Knowledge base / Chatting

Images, video and speech

Three commands that produce a file rather than a paragraph. All are charged per generation rather than by length, and a failed generation costs nothing.

Generating images

?image a lighthouse at dusk, long exposure returns a picture. It arrives in the conversation like any other message, encrypted the same way, and you can save it from there.

On the standard tier this uses the built-in generator. With a Pro model selected you get the frontier generators instead.

Choosing a generator

?image models lists what is available with prices, and costs nothing. With a Pro model selected, ?image --model <name> <description> picks one: Nano Banana Pro, Nano Banana 2, Imagen 4, FLUX 2 Max, FLUX 2 Pro, Seedream 5 Pro, GPT Image 2, Grok Imagine and Recraft v4 Pro.

They differ in what they are good at rather than in quality — text rendered inside the image, photographic realism, illustration, and how literally each takes a long prompt. If one misreads a description, another usually will not.

Pinning a generator

Picking a generator in the model picker pins it. From then on every ordinary message you send is a generation with that model, so a run of pictures is a run of plain descriptions rather than the same command typed over and over. The composer says what it is about to make, and the generator takes over the model chip in the toolbar.

Because every generator but the built-in one is Pro work, pinning it selects the Pro tier too, carried by the cheapest chat model when you have not chosen one.

Check Only for this chat before picking to pin it to the conversation you are in rather than everywhere. Clear it by picking the same generator again, by switching back to standard, or with ?image off and ?video off. A message that starts with ? or ! is always left alone, so the commands still work while a generator is pinned.

Editing a picture

Attach a picture, or link one, and send ?image with what to change: ?image make the sky stormy. The generator starts from your picture rather than from nothing. The same works with a generator pinned: attach the picture and describe the change, and the composer prompts you for it.

Editing is Pro work, and it is priced per generation, the same as making a new picture. Not every generator can edit; ?image models and the picker say which ones can. If the one you pinned cannot, Nymbot hands the edit to the default editor, or to the cheapest one that costs no more, and the reply says which one made it. If every generator that can edit would cost more, it tells you rather than charging more.

Video

?video a lighthouse beam sweeping across a storm returns a short clip. It is a Pro command: every video model available is provider-hosted, so unlike ?image there is no standard-tier generator to fall back on. Select a Pro model first.

?video models lists the generators with their prices and costs nothing. ?video --model <name> <description> picks one: Veo 3.1, Seedance 2.5, Hailuo 2.3, Wan 3.0, Grok Imagine Video, Pixverse v6, LTX-2.5, Vidu Q3, FLUX 3 Video and Runway Gen-4.5, among others.

Send a picture in the same message and the generators that take a reference frame will animate it rather than starting from nothing. A clip takes longer than a picture — a render still going after a couple of minutes is given up on, and nothing is charged.

Speech

?speak <text> returns a spoken clip of up to 800 characters. Longer text is refused rather than truncated, so you never pay for half a sentence.

What they cost

Because these are flat per-generation charges, the length of your description does not change the price. See credits and pricing for how the two balances work.