Vellria Flash
The fastest and cheapest image model. For drafts, variations and quick tests.
Text to image1 credit
Image generation
AI image generation with no subscription, billed in prepaid credits
Vellria is an AI image generator you pay for in prepaid credits rather than by the month: a catalog of image and editing models you run from a model's own page or over a REST API, where the cost of a run is a property of the model and is stated before you commit to it. Nothing renews, credits come off when a run starts and return automatically when a run fails on the provider's side, and finished files are held for a limited retention window before they are removed from the server.
What you are choosing between
The catalog holds two kinds of image model, and the difference is what you are allowed to hand it. One kind works from a written description and nothing else. The other takes pictures you supply as references and reworks them: a wardrobe change, a different background, light falling another way, with the rest of the frame left where it is.
Which kind you are on is not a soft preference. A model built for references refuses a request that arrives without them, and a model that takes none refuses a request that carries them, rather than dropping the attachment and rendering anyway. Ignoring a picture you paid to send would leave you holding an output that never used it, with nothing on screen to tell you so.
Every model has its own page, and that page carries its own list of valid sizes with the credit cost of each, read from the server as the page loads. Those sizes are not shared across a family by assumption; they are what each model actually accepts. It is the same reason the comparison table above is drawn from the catalog instead of written into this page: the catalog grows, and a list retyped into prose is wrong the day it does.
Models
Seedream 5 Pro
The flagship image model, focused on detail and composition. 2K output only.
Text to image9 credits
Seedream 5 Pro Edit
The flagship editing model. Takes up to ten references and preserves identity and lighting.
Image to image9 credits
Qwen 3 Pro
Strong on long, complex prompts; stays faithful to scene composition and text.
Text to image6 credits
Qwen 3 Pro Edit
Edits using the images you provide as references, preserving character and scene.
Image to image6 credits
Seedream 5 Lite
Fast, cheap Seedream. Reaches 4K and is good at typography and poster work.
Text to image6 credits
Seedream 5 Lite Edit
Takes up to fourteen references and redraws the images you provide.
Image to image6 credits
What happens between pressing start and getting a file
Credits come off when the generation starts, not when it finishes. That is the honest order, since the work is commissioned at the front, and it is why the failure path matters as much as the successful one.
The run is asynchronous. A generation moves from queued, to processing, to a terminal state, and you read the result back from the record you created rather than holding a connection open; closing the tab does not lose the job. If the run ends in an error from the provider, the credits taken at the start return to your balance the moment it is marked failed, with no support request and no form to fill in. The failure and the refund both stay in your history.
Finished files land in your Gallery and download from there or over the API. They are held for a limited retention window and then removed from the server. The billing record outlives the file, so you can still see what a generation cost after the image is gone, and a request for an expired file answers as expired rather than as missing, because a deleted file and a file that never existed are different things to anything calling this from code. Download what you mean to keep.
Calling it from your own code
The console issues API keys, and you send one as a Bearer token in the Authorization header. From there the whole catalog sits behind one shape of request: a model id, a prompt, and the settings that model accepts. Start a generation, take the id you get back, and poll that record until it reaches a terminal state.
Read the catalog endpoint rather than copying values into your code. It is the same source the model pages render from, so a size or a cost that changes there changes for you at the same moment instead of drifting quietly into a mismatch you find out about from a rejected request.
Reference images have their own endpoint: post the file, take the address you get back, and pass that address in the generation body. The address carries an unguessable id and is reachable without signing in, because the provider doing the render has to fetch it, and it expires on the same schedule as generated files. Requests are rate limited per account, and when you cross the line the response tells you how long to wait.
Paying for it
There is no plan and nothing renews. You add credits with a card or with cryptocurrency, the balance rises once the payment confirms, and it falls only when you start a generation. Nothing bills in the background, and there is no monthly floor to justify between projects.
Each method carries its own set of top-up packages, so the choice is worth making before you get to the checkout rather than after. Neither of them buys a tier: a balance is a balance, and every model in the catalog answers to it on the same terms.
What a given run costs is a property of the model rather than of this page. Each model page states its own cost before you commit to it, and the pricing page lists the catalog in one table, both read from the same source the server bills from.
Where the rules come in
What is permitted here is written down, and so are its edges. Fictional adult content is allowed, the site is for adults, and the refusals that remain are absolute: any depiction of a person who is not of age, and any real, recognisable person used without their consent. A breach of those rules closes the account without notice.
Neither of those is a setting, and neither is what an ordinary creative prompt runs into. For the policy in full, including what happens to unused credits and who can see your work, read the content rules. For what happens to your wording on the way to the model, and what silent rewriting looks like from the outside, read the page on prompt control.
Frequently asked questions
Is there a subscription?
No. You buy credits up front, with a card or with cryptocurrency, and they are added once the payment confirms. The balance moves down only when you start a generation, so nothing renews, nothing bills in the background, and an idle month costs nothing.
What does a single generation cost?
Whatever the catalog says for that model at those settings. Each model page states its cost before you commit, read from the same source the server bills from, and the button you press names the amount it is about to take. Extra reference images, where a model accepts them, are priced into that same total.
What happens if a generation fails?
The credits taken at the start return to your balance automatically, at the moment the generation is marked failed, without a support request. Both the failed run and the refund remain visible in your account history, so the record of what happened survives the failure.
Can I run this from my own code?
Yes. Create a key in the console, send it as a Bearer token, and the whole catalog answers to one shape of request. Generation is asynchronous, so you start a run and poll the record it returns until it finishes. Read the catalog endpoint for valid options and costs rather than hardcoding them.
How long do generated images stay available?
For a limited retention window, after which the file is removed from the server and cannot be brought back. Your billing record stays, so the history of what cost what is intact, but the image itself is gone. Uploaded reference images expire on the same schedule, so download anything you intend to keep.