Documentation

Models and options

The built-in image generators, with their resolutions, aspect ratios and model-specific controls.

Every model exposes its own controls, because the services genuinely differ. Google and ByteDance models think in aspect ratios and a resolution tier; OpenAI models think in fixed pixel sizes; the OpenRouter models take an aspect ratio and set the resolution themselves. Monet shows you whichever set applies to the model you picked.

The table lists the models Monet ships with. It is not necessarily the whole list: providers you add yourself pull their own models in, and Settings → Models keeps the full, current registry — where you can hide a model from the picker, rename it, choose its colour, or set its default options. See Settings.

ModelServiceResolutionNotes
Nanobanana Progemini-3-pro-image-previewGoogle Gemini1K, 2K, 4K10 aspect ratios, from 1:1 to 21:9. Defaults to 2K.
Nanobanana 2gemini-3.1-flash-image-previewGoogle Gemini512, 1K, 2K, 4K14 aspect ratios, adding banner and panorama shapes such as 4:1 and 8:1. Defaults to 1K.
GPT Image 2gpt-image-2OpenAIUp to 3840×2160Fixed sizes rather than ratios. Quality, background, output format (PNG, JPEG, WebP) and moderation level.
GPT Image 1.5gpt-image-1.5OpenAIUp to 1536×1024Quality, background (including transparent) and output format.
Seedream 5.0 Proseedream-5.0-proByteDance, via fal.ai1K, 2K14 aspect ratios and reference-image editing. Defaults to 2K.
FLUX.2 Problack-forest-labs/flux.2-proOpenRouterSet by the model10 aspect ratios, from 1:1 to 21:9. No resolution tiers; defaults to 1:1.
Riverflow V2.5 Prosourceful/riverflow-v2.5-proOpenRouterSet by the modelThe same 10 aspect ratios as FLUX.2 Pro, defaulting to 1:1.

Shared options

  • Aspect ratio. Square, portrait, landscape and widescreen shapes, up to ultra-wide 21:9. Nanobanana 2 and Seedream add extreme banner shapes such as 4:1, 1:4, 8:1 and 1:8.
  • Resolution. 512 for quick drafts, 1K for everyday work, 2K for detail, and 4K where the model supports it. Larger images take longer and cost more.
  • Reference images. Attach existing images to steer the result. This is what keeps a subject recognisable between runs, so it matters most when you are regenerating a layer.

OpenAI-only options

  • Quality. Auto, high, medium or low, trading detail against speed.
  • Background. Auto or opaque, and on GPT Image 1.5, transparent, which is handy for layers you plan to composite.
  • Output format. PNG for lossless and transparency, JPEG for smaller files, or WebP.
  • Moderation. Auto or low, controlling how strict OpenAI’s content filtering is.