AI-generated images are everywhere now. But the interesting question isn't the image. It's what did they actually type to get that?
That part usually stays private. The finished piece gets posted; the prompts and the dead ends stay on the maker's screen.
XBRUSH's new Open Studio leaves that screen open. Walk into a running task and you can see who asked for what, the exact prompt that produced each result, and what the agent said back. No login required.
Browsing, with no account
Click Open Studio in the top navigation and this is the landing page.
The studio is open. Watch the work as it happens, create together for free.
Two buttons: Browse tasks jumps to the lists below, Start for free opens a task of your own. A status line underneath tells you where things stand — something like "No live tasks right now — 3 open to join" — so you know immediately whether anything is running and how many tasks you could join.
There are two lists.
| List | What's in it |
|---|---|
| Open to join | Tasks currently accepting participants. Start here |
| Recent tasks | Recently finished or paused. A card with no thumbnail hasn't produced output yet |
Every card carries a thumbnail, a title and description, the team name, and elapsed time. Tasks run by XBRUSH itself get an Official badge; tasks still accepting people get an Open to join badge.
All of this is visible without an account. Open a card and see how something was made before you decide whether to sign up.
Spectating — you see the prompt, not just the picture
Click a card and you enter as a spectator. The top left reads "Spectating — you're watching the host's live workspace." Not a recording. The screen the host is looking at right now.
Reading the status bar
| Item | Meaning |
|---|---|
| LIVE | The task is running right now |
| 👁 1 | How many people are watching |
| 👥 1/10 | Participants / maximum |
| 💳 9,992 | Shared credits remaining. It drops whoever generates |
| Copy link | Copy the task link to share |
The conversation stream is the point
Every generation is logged with four things attached.
- Who ran it — the participant's name
- What they ran — the Generate label and the engine used (GPT Image 2, for instance)
- The prompt — the full text that produced the result
- The output — the generated images, with a timestamp
Take the "toy image generation" task that was running recently. One participant produced a sneaker advertising shot with GPT Image 2, and the prompt sits right there in the open — a long English instruction specifying an extreme low angle, a 45-degree product angle, cinematic rim lighting from the top right, and a navy-and-orange color palette.
Between generations, participants talk to the agent in plain language. When someone asked for a "kkojingi" drawing, the agent replied:
I'll generate 2 images of "kkojingi" — a cute hybrid character mixing the charm of a cat and a dog — using the gpt-image-2 model. (roughly ~4.8C)
It states what it will do, which model it will use, and roughly what it will cost, and only then runs. You see the reasoning that produced the image, not only the image — that is what makes spectating worth the time. How to phrase a prompt so it gives you what you wanted is something you can simply read off someone else's screen.
Anything inappropriate can be flagged with Report on the individual entry.
Joining — what you can actually do
This isn't a viewing gallery. Log in and you can step into someone else's task and generate.
The left panel is the Workspace panel, unchanged
Open Studio's generation panel is identical to the paid Workspace panel. Whatever you learn here transfers directly.
| Control | Options |
|---|---|
| Tabs | Image · Video · Audio |
| Sub-tabs | Generate · Edit · Outpaint · More |
| Engine · style | AI engine selection (Z-Image Turbo with a FREE badge, among others), Apply Style |
| Output | Image size, number of images (1–4) |
So image generation, editing, outpaint, video, and audio — everything the Workspace does works here. Three ways to take part:
- Type an instruction in plain language — the box suggests "What should we make? e.g. 4 samples of kids swimming"
- Drag a file in — dropped images immediately become source material for that task
- Drive the left panel yourself — pick the engine, size, and count, then hit Generate
What kind of tasks show up
Hosts decide what their task is about, so the list changes constantly. Recent examples give a sense of the range.
