Unrestricted AI video

    The video models that don't argue with the brief.

    OpenGradient's Video Studio runs open-weights models hosted without provider-side prompt moderation - the Wan family through Wan 3.0 and Happy Horse - next to the mainstream lineup. Type the prompt you actually want, pick aspect and duration, generate. Private by architecture, 18+ capabilities behind their own gate.

    Free messages to start · no card · no identity attached

    The unrestricted lane

    Open weights, badged in the picker.

    Every model below runs as an open-weights deployment with no provider-side prompt review - clearly badged in the picker so you always know which door you're using.

    Wan 3.0

    New · Unrestricted
    30s single takeaudiono prompt filter

    The flagship of the lane: one continuous take up to 30 seconds with audio. Reads plain visual description and doesn't fill gaps - say exactly what happens.

    Wan 2.7

    Unrestricted
    audioup to 10s

    The strong all-rounder: motion, detail and audio without the filter - and the model the picker's best-quality unrestricted setting selects for you.

    Happy Horse 1.1

    Most Unrestricted
    zero filters21:9 widescreen

    The least filtered model in the Studio, specializing in realistic humans, with widescreen support. Zero filters is always Happy Horse.

    Wan 2.2 14B

    Draft lane
    88 credits per 5s draft480p

    Iterate on the awkward brief cheaply: 5-second 480p drafts for 88 credits, then carry the winning prompt up to Wan 3.0.

    The moderation wall

    Mainstream video tools review your prompt before the model sees it.

    Hosted video products moderate twice - your prompt on the way in, the frames on the way out. Legitimate work gets bounced daily: combat choreography, horror, medical scenes, romance, edgy satire. The refusal isn't a model limitation; it's a policy choice made for you.

    Refused before rendering

    Prompt classifiers reject briefs that any film school would call ordinary - and every retry burns time rewording a scene you had already written.

    Flagged prompts, filed under you

    Every rejected attempt is logged against your account. The tools keep a tidy record of everything they refused to make for you.

    Briefs people bring here

    “A brutal medieval duel in the rain - real contact, real consequences.”“Body-horror transformation scene, practical-effects aesthetic.”“A romance scene mainstream tools keep watering down.”

    Here's how OpenGradient is different

    Open weights, open door

    No prompt moderation on open models

    Wan 2.5 through Wan 3.0 and Happy Horse 1.1 run as open-weights deployments with no provider-side prompt review - clearly badged in the picker so you always know which door you're using.

    The mainstream lineup too

    Seedance 2.5, Veo 3.1 and Kling v3 Pro sit in the same picker - use frontier polish when the brief is safe, open models when it isn't.

    Anonymous by architecture

    The anonymity layer splits who you are from what you generate, and share links expire 24 hours after generation.

    How privacy works

    Free messages to start · no card · no identity attached

    The anonymity layer

    Creative freedom needs privacy to be usable.

    Not a policy promise - a protocol. No single party, including us, ever holds both your identity and your question.

    Read the full security architecture
    1. Encrypted on your device

      Messages are encrypted locally before they leave your browser.

    2. Anonymized by Oblivious HTTP

      Network relays split who you are from what you're asking. No one sees both.

    3. Unlocked only in a secure enclave

      The model reads your prompt inside attested hardware that never learns who sent it.

    Questions, answered

    Frequently asked

    What does “unrestricted” actually mean?

    The open models (Wan family, Happy Horse) are hosted without provider-side prompt moderation, so prompts that mainstream hosted models refuse will run. It does not mean anything goes: illegal content is banned, explicit capabilities are 18+ gated, and non-consensual content involving real people is prohibited.

    How good are the open models compared to Veo or Kling?

    Genuinely competitive and improving fast - Wan 3.0 renders one continuous take up to 30 seconds with audio, and Happy Horse specializes in realistic humans with zero filters. For maximum cinematic polish on safe briefs, the frontier models are in the same picker; many users audition both.

    Do I need to verify my identity?

    No. Several mainstream video tools now require a government ID, a selfie, or a face scan before you can generate anything. Here you don't even need an email to try it - and the anonymity layer keeps every prompt structurally unlinked from you.

    What does a clip cost?

    Prices are shown per model before you generate; open models are the budget path - a 5-second Wan 2.2 14B draft is 88 credits, Happy Horse at 5 seconds is 910. Credits work across everything: 1,000 credits = $1, packs from $10.

    FREE TO START • NO CARD

    Render the scene they rejected

    Free to start, no card. Pick an unrestricted model from the picker and type the brief exactly as you wrote it.

    Unrestricted means no provider-side prompt moderation on the open models - not no rules. Illegal content is banned, explicit capabilities are gated 18+, and sexual content depicting real people without their consent is prohibited.