AI video generation: choose a workflow for your starting material.
Published by ClapGen · Updated
A scene described in words, a still image and a source performance provide different kinds of direction. ClapGen Video Studio exposes compatible workflows and controls through the selected model. The public release currently uses simulated requests. Live generation and paid subscriptions are unavailable.
Choose between words, images and a source clip
Start with text-to-video when you want the model to invent a scene from a written brief. Start with image-to-video when the appearance or opening composition is already in a still image. Use a source-video workflow when an existing clip supplies movement or an editing starting point.
These inputs are not interchangeable. A general image reference need not define the first frame, and a still image does not supply a recorded performance. Pick the workflow before assembling the media.
Set up a request in the studio
1. Open Video Studio and choose a model. Read the workflow choices and any unavailable-state message for that selection.
2. Select a compatible workflow, such as Text, Frames or Reference when shown. Changing the workflow can change which references and controls are valid.
3. Add the required media in its assigned role. A first-frame image belongs in the first-frame slot; a reference clip belongs in the relevant video-reference or motion workflow.
4. Write a brief for one understandable shot, then choose from the duration, format, resolution and audio options that remain available.
5. Check the estimate and input messages. In this release, requests are simulated rather than sent for live AI generation.
Separate subject movement from camera movement
Subject movement is what happens inside the scene: a tram passes, a curtain lifts or a person turns. Camera movement changes the viewpoint: a pan follows a subject, a tracking move travels beside it or a static shot holds its position.
Example brief: A yellow tram crosses a quiet, tree-lined avenue in soft morning light. The camera stays at street level in a wide, static view as the tram moves from left to right. This is a planning example, not a demonstrated output.
Camera presets offer starting language such as tracking, orbit, static, cinematic and aerial movement. They communicate intent; they do not create a precise physical camera path or a guarantee of identical motion.
Read the controls for the selected workflow
Duration, aspect ratio and resolution are model-specific. Some source-video workflows follow the source duration or framing and therefore hide an independent control. A missing setting may reflect that workflow rather than an incomplete form.
Audio is also model-specific. Mentioning music or dialogue in a prompt does not give a model an audio capability that its controls do not offer. Check the available option and review the future output rather than assuming sound is included.
A simulated quote or status is not a completed live clip or an opened subscription. Use Pricing for the current payment status.
Plan a sequence as understandable shots
For a short sequence, describe what each shot needs to show and how it should end. Keep a record of the same character, clothing, scene and light so that the next brief does not accidentally introduce a different identity.
A long paragraph of cuts, transformations and competing actions is harder to review than a focused shot. A request is not a promise of an editable timeline. Use the controls actually present in the chosen workflow and assess each resulting clip separately when live generation becomes available.
Review a clip before reusing it
Check the opening, middle and ending as well as playback. Look for changing facial features, extra objects, sliding feet, implausible contact, unstable backgrounds and a camera move that contradicts the brief.
If the subject moves incorrectly, clarify the action. If the viewpoint moves incorrectly, simplify the camera direction. If appearance changes, review the starting frame or identity references. Revise one cause at a time.
If a request has an uncertain status, check its history before starting another. A new request can be a separate action; uncertainty is not proof that nothing was submitted.