5.5 is built for complex professional work: writing, research, data analysis and code. It gives warm, readable answers without heavy prompting, matches GPT-5.4's speed at a higher level of intelligence, and has a context window of over 1M tokens. Choose a reasoning level, add web search for current facts, or request structured JSON output.

Anthropic's Claude Opus 4.8 builds on Opus 4.7 for long, multi-step work, knowledge tasks and vision. It is more likely to flag what it is unsure of than to make unsupported claims, and it spends time thinking only when a request needs it. Use it with 1M tokens of context, five effort levels, web search, image, text and PDF attachments, and structured JSON output.

Anthropic's Claude Opus 4.7 is tuned for complex reasoning and for following instructions to the letter. It sees fine detail in images, works across 1M tokens of context, and produces polished documents and analyses. Pick an effort level to trade speed for depth, attach images, text files or PDFs, add web search for current facts, or request structured JSON output.

GPT-5.4 pro is the version of OpenAI's GPT-5.4 that uses more compute to think harder on tough problems that need deeper reasoning. A reply can take several minutes. It has a context window of over 1M tokens and is especially strong at persistent web research for hard-to-find information. Choose a reasoning level, add web search, or set a system message.

GPT-5.4 mini brings the strengths of OpenAI's GPT-5.4 to a faster, more efficient model for high-volume work. It is at its best on clearly structured tasks and takes instructions literally, at a lower cost than the full model. It offers a 400K-token context window, adjustable reasoning, web search and structured JSON output.

Anthropic's Claude Opus 4.6 is built for deep reasoning and long, multi-step work across large sets of documents. It keeps up to 1M tokens in context, enough for whole reports, contracts or research files in one conversation, and writes up to 128K tokens per reply. Turn on thinking for harder problems, attach images, text files or PDFs, ground answers with web search, or ask for structured JSON output.

Anthropic's Claude Sonnet 4.6 balances speed and intelligence for everyday work: drafting, analysis, research and code, at a lower price than Opus. It keeps up to 1M tokens in context, enough for long contracts or dozens of papers at once, and follows through on multi-step tasks more consistently than earlier Sonnet models. Turn on thinking for harder problems, attach images, text files or PDFs, or add web search.

OpenAI's GPT-5.4 is a general-purpose model for professional work that moves between writing, analysis, coding and research. It has a context window of over 1M tokens, follows instructions closely, and keeps its tone steady across long answers. Choose a reasoning level, add web search for current facts, or request structured JSON output.