No videos generated yet. Upload an image to start!
No videos generated yet. Upload an image to start!
2.VIDEO brings several video model families into one workspace. Each version exposes its own input types and output settings, so select the model after you know whether the scene starts from text, a frame or reference media.
Use the available Sora 2 versions for prompt-led video generation. The model selector shows the aspect ratios, resolutions and durations that can be submitted for the version you choose; unsupported combinations are not presented as universal options.
Veo 3 and Veo 3.1 variants are available for supported text and media-guided workflows. Choose a concrete version first, then use only the reference fields and output controls shown by the generator form for that version.
MiniMax Hailuo H3 supports model-specific reference workflows alongside prompt generation. The workspace keeps its requirements visible so you can prepare the right material instead of assuming every model accepts the same files.
Seedance 1 and Seedance 2 variants are included with their own supported controls. Use the prompt and any reference inputs exposed for the chosen version, then review its duration, resolution and credit estimate before submitting.
Start with the intended scene, then let the selected model determine which inputs and settings are valid.
Open the model selector and compare the available Sora, Veo, MiniMax, Seedance and Wan options. Select the version that supports the input method you need; its visible controls define the valid choices for the rest of the form.
Write a prompt that identifies the subject, action, setting and visual treatment. A focused scene is easier to evaluate than a long script containing several scene changes, so separate unrelated ideas into later generations.
If the selected model exposes first-frame, end-frame, image, video or audio reference fields, add only the material that helps define the scene. Text-only generation remains available without attaching media.
Choose from the aspect ratios, resolutions and durations displayed for the model. Check the calculated credit cost, submit the task, then preview or download the completed video from Recent Videos.
The generator keeps model selection, inputs, output settings, task history and credit information together without pretending every provider has the same contract.
Use a prompt by itself or add the frame, image, video or audio references supported by the selected model. Media fields appear as capabilities, not as promises that every model accepts every format.
Good results begin with a narrow creative decision and inputs that agree with one another. Use this guide to prepare the prompt, reference media and output settings before spending credits.
Begin by deciding what must stay fixed. A text-only concept leaves more visual decisions to the model. A first frame can anchor the opening composition, while an end frame can define where a supported transition should arrive. Image, video or audio references can provide additional context only when the selected model exposes those fields. 2.VIDEO therefore treats the model version as the first decision, not a decorative label. Read the visible input area after changing models, remove material that no longer applies and avoid assuming settings from a different provider will carry over. If two models support the same general idea, run separate, clearly named attempts and compare the completed clips rather than changing several variables inside one prompt.
Describe what the viewer should see: the main subject, the action, the environment and the intended camera treatment. Concrete relationships are more useful than a stack of adjectives. State where the subject begins, how it moves and what should remain visually stable. Add lighting, time of day, material or atmosphere only when those details change the scene. If the prompt asks for several locations, characters and edits at once, split it into smaller clips so each result has a clear success criterion. When reference media is present, use the prompt to explain desired motion and change instead of repeating every visible detail. This makes later revisions easier because you can identify whether the problem came from the prompt, the media or the selected model.
Choose clean, relevant files and check that their content does not contradict the prompt. A first frame with a centered product and a prompt demanding an off-screen subject creates competing instructions. An end frame should represent a plausible destination rather than an unrelated scene. When a model supports multiple references, add only those that contribute a distinct constraint; extra material can introduce ambiguity without adding useful control. Review previews before generation and remove an accidental attachment instead of relying on the model to ignore it. If you want a pure text experiment, use the dedicated Text to Video page, which clears media fields at initialization, editing and submission boundaries so a hidden reference cannot enter the request.
Available durations generally range from 5 to 15 seconds, depending on the selected model version, and resolution or aspect-ratio choices also vary. Treat those controls as creative constraints: vertical framing changes subject placement, wider framing creates more environmental space and longer duration asks the model to maintain motion for more frames. Review the credit estimate before submitting because model and settings affect cost. After completion, preview the result at full size and decide what needs to change. Keep the original task in Recent Videos, adjust one factor such as the prompt, model or duration, and regenerate. A short sequence of controlled comparisons produces more useful evidence than repeatedly rewriting every field.
Choose a model, write the prompt, add only supported references and review the exact credit estimate before generation.
Text or supported reference media
Model-specific output controls
Recent videos and regeneration
These answers describe the controls and limits implemented in the current 2.VIDEO workspace.
Every task can begin with a text prompt. Depending on the selected model version, the generator form may also show first-frame, end-frame, image, video or audio reference fields. A field that is not shown for the current model is not treated as supported input.
The current selector includes versions from Google Veo, OpenAI Sora, MiniMax Hailuo, ByteDance Seedance and Wan. Model availability and individual versions can change with the configured service, so use the live selector as the authoritative list when preparing a task.
You control the prompt and the aspect ratio, resolution and duration options exposed for the selected version. Reference media adds another form of control when supported. 2.VIDEO does not claim a universal storyboard, caption, voiceover or brand-preset editor inside this generator.
Available duration choices depend on the model and currently fall within a short-clip range, generally 5 to 15 seconds. The generator form only offers valid choices for the selected version rather than promising one duration across all models.
The required credits are calculated from the selected model and settings and displayed before submission. Check that amount before starting the task. Available subscriptions and credit packs are shown in the pricing flow.
2.VIDEO does not make a blanket commercial-rights promise on this page. Usage rights can depend on the model provider, source media and your account or plan terms. Review the applicable terms and confirm that you have rights to any uploaded material before publishing.
First identify one problem: subject, motion, composition, timing or style. Clarify that point in the prompt or replace the conflicting reference, then regenerate with the remaining settings unchanged. Controlled revisions make it easier to learn which change improved the clip.
Recent Videos shows your generation history. Completed results can be opened for preview and download, and saved task parameters can be used to start another generation. A regeneration creates a new task and does not overwrite the prior video.
