Kadraris FAQ - Studio, pricing, models and API

Studio and generating

What does Kadraris actually do?

Kadraris brings dozens of AI models for video, images, music and audio, plus chat and coding models, under one account with one prepaid balance. Work with them by hand in Kadraris Studio, which runs in any browser, or send requests from your own software via the Kadraris API. Both spend from the same balance, so there is nothing to split between products and you can jump between models at any time. The full catalog with prices is on the models page.

Can I try Studio before I sign up?

Yes. Without logging in you can open Studio, choose a model, type a prompt and see the exact price on the "Generate" button before anything is sent. You only need an account and a balance at the moment you generate, upload a file or chat with a model - Studio then asks you to sign in and keeps what you already set up. To register, type your email and enter the six-digit code that arrives in your inbox (or use Google). No password needed. You can also browse the models, pricing and docs freely, and the public GET /v1/models endpoint returns the catalog with prices without a key.

Do I have to write code?

Not if you stay in Studio. You pick a model, describe the scene, set the clip length or the frame's aspect ratio and generate. The API is there for people who want to wire models into their own app, store, script or automation. Both live on the same account.

How do I get to my first generation?

• Sign up with an email code or a Google account.
• Add funds; $10 is the smallest amount accepted.
• Open Studio, choose a category (video, image, music, chat) and a model.
• Check the price on the button and click "Generate".

If you would rather start with code, the quickstart walks you from creating a key to getting an image back, with copy-ready snippets for curl, Python and JavaScript.

Do I get anything free to start with?

No. Kadraris has no free tier, no welcome credit and no monthly plan; every generation is paid, and adding funds starts at $10. What you get instead: money on the account has no expiry date, and a generation that ends in an error costs nothing. Every model's rate is on the pricing page, so it is easy to work out how many clips or images a top-up buys.

What can Studio make?

• Images from a text description or from your own reference photos.
• Video clips built from a prompt or from an opening image.
• Songs, music and audio clean-up (for example vocal removal).
• Conversations with chat models.

You see the cost before each click, and every task lands in the history on your account. The Studio page has an overview.

How do I animate a photo into a video?

Go to video in Studio and choose a model that can open the clip with your image - for example Kling 3.0, Seedance 2.5 or MiniMax H3. Upload the photo, use the prompt to describe the camera move and the action, and set the aspect ratio to match the photo so nothing gets cropped. The price shows up before you click. For a walkthrough read the image-to-video guide; the broader process, from a rough idea to an exported file, is in how to make an AI video.

Can I use my own photo as the starting material?

Yes, as long as the chosen model accepts that input. Image models often take reference photos for editing or for holding a style, and video models can begin a shot with a still you supply. Each model page lists exactly what it accepts. Upload only files you have the rights to (real people are covered further down).

Do prompts have to be in English?

Chat models handle many languages and reply in the one you write in. Image and video models often follow other languages too, but an English prompt tends to land closer to your intent, and a few makers only officially support English and Chinese input. When a result drifts from the description, translate the prompt into English and compare the two. Prompt structure is explained in the AI image prompt guide.

How long is the wait, and does the tab need to stay open?

Most images finish within a minute; videos take a few minutes. Timing depends on which model you use, how long and how sharp the clip is, and how busy things are. Processing happens server-side, so feel free to shut the browser or put the phone away - the output will be waiting in your Studio history. Over the API the result arrives by webhook or you read it from GET /v1/tasks/{id}, as described in the tasks docs.

My generation failed. What went wrong?

The usual causes:

• the safety system of the company behind the model blocked the prompt or picture (celebrities, brands, violence, nudity and similar),
• the model does not support a parameter you picked,
• a temporary outage on the model side.

The task is marked failed, Studio explains why with a "not charged" note, and every cent of the hold returns to your balance. Reword the prompt, compare your settings with the model page or switch models. Trying to get around the filters is against our Terms of Service. API error codes are in the errors docs.

How long do you keep my files?

