Nano Banana 2 is Martini's default image model and the best all-rounder for most users. It supports both text-to-image and image-to-image editing, accepts up to 10 reference images, outputs at up to 4K resolution, and costs as little as 10 credits per image. Where Midjourney prioritizes aesthetics and FLUX prioritizes prompt fidelity, Nano Banana 2 balances both — producing photorealistic, detailed images that closely match your description.
Nano Banana 2 offers three resolution tiers: 1K (10 credits, fine for social media and web thumbnails), 2K (15 credits, the default — good for most uses including blog headers and presentations), and 4K (30 credits, for print-quality art and detailed illustrations). The detail jump from 1K to 4K is dramatic — fine textures like fabric weave, skin pores, and leaf veins become visible at 4K.
Connect up to 10 images to the node as references. Nano Banana 2 can use these for image-to-image editing (change backgrounds, modify elements) or style transfer (apply the aesthetic of one image to new content). This is especially powerful for creating art series with consistent visual language.
Nano Banana 2 excels at rendering specific materials and lighting. Instead of "a castle," write "a weathered limestone castle at golden hour, warm directional light casting long shadows, moss growing in mortar joints." Camera terminology works well too: "85mm lens, shallow depth of field, shot on medium format" pushes results toward photorealism.
Toggle Web Search to "On" (+3 credits) when referencing specific artists, art movements, or contemporary visual styles. This helps the model access up-to-date references beyond its training data — useful for prompts like "in the style of Studio Ghibli" or "Bauhaus poster design." For generic prompts, leave it off to save credits.
Material-focused photorealism — Nano Banana 2 renders marble, water droplets, and plant textures with exceptional fidelity. The camera terminology cues ("medium format", "tonal range") guide the model toward photographic rendering.
A hyperdetailed sculpture garden at golden hour, marble statues draped in morning glory vines, dew drops catching prismatic light, shot on medium format camera with extreme depth and tonal range
Fantasy concept art at 4K — even with an imaginative subject, Nano Banana 2 applies realistic material physics: the stone has weight, the waterfalls have correct flow dynamics, and the volumetric light scatters naturally through clouds.
Digital concept art of a floating sky city, massive stone platforms connected by rope bridges, waterfalls cascading into clouds below, dramatic volumetric lighting at sunset
4K costs 3× more than 1K (30 vs 10 credits) but the detail increase is nonlinear — for print or zoom-in use cases, 4K is always worth it. For social media posts, 1K is sufficient.
Web Search adds only 3 credits but dramatically improves results when referencing specific artists, films, or contemporary design trends. It's the best value parameter on the platform.
PNG output (default) preserves full quality for further editing. Switch to JPEG only for final web delivery where file size matters.
Nano Banana 2 supports the widest range of aspect ratios on Martini — including extreme formats like 1:8 and 8:1 for banners, panoramas, and vertical story formats.
Nano Banana 2 is Martini's default model for good reason: it's the most versatile, supports the widest aspect ratio range (including extreme 1:8 and 8:1 formats), accepts up to 10 reference images, and offers three resolution tiers. It generates 1 image per request at consistent, high quality. The model is less "artistic" than Midjourney and less precise at text rendering than Ideogram, but for the majority of AI art tasks — illustration, concept art, character design, environment art — it produces excellent results at the lowest per-image cost.
Connect Nano Banana 2 with other AI models on Martini's infinite canvas. No GPU required — start free.
Get Started FreeBlack Forest Labs
FLUX.2 is the go-to model when you need your prompt followed precisely. Unlike Midjourney, which interprets and embellishes, FLUX.2 renders exactly what you describe — every element, spatial relationship, and style directive is respected. This makes it the strongest choice for concept art with specific compositions, multi-subject scenes, and illustrations that need to match a creative brief.
View guideMidjourney
Midjourney v7 is the most aesthetically opinionated image model available. Where other models faithfully reproduce your prompt, Midjourney actively interprets it — adding dramatic lighting, compelling composition, and artistic flair that transform simple descriptions into gallery-worthy images. This makes it ideal for concept art, illustration, and any project where visual beauty matters more than literal accuracy.
View guideIdeogram
Ideogram V3 is the only AI model that reliably renders readable text inside images. Every other model — FLUX, Midjourney, GPT Image — struggles with text accuracy, often producing garbled letters. Ideogram V3 solves this, making it the clear choice for poster art, book covers, logo concepts, infographics, and any visual design where typography is part of the composition.
View guideOpenAI
GPT Image 1.5 is built on OpenAI's language model architecture, giving it the deepest natural language understanding of any image generator. While FLUX and Midjourney interpret prompts as visual keywords, GPT Image 1.5 reads them as full sentences — understanding context, metaphor, spatial relationships, and narrative intent. This makes it the best choice for complex scenes with specific compositional requirements, abstract concepts, and multi-element illustrations.
View guide