FAQ
Answers to the questions we get most often — pricing, protocol errors, stuck jobs, content moderation, accounts, and limits.
General
What is Atlas Cloud?
Atlas Cloud is an AI API aggregation platform. One API key and one balance give you access to 400+ models — language models, image, video, audio, and 3D — from many providers, without signing up for each one separately.
How do I get started?
- Create an account
- Create an API key
- Follow Get Started for your first request
What models are available?
Browse the Model Library. The catalog changes often, so it is the authoritative list — the docs deliberately avoid hard-coding model names that go stale.
Is the API compatible with OpenAI?
Yes, and with more than that. Atlas Cloud accepts six request formats: OpenAI Chat Completions, Completions, Responses, and Images, plus Anthropic Messages and Google Gemini. See LLM API Protocols.
API & Technical
I'm calling a model and getting a 400. Why?
The most common cause is using the wrong protocol for that model. Not every model accepts every format:
- Anthropic-family models expect
POST /v1/messages, not/v1/chat/completions - Some OpenAI reasoning and codex-family models expect
POST /v1/responses, not/v1/chat/completions
Each model publishes the protocols it supports. Check the model's API reference before switching formats, and see LLM API Protocols for the differences between them.
What are the API endpoints?
| Purpose | Endpoint |
|---|---|
| LLM chat | POST /v1/chat/completions |
| LLM (Anthropic format) | POST /v1/messages |
| LLM (Responses format) | POST /v1/responses |
| Image and 3D generation | POST /api/v1/model/generateImage |
| Video generation | POST /api/v1/model/generateVideo |
| Audio: speech, music, transcription | POST /api/v1/model/generateAudio |
| Check job status | GET /api/v1/model/prediction/{id} |
| Upload a file | POST /api/v1/model/uploadMedia |
| Balance and usage | GET /public/v1/balance |
Base URL: https://api.atlascloud.ai
My job has been processing for a long time. Can I cancel it?
Image, video, audio, and 3D generation are asynchronous. Submit the job, then poll GET /api/v1/model/prediction/{id} until it reaches a terminal state. Video jobs in particular can take minutes.
There is no cancel endpoint. Jobs that exceed the platform timeout are terminated and refunded automatically.
If you are polling frequently, switch to webhooks instead — you get a callback when the job settles rather than holding a polling loop open.
Do you support webhooks?
Yes, for image, video, and audio jobs. See Webhooks for the payload format and signature verification.
How do I upload files?
Use POST /api/v1/model/uploadMedia and pass the returned URL into your generation request. You can also supply a public URL directly, or inline Base64 for most image and audio fields. See Upload Files.
Are there rate limits?
Yes, applied per account and per model. Exceeding one returns 429.
Note that LLM and media endpoints do not return rate-limit headers, so you cannot read your remaining quota from a response — implement exponential backoff instead. See Errors & Rate Limits.
Need higher limits? Contact us with your expected volume.
I'm getting 401 errors. How do I debug?
Check in this order:
- Is the key correct and still active? Check API Keys
- Is the key scoped correctly? A Coding Plan key cannot call models that do not accept one
- In a team context, are you using a key created in the right account?
- Is the URL path correct? The gateway authenticates before routing, so a typo in the path also returns
401rather than404
What is the MCP Server?
An official MCP Server that exposes Atlas Cloud as tools inside AI-assisted IDEs — generate media, chat with models, and check your balance without leaving the editor.
Pricing & Billing
Is the price on a model card the total cost?
No — for image and video models it is a unit price. The final cost depends on parameters such as duration, resolution, and number of outputs. A video model priced per second will cost that rate multiplied by the clip length.
To get an exact figure before you spend anything, use the pricing estimate before submitting. See Model Billing for worked examples.
Why doesn't the price I saw match my bill?
Three things can differ:
- The model card shows a list unit price
- The playground recalculates as you change parameters
- Your bill also reflects account discounts and promotions
Image and video use asynchronous billing: an amount is held when you submit and settled against actual usage when the job finishes. A pending hold is not a final charge.