Output files are available for 30 days after they are generated, while the task log with prompts, settings, model and cost is kept for 90 days. Kadraris is not cloud storage, so an expired file link does not come back. Save what matters to your own drive as soon as it is done.

Does it work on a phone?

Yes. Studio and the dashboard open in Safari, Chrome or any modern phone browser. To get an app-like icon, pin the site: in Safari tap the share icon and pick the home screen option, in Chrome open the menu and choose to add it to the home screen. Upload pictures from your gallery and save results to the phone when they are ready.

Can I use the results in ads or products I sell?

Kadraris claims no rights over what you generate. Whether you can sell or advertise with them comes down to each model's license and the law of the country you publish in. Makers write their own conditions, and fully machine-made material often gets no copyright protection. Responsibility for what you publish sits with you, including other people's trademarks, protected characters and personal rights. Before you launch a campaign, look up the maker on the model page and go through the Terms of Service.

Am I allowed to generate real people?

Only with their consent. Without consent you may not make deepfakes, impersonate anyone or create intimate or sexual images of real people. Sexual content involving minors is banned with no exceptions. Breaking these rules gets the account closed. Several models also turn down prompts that mention celebrities by name. If you publish lifelike synthetic media, add a disclosure wherever regulations such as the EU AI Act call for one.

Payments and billing

How do I pay?

By card. Payments are handled by Stripe, which means Kadraris never stores or even sees the full card number. Funds appear once Stripe confirms the charge, typically a few seconds later. If you are in Poland, the Polish version bills in zloty and also takes BLIK.

How much can I top up?

At least $10. The form offers quick amounts of $10, $25, $50 and $100, and you can type any sum above the minimum. There are no top-up bonuses: what you pay is what you get. Details on the pricing page.

Is there a subscription?

No. You load money in advance and use it at your own pace; there is no monthly fee and no minimum spend.

Can my balance expire?

No. Funds have no expiry date and we charge no inactivity fees.

Do I get an invoice?

Yes. Every top-up comes with an invoice or receipt, which you download from the billing section of the dashboard. Paying as a business? Fill in the legal name, address and tax number in the billing details first - the document is generated with whatever is saved at payment time.

Do you add VAT or sales tax?

No tax is added on kadraris.com: prices are in US dollars, and the checkout total equals the credit you receive. Companies that need a Polish VAT invoice, or EU businesses that want reverse charge on a verified VAT number, should sign up on the Polish version, where prices are in zloty including 23% VAT.

Which currency is my account in?

Sign up on kadraris.com and your account runs in US dollars. Register on kadraris.com/pl (kadraris.pl redirects there) and it runs in zloty with Polish VAT invoices. The choice is locked at registration, and money cannot be transferred between a dollar account and a zloty account.

Can I get my money back?

You can ask for one during the first 14 days after adding funds. Send the request from the address linked to the account. If you have spent nothing from that top-up, we return all of it; if you have spent part, we return the rest. You receive it on the card you paid with, no later than 14 days after asking. Referral commission is excluded. Past the 14-day mark a top-up normally stays on the account (the Terms of Service list exceptions, for instance if we shut down), though it never expires. And since failed tasks cost $0, there is nothing to reclaim for them.

Models and prices

Which models are in the catalog?

• Image: for example GPT Image 2.5, Nano Banana Pro, FLUX.2 [pro].
• Video: for example Seedance 2.5, Kling 3.0, Veo 3.1 (Quality).
• Songs with lyrics: Mureka V9.
• Chat and code: for example Claude Opus 5.5, GPT-6 Sol, Gemini 3.6 Flash, DeepSeek V4 Flash.

The list grows as new models launch. The current state is on the models page and in GET /v1/models.

Why use Kadraris when model makers have their own apps?

Apps from makers like Kling AI or Gemini run on monthly plans with usage caps. In Kadraris, models from many companies sit under one account and one balance, and you are billed per image or per video second instead of a plan. You can run the same scene description through Kling 3.0, Veo 3.1 (Quality) and Seedance 2.5 without opening three accounts. We build no models of our own; each model page tells you which company made it.

