Make images without learning prompts: hands-on tutorial for generating images by chatting with the assistant (zero barrier, with billing notes)

Interested in AI image making, but prompt tutorials make your head spin, and you do not want to memorize a dozen tool entries: go straight to the assistant. It is the site's flagship feature, positioned as fully zero-barrier. Describe the picture you want in everyday words inside the chat, and the assistant understands and directly calls the site's text-to-image, image-to-image, video and audio capabilities to deliver finished work, then keeps revising along the conversation. This tutorial covers no parameters and no jargon. Follow it and you will hold your first image.

What the assistant can do

In one sentence: say the idea out loud, and images, video and music arrive in one step. No tools to learn, no switching back and forth between pages.

Type to generate: describe the scene in words and the assistant generates directly, e.g. "orange cat sunbathing on a windowsill, soft morning light, shallow depth of field".
Bring a picture to edit: paste a photo from your phone into the chat and say "change the background" or "make it illustration style", and it edits your picture.
Follow up until satisfied: if the first version disappoints, just say "revise it" or "try another style". Context carries over, no need to restate the request.
Beyond images: the same entry also makes short videos, voiceovers and music, plus multimodal Q&A that understands images and text together.
External Agents welcome: an MCP endpoint is also provided, so external Agents like Cursor call the site's capabilities through the same toolset.

Step 1: Open the assistant and say the idea

The entry is the assistant. After opening, write the request straight into the input box. No model selection is needed upfront. Models are configured server-side; you only need to describe the picture clearly.

The beginner formula for a good request is "subject + scene + feeling", plus one line on what it is for:

Writing "an orange cat sunbathing on a windowsill, soft morning light, shallow depth of field" beats writing "cat" by far.
Stating the use works too: "make a Moments cover" or "draw a birthday card picture for my baby", and the assistant delivers to finished-work standards.
When unsure, do not write it perfectly in one go: give a rough first version, follow up after seeing the image, and two or three iterations usually land it.

Step 2: Click options when you see them, no typed guessing

When a request has several possible approaches, the assistant does not make you type guesses. It pops up a few tappable options instead. Tap one, and the input box stays free for extras. If you say "make an avatar", it may ask realistic or cartoon. Tap one. When you see options, tap them; do not answer with letters or numbers in the message body, only button taps count.

Step 3: Keep talking when unsatisfied, no need to restart

This is where the assistant saves the most effort. Three ways to revise:

Say it directly: comment on the fresh image, "brighten the background a bit" or "switch to a warmer tone", and the assistant continues from context.
Paste an image to pin down: paste the image to revise into the chat (referencing images from earlier rounds is also supported) and say "keep this composition, make it a night scene". That image gets revised, nothing else moves.
Circle a part to fix: for a small patch (removing an extra hand, changing a clothing color), open the box-select edit canvas, circle the area, write one line on what to change, and after confirming only the circled part is repainted while the rest stays.

There is no parameter panel anywhere, no professional vocabulary. Every unsatisfied remark works the same, "one more version" or "another style", until the image you want appears. Then download it.

Attachments and limits

Images, audio and video can all be attached to a question together. A single file must not exceed 8MB; compress oversized files first.
Images from earlier rounds are not resent to save consumption. To reuse an old image, take it from the in-chat references, or attach it again.
Doodles and hand sketches can be pasted too: ask to "turn the doodle into a polished illustration", and the assistant opens the doodle canvas for hand drawing, then generates from what was drawn.

Billing notes

Pay per use, no card binding, no subscription, no monthly fee. A minimum of 10 credits produces one image (as low as ¥0.1/image). Prices vary slightly across capabilities, but the page always shows this run's consumption before generating, and deduction happens only after confirmation. Signing up gives 50 credits, enough to run this tutorial's full flow several times over. Top-up credits never expire. See Pricing for prices and FAQs.

Three beginner pitfalls

Writing half a sentence: descriptions like "give me a picture" or "make it nicer" most likely come back off-brief. Spend ten extra seconds on subject, scene and feeling, and rework rounds drop by half.
Answering options in the message body: when tappable options appear, always tap the button. Typing letters in the message body may be taken as extra instructions.
Studying each tool's tutorial first: while the request is still unshaped, talk it through in the assistant first. Once you know what you want, go to single-purpose tools like text-to-image for batch runs. Reversing the order invites rework.

Start now

Assistant: talking to generate, editing with images and multi-round follow-ups all live here, and the starting point of this tutorial.
Text-to-image: already know what you want and just need one image fast? Write the description and generate.
All tools: twelve tools at a glance, pick by scenario once you are comfortable.