- Toy image generation — described by its host as a walkthrough of building a designer toy figure and generating a turnaround sheet. The whole multi-angle process was left open
- 12-month art calendar — a set of monthly illustrations built as one deliverable
- People and product shots — advertising portraits and product photography, the kind of output that goes straight into real work
The list refreshes daily. Any card with the Open to join badge lets you jump straight in, up to 10 participants by default (👥 0/10). Tasks the host has closed show as Participation closed — still watchable, just not joinable.
Opening your own task is free too
Hit Start for free on the landing page, or Create with XBRUSH at the top left while spectating, and an empty task opens. It sits on "Start anything with a chat" until your first instruction arrives.
Two settings are yours to control:
- 👥 0/10 — current and maximum participants
- 🔒 Participation closed — whether others can join. Closing it still leaves the task watchable
If you plan to share the link, name the task first. The title is what makes people open your card in the list.
Shared credits — your balance is never touched
This is the rule to understand before anything else.
- Generation runs on a shared pool the company provides. It is never deducted from your own balance
- But it is shared by everyone. The bar at the top drops whoever generates, and when it empties, generation stops
- It resets daily at midnight KST (Seoul time). Even if it runs dry, it is full again the next day
So there may be little left when you arrive. If generation won't run, come back after the reset — or move to the Workspace, where your team's credits are always available.
Open Studio vs. the Workspace
Both run the same AI agent and the same generation panel. What differs is whose space it is.
| Open Studio | Workspace | |
|---|---|---|
| Access | Anyone can browse; log in to generate | Your team, after login |
| Visibility | Public — anyone can watch | Private to the team |
| Credits | Shared pool. Yours untouched, but others draw it down | Entirely your team's |
| Best for | Trying things and learning by watching | Real production, where credits must be guaranteed |
Once the workflow feels natural, move to the Workspace: tasks stay private, your team shares folders, history, and storage, and the credits are nobody else's.
Four ways to use it
1. Evaluating a subscription? Spectate first
Reading one task end to end beats reading a feature page. Which engine gives which result, and what it costs, show up as actual numbers.
2. Copying prompts is the intended use
Find an output you like and read the prompt above it. The lighting, the angle, the color palette — the specifics are all written out.
3. Use it as a practice room before rolling out to a team
Send colleagues a link, have everyone join the same task, and generate side by side. Nobody spends a single team credit learning the controls.
4. Need a public demo? Open your own task
Write a title and description, share the link, and your audience watches your screen live without logging in. No screen-sharing tool required.
Frequently Asked Questions
Can I watch without logging in?
Yes. The task lists and the full contents of running tasks are visible without an account — the complete prompts, the engine used, the outputs, and the conversation between participants and the agent. A login is only needed to generate something yourself or to create a new task.
Does joining use up my credits?
No. Generation in Open Studio runs on a shared pool the company provides to all users, and your own balance is never deducted. Because the pool is shared, it drops as other people generate, and generation stops once it is empty. The pool resets daily at midnight KST.
What kind of work can I do in Open Studio?
The left generation panel is identical to the Workspace, so the Image, Video, and Audio tabs and the Generate, Edit, and Outpaint functions are all available. You can type an instruction in plain language, drag a file in as source material, or set the engine, size, and image count in the panel and generate directly.
Can I generate inside any task I find?
Only tasks carrying the Open to join badge. Participation is capped at 10 people by default and shown at the top as 0/10. If the host has closed joining, the task is marked Participation closed — you cannot generate, but you can still watch.
How is this different from the Workspace?
They run the same AI agent and the same generation panel, but the space is different. Open Studio is public and runs on shared credits, which suits experimenting and learning. The Workspace is private to your team and uses your team's own credits, which suits real production work where credits need to be guaranteed.
In short
- Watch — read every prompt, engine, output, and exchange in a running task, no login
- Join — log in and jump into an open task (10 people by default) to generate and edit image, video, and audio
- Credits — a company-provided shared pool. Your balance untouched, reset daily at midnight KST
- Your own task — Start for free, set the title, participant cap, and visibility, share the link
- Next step — once it feels natural, move to the Workspace for privacy and dedicated credits