Handbook

Using LU Cloud

How to create the account, use the studios in a browser tab, read your credits, set up an API key, and find every setting. Seven steps, and the fixes for what usually goes wrong.

What LU Cloud is, and what you need

LU Cloud is the LU studio running in a browser tab. The models run on managed GPUs, so the machine in front of you only has to render a web page: a phone reaches the same catalog as a tower with a big graphics card. You need a browser and an account. There is nothing to install and no hardware to buy.

The catalog is the same on every plan and on every credit pack: 47 chat models, 7 image models and 5 video models. Paying more buys a bigger credit budget to spend on that list, never a longer list.

To run models on your own machine instead, the desktop app is the route. It is free, needs no account for local use, and installs from the download page on Windows and Linux. There is no Mac build, so on a Mac the browser is the way to use LU.

Create an account and sign in

Everything starts at the login page, one card with three ways in.

Press “Continue with Google” or “Continue with GitHub”. The account is created on the way back.
Email and password: the card reads “Create your LU account”. Type your address, then a password of at least eight characters, and press “Create account”. Next time the same card reads “Log in to LU” and the button reads “Log in”.
No password at all: press “Email me a magic link instead”, type your address and press “Send magic link”. Open the link from the mail, or type the code it carries into the “6-8 digit code” box and press “Verify code”.

Forgot the password? Put your address in the email box first, then press “Forgot password?”. A reset link arrives by mail. Pressing it with an empty box answers “Enter your email above first, then click "Forgot password?".”

After signing in a page reads “Setting up your account.” while your username is assigned. It then sends you into the studio if your plan is active, and to pricing if it is not. You can change the username later on the account page.

The studios in your browser

The studio lives at /app. Its header carries four entries: “Chat”, “Create”, “Compare” and “Settings”.

“Chat” is the model picker and the conversation. Above the conversation list the sidebar has two tabs. “Chat” is an ordinary conversation. “Code” is the coding agent, which works on a GitHub repository you have connected and hands its work back as a pull request. “New Chat” starts a fresh conversation. The Benchmark and Models entries belong to the desktop app and do not appear here, because they manage local backends.

“Compare” opens A/B Compare. Pick Model A and Model B, send one prompt, and read both answers side by side with their speed and token counts.

“Create” is the media studio, and every lane in it runs on hosted GPUs:

Chat & code: Chat with open models, Coding agent, Thinking mode and Voice in & out.
Image: Text to image, Image editing, Background removal, Object eraser, Upscale and Train your own LoRA.
Video: Text to video, Animate an image, Extend a clip, Motion transfer, Upscale and Lipsync avatars.
Audio: Music generation, Text to speech and Speech to text.

Plans, packs and credits

Everything you run in the browser is measured in credits, drawn from one wallet shared by chat, image, video and audio. Each model draws at its own rate, so the same question costs a fraction on a small model and much more on a flagship. The account page puts it in the same words: “One shared pool for chat, images & video.”

There are two ways to hold credits. A monthly plan grants a credit budget that refills on your renewal date; the monthly prices are Hosted at €19, Hosted Pro at €49 and Hosted Max at €99, and a yearly option sits on pricing. A credit pack is a one-time purchase from €15, in 5 sizes, with no subscription behind it. Pack credits never expire.

If you hold both, your plan credits are spent first and the pack credits only afterwards, so a pack bought mid-month is not burned before the budget it was meant to extend.

Every plan and every pack carries every model, which is why the cheapest way in is a €15 pack without a subscription, and the entry plan at €19 a month buys more credits rather than more models.

Your own balance sits on the account page under “Subscription”. “Credits this billing period” shows what is left against what the plan grants, with the refill date underneath, and “Buy more credits” goes straight to the packs.

The Flash chat allowance

Hosted includes 11 Flash models without credits; Pro and Max include 12. Unlimited prompt counts, up to 500,000 combined input and output tokens per account per UTC day in app sessions. All 47 chat models remain available on every plan. GLM 5.3 Flash uses credits on Hosted and is included in the daily allowance on Pro and Max. API keys and accounts without an active subscription always use credits.

On Hosted, 11 chat models run without spending credits inside the apps. The allowance is 500,000 input and output tokens per account per UTC day, and it resets at 00:00 UTC. It is a benefit of an active paid plan, not a free tier, and it is a daily allowance rather than unlimited usage.

One free request runs at a time per account. A second one sent while the first is still going is refused rather than quietly billed. A request whose token budget is larger than what is left of the allowance falls back to ordinary credit pricing, with a notice in the app.

Without an explicit output limit, a free request may produce up to 8,192 output tokens, and each one runs for at most 240 seconds. Requests made with an API key always use credits, whichever model they name.

