Docs  Images

Creation

Images

Pictures from words or from a reference, edited, varied and upscaled, on the model that fits the job, with the price shown first, every version kept, and the result checked before it is called done.


Two ways in

  • Creation, Image tab. One prompt box holds your words, a start picture, the style and shape, the price, and Make (or Cmd+Enter, Ctrl+Enter on Windows). The settings panel on the right shows only what the chosen model supports.
  • Any chat. Ask "make a hero picture for the landing page" or "turn this photo into a watercolour" in any thread. The job runs in Creation's queue, the picture shows in the reply, and it lands in the Library with its versions.

Both paths use the same queue, the same price rules and the same monthly count, so a picture made in chat is never cheaper or freer than one made by hand.

Choosing the model

The model picker lists the picture models of every provider you have set up, read from each provider's own list, so new models appear without an update. Each one carries a short line on what it is good at and how it is priced. A few of the families you will see: Google's Nano Banana models, OpenAI's GPT Image, Black Forest Labs' FLUX, ByteDance's Seedream, Recraft for illustration and clean design, Qwen Image and xAI's Grok Imagine.

When you ask in chat without naming a model, Mori picks one for the job: a transparent background, text that must be spelled right, a photograph, an illustration, a draft. Name a model and that model is used, every time.

You can also type any OpenRouter model id into the picker.

Settings that matter

  • Shape and size: square, portrait, landscape and the model's own sizes, up to 4K where the model offers it.
  • How many: one, or as many at once as the model allows.
  • Style: a preset look, or none.
  • References: pictures you add, each marked Subject (keep this person, product or object) or Style (borrow this look), up to the model's own limit.
  • More settings: a seed to repeat a result, what to leave out, and which project to file it under.
  • What Mori sends shows the full text of the request before you press Make, so nothing goes out that you cannot read.

Edit, variations and upscale

Every picture in the Library has Edit, Variations and Upscale, plus a Change it in words box: "make it golden hour", "remove the car", "make it night". Ready changes are one press away.

A change always makes a new version and leaves the old one as it was. The numbered Versions strip (v1, v2 and on) lets you step back, and Where it came from lists each step that led to the picture: made, edited, upscaled, animated.

From a reference

Show Mori a picture or a link and say what you want. She works out, from your words, which of these you mean:

  • copy it exactly: the same thing, as close as possible;
  • make one like it: clearly close to it, done her way or yours;
  • vibe only: take the feel, make it your own;
  • use it as one part of something bigger;
  • just study it: describe or explain it.

She also notes what must match ("the colours", "the layout", "everything") and what must stay yours ("my product", "darker"). When she cannot tell, she asks one short question instead of guessing. When you show a reference without saying more, the result has to actually look like it.

One rule holds for every reference: she takes the design (the layout, type, spacing, colour, motion and pacing), never someone else's logos, photos, video, music or words, and she says what she took from it.

How she checks the result

A picture the model returns is looked at before it is announced. A transparent background that came back opaque is not called a cut-out, and a picture made from a reference is put side by side with it.

For work made to match a reference, two separate looks must both pass: one asks does it match on what you said matters (and, for vibe only, is it clearly its own), the other asks is it professionally good, as an art director would. Measured checks have the last word on things a model could talk past, such as colour distance. A round only passes when both looks pass, and if the looking itself could not run, the result is never reported as a pass.

When a model is busy

If a model's own provider is over capacity or rate limited (not your key, your credit or your plan), the picture moves once to the next featured model from a different company, and Mori says so in one plain line. This only happens when you did not name the model; a model you chose is never swapped.

A job never moves to a pricier route than the price you saw, never after a provider refused the request on its content, and never after money was charged.

Price, spend and the monthly count

The price shows beside Make before anything is spent, and the header keeps a running total for the session. In Settings > Spend you can set your own amount ("Ask me before a job over"): above it, the first press turns the button into Make for $x and the second press spends. With no amount set, Make runs. There is no hidden default limit.

Pictures count toward the month on any key, so the count is fair whoever pays: Free 20, Plus 300, Pro 800, Studio 2,000 and Max 5,000. A job reserves its pictures when it is queued and gives back any it did not make. At the ceiling, Mori says how many are left, when they come back on the 1st, and what the next plan makes. See Plans and credits.

Keys and providers

On a plan, pictures run on the plan's credits with no key. On your own keys, an OpenRouter key alone reaches most picture models, and Providers > Media takes your own keys for Google, OpenAI, Black Forest Labs, fal, Replicate, Stability and the others, each with a free Test, an order to try them in, and a preference for quality, cost or speed. See Creation.