Which model fits a TikTok, Reel or Short?

For a vertical 9:16 clip with sound I would look at these:

• Seedance 2.5 - clips of 4-30 s with voices, effects and a soundtrack,
• Kling 3.0 - 3-15 s, talking characters whose mouths match the words, multi-shot clips,
• MiniMax H3 - 4-15 s with stereo sound.

Rough out the concept on a budget - drop the resolution or use the lighter Seedance 2.0 Mini - then do the final render in your main pick. For shot ideas see AI video prompts and camera moves; for the Seedance generations side by side, Seedance 2.0 vs 2.5.

Which model is good for packshots and ads?

Begin with a still image. Nano Banana Pro accepts as many as 14 references and renders up to 4K; GPT Image 2.5 edits from as many as 16 references and follows lengthy instructions faithfully. Then feed that still to a video model like Kling 3.0 or Veo 3.1 (Quality) as the opening frame. Compare branding, packaging text and proportions with the physical item every time - models tend to alter them. I describe the complete pipeline in the AI product ad workflow, and the three image models are tested against each other in GPT Image 2.5 vs Nano Banana Pro vs FLUX.2.

Which model handles lettering on an image?

Posters, covers and labels with text come out well in GPT Image 2.5, FLUX.2 [pro] and Seedream 5.0 Pro. The low-cost Z-Image Turbo ($0.02 per image) also manages short English lines. Wrap the wording in quotation marks and describe its placement and font style. Diacritics and scripts other than Latin are often mangled, so run one trial image before producing dozens. See text in AI images and the guide to Z-Image Turbo.

Can a clip have sound and a talking character?

Yes, a handful of models output sound in the same pass as the video:

• Seedance 2.5 - speech, effects, music,
• Kling 3.0 - dialogue matched to lip movement plus ambient sound,
• Veo 3.1 (Quality) - audio always included in the clip price,
• MiniMax H3 - stereo sound you describe in the prompt.

Type dialogue exactly as it should be spoken, inside quotation marks. How natural the voice sounds differs between models and languages, so try a brief clip before anything long. Each model page says whether audio is supported.

Can I make a song and a music video here?

Yes. Mureka V9 turns lyrics of up to 3,000 characters plus a genre and mood description into a complete track delivered as audio, at $0.15 per song. Once the song exists, Music Video 1.0 produces footage cut to the beat - provide the MP3 (10-300 s long) along with one to seven pictures of a character or a look. Tips on lyrics and style text: Mureka V9 music API guide. Tips on the visuals: AI music video.

Can I raise the quality or resolution of a video?

Yes, under Tools:

• Topaz Video Upscaler doubles the resolution and removes artifacts at a fixed 2x factor; it accepts MP4, MOV or MKV files of 50 MB or less and charges by source duration: $0.11 per second,
• Enhance Video 1.0 targets a resolution you choose, as high as 8K (except on lite), across three tiers - lite, standard, professional - and cleans noise and color on the way.

Handy when a video model gave you 720p and the edit needs more. See also AI video upscaler to 4K.

Where does the price of a generation come from?

Each model publishes one unit rate - an image, a video second, a song or an audio second - and chat models are priced per million tokens in and out. To give a few figures, Seedance 2.5 is from $0.261 per second, GPT Image 2.5 is $0.05 per image and Claude Sonnet 5 is $3.24 / $16.20 per 1M tokens. Studio and the API both display estimated_cost up front and reserve that sum on your balance. Chat charges follow real token usage. The full list is on the pricing page.

What will a specific video or image cost me?

Take the rate and multiply it by seconds or by image count. Per video second you pay from $0.0963 per second in Kling 3.0, from $0.261 per second in Seedance 2.5 and from $0.161 per second in MiniMax H3. Veo 3.1 (Quality) charges per clip: from $2.53 per video. An image is for example from $0.07 per image in Nano Banana Pro or $0.02 per image in Z-Image Turbo. I worked out sample totals for different resolutions and durations in separate guides: Kling 3.0, Veo 3.1, MiniMax H3, Grok Imagine and Gemini Omni.

