The current flagship
OpenAI's state-of-the-art image model (April 2026): generation and instruction-following editing in one model — no separate inpaint pipeline to learn.
Model family · Image
Give notes, not masks.
GPT Image 2 is OpenAI's current flagship for images: one model that generates and edits by instruction, holds every reference at high fidelity, and renders text you can actually read. In MML ONE it answers your canvas and storyboard — four routes in the catalog.
By OpenAI
What it's best at
One model draws it. The same model fixes it.
What GPT Image 2 does best, per OpenAI's own documentation — including the parts that stay honest.
References held at high fidelity
Every input image is processed at high fidelity automatically — identity, composition, and lighting survive targeted edits. There is no knob to get this wrong.
Text you can read
Dense copy, small lettering, posters, infographics, UI mockups — legible in-image text is a headline capability, not a lucky roll.
A 4K-class canvas
Flexible sizing up to a 3840-pixel edge, with fast low-quality drafts on one end and final high-quality renders on the other.
It knows the world
Built-in world knowledge and reasoning: period-accurate, context-rich scenes from short prompts — the model fills in the era, not just the pixels.
In MML ONE
Four GPT Image routes, ready to run.
Exactly what our catalog serves today — vendor names, quality tiers included.
GPT Image 2The direct route: aspect-ratio control with 1K, 2K, and 4K-class output.
GPT Image 2 HighThe high-quality tier — final frames and hero stills.
GPT Image 2 MediumThe middle setting — quality and turnaround balanced.
GPT Image 2 LowFast drafts for exploration and coverage passes.
As served through our channels: the direct route takes an aspect ratio and delivers 1K/2K/4K-class output; the quality tiers ride a second route. The in-app catalog states what each route delivers.
In the MML ONE flow
Where GPT Image 2 earns its place.
Frames land on the canvas and in the storyboard as versioned assets. Poster-grade lettering, instruction-driven fixes, and period-accurate stills are what you route here.
The honest part
Know before you route.
Complex prompts can take up to around two minutes, and OpenAI marks output beyond 2K as experimental — temper the 4K expectations.
Recurring characters and brand elements can drift across generations, and precise layout placement remains imperfect.
No transparent backgrounds on GPT Image 2, and every prompt and output passes OpenAI content moderation.
Public Alpha
Bring one film into the graph.
Start with a premise, a screenplay, or a folder of references. We'll set up your provider keys and walk through the first scene with you.