Models
The models Meevra items are written for, where they run and how to prompt them.
Midjourney
Default since 2026-07-24; focuses on aesthetics, image quality and Personalization; ships the Edit Model.
Fastest Midjourney model (standard jobs about 4-5x faster); select with --v 8.1.
Anime and illustration model; more literal than earlier Niji versions.
Black Forest Labs
Production image generation and editing; up to 8 reference images via the API.
Highest-quality FLUX.2 tier; up to 8 references via the API (10 in the Playground).
Small, fast open-weight models (Apache 2.0 or FLUX Non-Commercial); up to 4 references.
Stable since 2026-10-06; Google's primary image generation and editing model; 1K, 2K or 4K output; thinking minimal, medium or high (default medium).
For the most complex visual tasks and professional text rendering; 1K, 2K or 4K.
Fastest and cheapest; 1K only; not optimised for many references.
OpenAI
OpenAI's default choice for most API applications: fast, high-quality everyday generation and editing. ChatGPT Images 2.5 is available on all ChatGPT plans, ChatGPT Work and Codex.
Most capable OpenAI image model, for precise multi-pass editing; longer generation times.
Runway image-to-video planned frame first: keyframe prompts with one style block, motion-only prompts for Gen-4.5, and a clear rule for which to fix.
Plan a 6-12 image FLUX.2 product series that looks like one shoot: reference plan, frozen product sentence, locked style block, shot list and a two-image drift check before you generate the rest.
Turn a messy list of image changes into ordered one-change edit passes, each with a keep-the-rest prompt, the details most at risk and a check, plus a stop rule before the image degrades.
A reusable dark-and-gold style block for premium product shots that keeps light, scale and camera height identical across a whole series.
Get posters, thumbnails and social cards whose text actually renders: copy cut to fit, a layout spec, a model-ready prompt, a blank fallback and a proofreading list.
Plan ad and social image variants around one test variable, with a shared lock block, safe areas per format and a pre-flight check, so the results tell you something.
Build a five-shot listing set from your real product photos with GPT-Image 2.5, with a fidelity lock and a side-by-side check so the product stays the one buyers receive.
Turn a few reference images into a calibrated Midjourney V8.2 style card: one suffix, one style sentence and a drift check the whole team reuses.
Combine product, character and scene photos into new Nano Banana images that keep identities intact, with a reference map, edit passes and an identity check.