Which video model comes out cheapest?

Veo 3.1 Lite has the smallest rate: from $0.177 per video buys an entire 8-second clip. Of the per-second models, the lowest prices belong to:

• Grok Imagine 1.0 Video (from $0.0299 per second),
• Seedance 2.0 Mini (from $0.0784 per second),
• PixVerse V6 (from $0.0405 per second),
• Seedance 2.0 Mini Eco at 720p (from $0.0789 per second).

Use them to try out concepts and prompt wording, then check movement, fine detail and audio against premium options like Kling 3.0 or Veo 3.1 (Quality). Lots of models drop their rate when you lower the resolution. For images, Z-Image Turbo and Grok Imagine sit near the bottom of the price list. The pricing page lists every rate in one table, and the AI cost calculator shows 5 s and 10 s clip totals for all models together.

Why does Studio hold extra money for vocal removal or upscaling?

These tools are priced on the duration of what you upload - Vocal Remover by the audio minute, Topaz Video Upscaler by the video second. Studio reads the duration and gives you an estimate, and a safety margin is reserved meanwhile (10 audio minutes, 120 video seconds, or 10 seconds in Motion Control). After processing, only the true cost is taken and the leftover reservation is freed at once.

Are cached prompt reads cheaper on chat models?

On selected models. GPT-6 Sol, GPT-6 Luna, Claude Haiku 4.5 and Kimi K3 have a separate cache rate, several times lower, listed on the model page. Elsewhere a cache read costs the same as normal input, and a cache write is the input rate times 1.25.

Do I pay when something fails?

No. A task with status failed costs $0, and the hold placed at the start is released in full - for provider errors, timeouts and prompts the model rejected. A task that succeeds is billed whether or not you like what came out, since the compute was used. An account that racks up 25 failures in 24 hours is temporarily blocked from new tasks (error 429) until the oldest failures age out.

Can model prices move?

They can, for example after a provider reprices or a new model version ships. Check the model page or GET /v1/models for today's figure. Anything you already started is billed at the price displayed at launch.

API and automation

Will the OpenAI or Anthropic SDK work?

Yes. Chat supports /v1/chat/completions in the OpenAI format (streaming too, with stream: true) and /v1/messages, which follows Anthropic's Messages spec. Most SDKs only need two changes: point the base URL at https://kadraris.com/v1 and supply a Kadraris key. Media jobs (image, video, audio) use /v1/generate; see the generate docs.

How does a finished video reach me over the API?

Media jobs are asynchronous. Calling POST /v1/generate gives you back a queued task carrying estimated_cost. Then either check GET /v1/tasks/{id} periodically or let the completed task be delivered to your webhook_url. Quick jobs like images can use ?wait=60 (capped at 120 seconds), which holds the connection until the output is ready. More in the tasks and webhooks docs.

Can I plug it into Claude Code, Cursor, n8n or Make?

Yes. If a tool accepts its own endpoint address for OpenAI-style or Anthropic-style calls, a Kadraris key will work in it. The variables for Claude Code are in the Claude Code docs, n8n, Make and Zapier in the no-code docs, and agents over MCP in the MCP docs.

What does Claude Code cost on Kadraris?

You pay per token, no subscription: Claude Sonnet 5 costs $3.24 / $16.20 per 1M tokens and Claude Opus 5.5 costs $3.78 / $18.90 per 1M tokens. What a month costs depends on how long you work and how many files the agent pulls into context; the dashboard breaks spend down by model over the past 30 days. I recommend a dedicated key for Claude Code with a monthly ceiling. Configuration: Claude Code docs. Pay-per-token versus a subscription: Claude Code cost: pay per token.

How many requests can one key handle?

By default 60 generations per minute per key. Above that the API answers 429 - wait and retry with growing (exponential) backoff, not in a tight loop. Need more? Tell us the volume you expect and what it is for. All codes are in the errors docs.

