NS-Image: our image model for 0.25 tokens

Briefly about the main thing (BLUF)

NS-Image is our own image model and costs a quarter of a token per image. He draws according to the usual Russian request, and the most interesting thing is in editing the finished pictures. We tell you what the model can do, how much it costs and where it is not exported.

A quarter of a token per image. This is how much it now costs to generate NS-Image, our own image model, which we rolled out to the site this week.

To make it clear how much this is: the free tokens that drop into your account every day are enough for about three dozen pictures. Every day. Without paying a penny.

What is she drawing?

It produces 2048x2048 - this is enough for the cover of an article, a banner, and a product card. Holds details: the texture of fabric, reflections on glass, grain on wood. A long description of three sentences is read in its entirety, but the first five words are missing from it, as happens with simpler models.

The aspect ratio is directly selectable generator: horizontal for the cover, vertical for the story, square for the avatar. You can leave “auto” - then the model itself will decide according to the description.

Editing is where it gets interesting

Honestly, now everyone has generation from scratch. But not everyone knows how to remake THEIR picture with words, and NS-Image does it surprisingly accurately.

It works like this: attach your image, write in plain text what to change. Replace background. Change a cloudy morning to a warm evening. Add an item to the table. Remove unnecessary things. The model redraws exactly what is asked for and does not redraw the rest of the frame.

Here is a couple of “before/afters” - we didn’t select anything, this is the first run. First the original frame:

And now the same mug and the same table, but using the text command “replace the cloudy morning outside the window with a warm evening city with lights, add the light of a table lamp and a couple of eucalyptus leaves nearby”:

The mug remained the same. The table is the same. Exactly what was asked for changed. Previously, this took an hour in the editor - now it takes a minute and a quarter of a token.

You can attach up to ten pictures at once - if you need the model to pick up the style or collect several objects into one scene.

What is the price

0.25 tokens per generation. For comparison: one short video costs several tokens, that is, for the price of one video you get three to four dozen pictures.

There is no need to pay separately for the model - it works from the general balance, as chat And video. Top up once, spend on whatever you want.

Where it is not taken out

Without rose-colored glasses. The model still draws the text in the picture crookedly - if you need an inscription, it is better to add it later in any editor. Small details in the background sometimes float. And like all image models, it doesn’t always hit the mark the first time: sometimes the description has to be rewritten in other words.

But at a price of a quarter of a token, redoing it three times is less than one token. So there is no shame in experimenting.

Open the generator with NS-Image already selected and try it on your picture - the most interesting thing is the editing.

Frequently asked questions (FAQ)

Question: How to get the best result from a neural network?
Answer: Use detailed prompts (descriptions) in English, set the style and details of the scene.

Question: Can these materials be used for commercial purposes?
Answer: Yes, the generated content is entirely yours.