Do failed requests cost money?
No. Jobs that fail, time out, or are rejected by content moderation are not billed. If you see a charge you cannot account for, contact support with the prediction ID.
Can my balance go negative?
It can dip slightly negative under high-concurrency LLM usage, since language model calls are not pre-authorized. A negative balance blocks further requests. Enable auto top-up to avoid interruptions.
Can Coding Plan credits be used for image or video?
No. A Coding Plan is a separate subscription allowance for language model usage, and it cannot be spent on image, video, or audio generation. Those draw from your regular balance. Coding Plan credits also cannot be converted into account balance.
How do I get an invoice?
Request or re-issue one from the billing history in your dashboard. Download links are time-limited, so save the PDF when you generate it.
What payment methods are accepted?
Credit and debit cards (via Stripe), WeChat Pay, Alipay, and cryptocurrency. Payments are processed in USD.
Are there sign-up bonuses or referrals?
Atlas Cloud runs top-up promotions from time to time — check the billing page for what is currently active. There is also a referral program: share your link and earn credits when someone you refer completes their first top-up. See Balance & Credits.
Do credits expire?
Prepaid credits are valid for a defined period from the date of purchase, and promotional credits may carry their own terms. See Balance & Credits and the Acceptable Use Policy.
Content & Moderation
My video generation failed with a copyright or sensitive-audio error. What do I do?
This usually comes from the audio track, not the video. Options:
- Turn audio generation off if you do not need it — many video models accept a flag such as
generate_audio: false - Supply your own royalty-free audio as the soundtrack instead
- Rephrase the prompt to ask for original music rather than naming an existing track or artist
Requests blocked by moderation are not billed.
My generation was blocked by content moderation. Why?
Prompts describing real people, minors, or otherwise restricted content are rejected. Some models support reference-driven generation with material you upload, which is the supported path for likeness work. See the Acceptable Use Policy.
A model I was using has disappeared. What happened?
Models are occasionally retired when a provider discontinues them or reliability drops. The Model Library always reflects what is currently available. If a model you depend on is gone, contact support and we can suggest a replacement.
Accounts & Teams
I signed in with GitHub but can't add credit.
GitHub does not always return an email address, and billing requires one. Add an email from account settings, then top up as normal.
I signed in and my balance is missing.
You may have created a second account. Signing in with Google, GitHub, or email produces separate accounts even with the same address unless they have been linked. Sign in the way you originally registered. If duplicates exist, contact support.
What can teams do?
Team accounts share a balance, support multiple members with roles, and let you create team-scoped API keys. Spend can be broken down per key and per member, and balance can be transferred between a personal account and a team. See Team Accounts and Roles & Permissions.
How do I delete my account?
Contact support from your registered email address with your account UUID. Note that:
- Remaining balance is not refundable — spend it first
- There is a cooling-off period during which you can cancel the request
- After closure, the same email cannot be used to register again
If you only want to remove generated content rather than close the account, you can delete history records directly. See Data Deletion.
Data & Security
How is my data handled?
Requests are transmitted over TLS. Atlas Cloud retains request content only briefly for service delivery and billing, and each upstream provider has its own retention policy. See Data Retention and the Privacy Policy.
Is my content used to train models?
No. Atlas Cloud does not use content submitted through the platform to train, fine-tune, evaluate, or benchmark its hosted models unless you authorize it in writing.
Can I control how long my data is kept?
Yes. Media generation requests accept per-request retention headers that shorten how long generated files and request records are kept. See Data Retention.
What about data protection rights?
You can request access, deletion, or export of your personal data at any time. Contact [email protected]. See Data Deletion for what gets removed and what must be retained for legal and accounting reasons.
Getting help
- Documentation — browse the sidebar, or start with Get Started
- Email support — [email protected]
- Community — Discord
- Sales and volume pricing — [email protected]
- Privacy requests — [email protected]
When reporting an API problem, include the X-Request-ID header from the failing response, or the prediction ID for a generation job. See Errors & Rate Limits.
Last updated on