How do I cap what one key can spend?

The key creation form under API keys has an optional field for a monthly ceiling, set in the currency of the account. Once a request would exceed it, the API responds 403 (permission_error) and nothing is taken from the balance. A separate capped key per workflow, staging setup and coding assistant means a script stuck in a loop can only burn its own allowance. Details in the authentication docs.

How do I keep a key safe?

• Keep it on the server side: in an environment variable or a secrets manager.
• Never put it in browser code, a mobile app or a public repository.
• Give each project its own key, so revoking one does not stop the others.
• After a leak, kill the key in the dashboard immediately and issue a fresh one; usage before revocation is on you.
• On incoming webhooks, check x-kadraris-signature (an HMAC-SHA256 over the raw body) before acting on the payload.

More in the authentication docs.

Can I build a service for my own customers on the API?

Yes. Kadraris can sit behind your app, shop or workflow while customers buy your product, not ours. Keep the key server-side; customers never get it. What the Terms of Service forbid is selling plain Kadraris access or handing out keys so people skip creating an account, unless we sign off on it in writing. Read the licensing rules of the model makers you rely on as well.

What if a model provider goes down?

Certain models are available through a backup infrastructure provider, and traffic fails over on its own, so a single incident rarely interrupts you. If every path is down, you get 502 and pay nothing for the task. Make your client retry 502 and 429 after a pause.

How do I read my balance from code?

Call GET /v1/balance; the figure comes back in the account's currency. When funds fall short of estimated_cost, you get 402 up front and the task never begins, so there are no partial charges. The rest is in the billing docs.

Account and privacy

Are my prompts used to train models?

No. Neither your prompts nor your results feed any training. The prompt travels to the servers hosting your chosen model purely to produce that single output.

Who sees what I generate, and how long is it stored?

Your task log keeps prompts and settings for 90 days - useful for you, and it lets us catch misuse. Server logs last 30 days; invoices and payment records last 5 years because tax rules say so. Staff look inside a task only when you ask support for help or when a breach of the rules is suspected. Everything is spelled out in the Privacy Policy.

Are my generations public?

No. There is no public feed or gallery; only you can see your history. One caveat: any file URL, whether an output or an upload, works for whoever has it until it expires, so share it with care.

What is off limits?

• Anything against the law.
• Sexualized content with minors, in any form.
• Deepfakes or intimate imagery of a real person who has not agreed to it.
• Content made to pass yourself off as another person or company.
• Malicious code, spam campaigns and phishing pages.

Filters can reject an individual request automatically; grave or repeated abuse means the account is shut. See the Terms of Service for the complete rules.

How do I close my account?

In the dashboard, at the bottom of the Overview tab, choose "Delete account", type the account email and confirm. You can also write to us from the address on the account (the contact is in the Terms). Use up the balance first or file for a refund if eligible, since money left on a deleted account cannot be recovered. Invoices are retained as long as tax law demands.

Referrals

What are the referral rules?

Your personal link sits in the dashboard and ends in ?ref=YOURCODE. For each new user who registers via that link you earn, over the 12 months after they join, 20% of what they put into image, video and audio models plus 5% of their chat spend. Earnings begin once that user has made a paid top-up, and they are credited to your own Kadraris balance. Full terms: affiliates page.

How much can I realistically earn?

Say your referral pays $250 for video and $50 for chat in their first year. Your credit is $50 plus $2.50, a total of $52.50. Commission follows money actually spent, so their failed tasks, which cost $0, contribute nothing. Spend the credit on whatever model you like.

How long does a referred person have to sign up?

30 days. That is how long the cookie remembers your code after the click. An account created later, or after cookies were cleared, cannot be assigned to you.

Can I cash out referral earnings?

No. It exists only as balance for generating in Studio or via the API, with no refunds and no transfers between accounts.

Can I refer myself?

No. Self-referrals, mass-created accounts and registrations the other person did not know about earn nothing. If we detect gaming of the program, the commission is withdrawn and the accounts in question may be closed.