Seedance 2.5 Video
A cinematic video model. Long scenes and natural camera movement.
Text to videofrom 28 credits / s
Video generation
An AI video generator without a filter on what you can describe
A video generator without a filter sends your description to the model unedited, so the clip you get back is built from your wording rather than from a sanitised paraphrase of it. Vellria runs video models this way, charging prepaid credits per clip with no subscription, refunding automatically when a run fails, and refusing only two things: anyone under age, and real people used without consent.
What a filter costs you in video specifically
Filtering hurts video more than it hurts stills, and the reason is arithmetic. A still that comes back wrong costs one render and a few seconds of your attention. A clip costs considerably more of both, and the thing that went wrong is spread across every frame rather than sitting in one image you can glance at and discard.
Rewriting also damages the parts of a description that video depends on. A prompt for a clip carries motion, pacing and camera behaviour alongside the subject, and a moderation pass that softens the subject frequently flattens the verbs around it too. What returns is not merely tamer, it is stiller: the described movement drains out along with the wording that specified it.
The third cost is iteration. Video work is a loop of watching a result and adjusting one variable, and that loop only teaches you something if the input you adjusted is the input that ran. When an unseen layer edits your text between the two, you are tuning against a moving target and the loop stops converging.
Models
Seedance 2.5 Image Video
Takes the image you provide as the first frame and animates it into a long scene.
Image to videofrom 28 credits / s
Seedance 2.0 Fast Video
The cheaper Seedance line, with seven aspect ratios to choose from.
Text to videofrom 5 credits / s
Seedance 2.0 Fast Image Video
Takes your image as the first frame, and still lets you choose the frame shape.
Image to videofrom 5 credits / s
MiniMax H3 Video
Fast, economical video generation. Straight from text to scene.
Text to videofrom 5 credits / s
MiniMax H3 Image Video
Takes the image you provide as the first frame and animates it.
Image to videofrom 5 credits / s
MiniMax H3 Max Video
The higher-priced MiniMax tier. Six aspect ratios, on a ladder that reaches 4K.
Text to videofrom 8 credits / s
MiniMax H3 Max Image Video
Takes your image as the first frame and animates it at the higher-priced MiniMax tier.
Image to videofrom 8 credits / s
HappyHorse 1.1 Video
A video model that reaches 1080p. Nine aspect ratios, 3–15 seconds.
Text to videofrom 11 credits / s
HappyHorse 1.1 Image Video
Takes your image as the first frame and animates it up to 1080p.
Image to videofrom 11 credits / s
Wan 3.0 Video
Video with sound you can switch off. A single scene up to 30 seconds.
Text to videofrom 8 credits / s
Wan 3.0 Image Video
Takes your image as the first frame and animates it for up to 30 seconds.
Image to videofrom 8 credits / s
Wan 3.0 Pro Video
The large version of Wan 3.0 Video. Higher detail and steadier motion.
Text to videofrom 11 credits / s
Wan 3.0 Pro Image Video
Starts from the first frame and continues with Pro's detail and motion quality.
Image to videofrom 11 credits / s
How a clip is actually produced here
Pick a video model, write what you want to see, and set it going. The balance drops at that moment rather than on delivery — the work is bought when it is ordered — and from there the job runs on its own. It queues, it renders, it finishes or it fails, all without you watching; shutting the laptop does not reach into the queue and stop anything.
Not every video model wants the same input, and the mismatch is not a detail you can shrug at. Some compose motion purely from a written description. Others begin from a still you hand them and animate outward from it. Give the second kind nothing to start from and it stops rather than inventing an opening frame, which is exactly right when the entire point was to move one particular picture and not something vaguely like it.
Durations, resolutions and what each combination costs sit on the model's own page, pulled live as it opens. Those numbers belong to the model, not to this piece of writing, which is also why the table further up is drawn straight from the catalog: a specification copied into a paragraph starts lying the first time the catalog moves underneath it.
When a run fails, and what it costs
Renders do fall over at the provider end, and when one does the balance is made whole automatically as the job flips to failed. There is no queue to join and no message to compose; the failure and its reversal both sit in the history afterwards. For video this arrangement earns its keep, because a wasted clip is worth several wasted stills.
What survives lands in the Gallery, downloadable there or through the API. Files are not archived indefinitely — after their window they are cleared, though the ledger entry sticks around, so the cost of a clip remains answerable after the clip itself has gone. Treat the Gallery as a hand-off point rather than a hard drive and pull down anything you want to keep.
There is no plan running underneath any of this. Balance goes up when a card or crypto payment settles and comes down when a job starts, and it does nothing at all in between. Weeks pass without a shoot in ordinary video work, and those weeks should be free rather than billed as an idle subscription.
The two refusals that remain
None of this means there is no line — the prompt path is open, the subject matter is not infinite. Fictional adult work is fine and the audience is adults. Two subjects are refused every time, with no setting anywhere that loosens them: minors in any depiction at all, and identifiable real people brought in without their agreement. Breaching either can end the account.
An ordinary creative brief never approaches those two, which is why they can be stated in a sentence and left alone. The whole policy, credits included, is on the rules page. The same argument about rewriting, worked through for stills instead of motion, is on the no-filter image page. Driving all of this from your own software instead of a browser is the API.
Frequently asked questions
What does without a filter mean for video?
That the description you write is the description that renders, including the motion and camera language in it. There is no rewriting pass that softens wording on the way to the model, so an adjustment you make to the prompt is the adjustment the model sees rather than an approximation of it.
Can I animate an image I already have?
Yes, with the models built to take a reference frame. Upload the still and the clip is generated from it. Those models refuse a request that arrives without an image rather than inventing a starting frame, which is what you want when the aim was to move one specific picture rather than something similar to it.
How long are the clips and what do they cost?
Both follow from the model you pick. Its page lists the durations and resolutions it supports with the price of each, read live from the catalog as the page opens, and the pricing page gathers the whole catalog into one table. None of those figures are typed into this page, because a number copied into prose is wrong the first time the catalog shifts.
What happens if a clip fails to render?
You are not charged for it in the end. The balance moves when the job opens and moves back when the job lands in failed, with nothing to submit and nobody to ask. Both entries remain visible afterwards.
Is there a subscription for video?
None. Balance is bought ahead with a card or crypto and spent only when a job starts, with nothing accruing in between. That suits how video work actually goes, since the quiet stretches between projects would otherwise be the expensive part.