Image Creator — Overview

Generate AI images from text prompts — what they cost, and what the download looks like.

What is the Image Creator?

The Image Creator turns text prompts into images. Pick a model, a style and a format, then generate — every image saves to your gallery automatically. Results can be refined afterwards with editing and inpainting.

Three modes

The Image Creator is three screens behind one page. You switch between them with the Generate / Chat / Gallery switch in the page header, at every screen size. The sidebar has the same three buttons:

Generate is the one you land on. Gallery carries the number of pictures you have made beside its name.

Generate — the full form: prompt, model, style and format, with a live cost summary. If the form is more than you want, there is a guided wizard that asks one thing at a time instead.
Chat — describe what you want in a conversation and iterate on results turn by turn. Chat work is organised into image projects — see Image Projects.
Gallery — browse, download and manage everything you have created.

The Wizard generates on its last step

The Wizard button on the Generate form opens the guided wizard. It asks five things, one at a time: the main subject (with quick ideas to pick from), the style, where it is (optional), the mood, and on the last step, Review & Generate, any extra details.

Your Kit sits right under the subject box: tap a saved person, brand, place or style to put its @tag in the box, and tap it again to take it out. On every later step a thin In this picture row keeps those tags on screen, each with a ✕, and Edit prompt jumps back to the box. When the picture is made, the people, brands and places you tagged are sent as pictures too, the same as attaching them on the form.

The last button, Generate Image, fills in the form and makes the picture in one press. There is no second confirmation, so the credits are spent on that press. Check the cost summary on the form first if the price matters to you.

Chat asks before it draws — but the conversation itself is billed

In Chat mode your message does not go straight to the image model. The assistant writes a prompt from what you said and shows it back to you with three buttons — Generate, Edit and Cancel — so you see the wording before a picture exists. No image is made, and no image is charged for, until you press Generate.

Read that precisely, because it is the part people get wrong: what waits for Generate is the picture. The conversation is a normal AI reply and is billed like one, every turn, the moment the assistant finishes writing back to you. Talking through five wordings before you generate anything costs five chat turns — not much each, but not nothing.

The amber Auto-generate button at the top of the chat turns the confirmation step off: with it on, your message is used directly as the image prompt and the picture is made immediately. It is the faster way to work and the easy way to spend credits without meaning to.

Most downloads carry a visible watermark

Pictures you download from the Generate screen, the Gallery and a new image chat all have a "VyroAI" watermark drawn into the bottom-right corner before the file is saved. There is no clean-download switch anywhere, on any plan.

Three places still hand you the picture untouched: the Download button inside a saved image project's chat, the Layers tab, and the download icon under a picture in AI Chat (skins included). A download whose watermark cannot be drawn also falls back to the plain file. Do not plan around that — it is an inconsistency, not a feature, and it can close at any time.

The watermark matters most for Minecraft textures: it is sized against the width of the picture, so on a 16×16 texture it covers the art. Generate at a larger size and scale down.

Choosing a model

The model dropdown lists the available image models, and one of them is preselected for you — there is no badge marking which is the default, so the one already showing when you arrive is it. The list always reflects what is currently enabled, so it can change over time.

A reference photo is free, and it now keeps the face

Attach a reference image and, on a GPT Image model or on Nano Banana, the picture really is sent to the model — it is what keeps the face or the object you gave it. A saved character works the same way: pick her and her frame travels along as the reference. This is new since 5 September 2026. Before that the upload was accepted and then dropped on the floor, which is why a saved character used to come back as a stranger.

DALL-E cannot take a reference at all. Rather than ignoring it quietly, the Generate screen stops before anything is sent — "This model cannot take a reference photo", with "Pick a GPT Image model or Nano Banana to use the picture, or remove it. Nothing was charged." underneath. Switch model and generate again.

Attaching a reference costs nothing extra on any model. The interface used to add 10 credits for it while the server never charged them; that phantom surcharge was removed on 29 August 2026, from the generate path and the edit path both. If you are following an older guide that mentions a Reference Image line in the cost summary, it is gone.

Inpainting is still the tool for changing part of a picture you already have: you paint a mask over the area and only that area is redrawn. A reference photo is for the opposite job — a brand-new picture that keeps a face. See Image Creator — Edit & Inpaint.

Minecraft skins, with a 3D preview

One of the styles makes a real, wearable Minecraft skin rather than a picture of one — a proper skin file laid out the way Minecraft expects, which you download and upload to your account.

What makes it worth finding: the result is shown as the flat skin sheet and, beside it, a rotating 3D character, so you can see the back of the head and the sides of the arms before you commit to it. You describe the character you want — there is no template picker on the screen today.

This works in AI Chat too — ask for a skin there and you get the same rotating preview in the conversation.

What the panel shows while a picture is being made

While the image renders, the preview panel shows the named steps the job is going through, each with its own timing, an estimate for the model you picked and a clock counting up, plus a recap of what you asked for and the credits reserved.

It is the same card the Video Studio uses, so the wait looks identical in both tools.

What the Ready card offers you

When the picture lands, the panel becomes a Ready card: the image, how many seconds it took and what it cost in credits, and a row of things to do next.

Download saves the file. This is the Generate-screen download, so it carries the VyroAI watermark described above.
Use in Video Studio hands the picture straight over: you arrive in the Video Studio with it already sitting in the first reference slot, with From an image picked, and a line telling you to describe the motion and generate. Nothing is generated and nothing is charged by pressing it — it is a handover, not a job. Note it moves the picture, not your prompt.
Make another simply clears the finished card away and leaves the form exactly as you set it, so you can change one word and go again. It costs nothing on its own.
Same prompt, new take does generate — immediately, on the settings already in the form, at the usual price. There is no confirmation between the click and the spend.
The bin deletes the picture.

When it fails instead

A failure gets the same card in a different colour, headed Could not make this image, with the seconds it ran. It prints the real reason rather than a bare failed, and says in plain words what happened to your credits. When nothing was taken, 0 credits shows beside the seconds.

One failure cannot say that: Could not reach the image service. The server may be busy or restarting — wait a minute and try again. The connection dropped before the image service answered, so the app cannot tell whether your credits were taken, and it does not pretend to know. The card then ends with Check your credit balance — if credits were taken without a picture, report it via Help. The same sentence appears after any other error the app cannot explain. Look at the credits in the top bar: if they went down and no picture arrived, contact support on /support with the time it happened. See Refunds When Something Fails.

Two buttons sit under the card: Try again, which re-sends the same prompt and settings, and Check status page, which opens VyroAI's status page. Use the second one when the reason sounds like the service rather than your prompt — several failures in a row usually means an outage, and trying again into an outage just fails again.

Pick a specific style rather than leaving it generic — it makes the biggest difference to how consistent your results look. See the Styles article for what each one is best at.

Also in Image Creator

Next in Image Creator: Styles
Making the picture: Formats & Enhancement — Square, landscape, portrait or a social-media preset: how to pick an image format, make it your default, choose SD or HD, and what each try costs.
Making the picture: Fancy Text — Type once, pick one of 32 styles, and copy it into Discord, a bio or a username. Free, with no AI involved.
Changing one you already have: Edit & Inpaint — Inpaint is the one that really edits your image. Read this before using Edit with Prompt.
Changing one you already have: Splitting an Image into Layers — The third edit tab: cut a picture into its parts, then hide or delete one without asking the model to invent what was behind it.

1 of 12 in Image Creator · Last checked 24 September 2026 · Previous: Reporting Content