The Hosted Flash models today: Llama 3.1 8B Turbo, Ling 3.0 flash, Gemma 4 26B, gpt-oss 120B, DeepSeek V4 Flash 0731, Qwen3 32B, Qwen 3.5 9B, Llama 3.3 70B Turbo, Gemma 4 31B Turbo, gpt-oss 20B, Mistral Small 3.2 24B. The list is read from the catalog itself, so it follows the catalog rather than this sentence.

Your API key and the endpoint

Any OpenAI-compatible tool can talk to LU Cloud: an editor plugin, LibreChat, your own scripts. Open the account page, find “Cloud API keys” and press “Create API key”.

The key appears once, under the line “Copy this key now. It will never be shown again.” Press “Copy” and store it somewhere safe; afterwards the list shows nothing but its prefix. Keys start with lu_. An account may hold five active keys at a time, and the bin icon beside a key reads “Revoke this key”.

Point the tool at https://lu-labs.ai/api/inference/v1 and give it the key as the API key. Two routes are served. GET /models lists the catalog with each entry’s context length, tool support and image support. POST /chat/completions answers. Authentication is the key as a bearer token in the Authorization header, and answers stream unless you send stream: false.

A key spends your plan’s credits and can do nothing else, as the settings panel says: it cannot read or change your account. Chat is limited to 60 requests a minute per account, and a refusal carries a Retry-After header saying when to come back.

Settings and your account

The account page is titled “Settings” and holds everything about the account. The same panel opens from “Settings” in the studio header, so there is one place to learn.

Profile: Your “Username” and the “Email” you signed in with. A free handle is picked for you so you can start right away; type another one and press “Update” to make it yours.
Subscription: Your plan, the credit bar and three buttons. “Change plan” sends you to the pricing page, which prices the switch first. “Payment method, invoices & cancel” opens the billing portal. If a cancellation is pending, a line reads “Your plan ends on” with the date, and “Keep my plan” undoes it.
Content policy: Three settings for how strict the rules for generated media are. “Strict” is the tightest we have, worth choosing if others use your screen. “Standard” is the default for every account. “Off” keeps the legal limits only. The setting applies to images and video generated in the cloud; text is unaffected. Material involving minors, and photographs of real people uploaded without their consent, are refused on every request whatever this is set to.
Security: “Change password” mails you a reset link and answers “Check your inbox for a reset link.” “Sign out” ends the session in this browser.
Connections: “Connect GitHub” is what powers the Code agent. You choose which repositories it can see, and changes only ever arrive as pull requests. “Disconnect GitHub” takes it back.
Danger zone: “Delete account” permanently deletes your account, cancels any active subscription, and erases your handle, credits and generation history. You type your email address to confirm, then press “Permanently delete”. It cannot be undone, so ask for a copy of your data first if you want one.

HELP

When something goes wrong

The payment did not go through: Three different sentences, three different cures. “This payment was not sent to your bank, so nothing was charged and your plan is unchanged. You can try another card, or contact support and we will sort it out.” means the payment never reached your bank, so another card usually works. “Your card was declined. Nothing was charged and your plan is unchanged.” came from the bank itself, so the bank is the one to ask. “Your bank did not confirm the payment. Nothing was charged and your plan is unchanged.” means the confirmation step was never finished. In all three, nothing was charged.
Credits have not arrived yet: After checkout the page reads “Checkout returned. Credits become available after payment is confirmed. Some payment methods take longer to confirm.” Give it a moment and reload the account page. Cards clear quickly, some other methods do not.
You are out of credits: The modal reads “Your plan credits for this billing period are used up. They refill on your next renewal date. Top-up credits are one-time, never expire, and are only used after your plan credits.” Either wait for the refill, buy a pack on pricing, or move up a plan.
Free chat says it is busy: “One free chat request is already running on this account. Wait for it to finish or stop it first.” The Flash allowance admits one free request per account at a time. Let the first one finish, stop it, or send this one on credits.
The sign-in code will not take: “Invalid or expired code” means the link or the code has been used already or has timed out. Go back and press “Send magic link” for a fresh one. Mail clients that pre-open links are the usual reason a link is spent before you press it, which is what the code box is there for.
The conversation is too long: A context_exceeded error means the conversation no longer fits in the model’s context window, even after older messages were dropped. Start a new chat, or pick a model with a larger window. In Code and Agent work the tool definitions take room in that window too, so turning off tools you do not need buys some back.
Generated media disappeared: Generated media is deleted on purpose after a week. Settings says: “Your chats stay in this browser by default. Generated media lives in your private cloud library and is auto-deleted after 7 days.” Download anything worth keeping with the download button on the result.

Anything still stuck, or a charge you cannot place: the FAQ covers the rest, and every page on this site carries the support address.