AI Stories: from script to video

Turn an idea or finished script into a multi-scene AI video. Prepare references, direct each scene, choose audio, correct shots, and download the finished story.

Create an AI Story Open project history Explore AI video workflows

Choose your video workflow

AI Stories generates new images and video clips for each scene, then composes them into one story. Sign in to save projects and generate complete videos.

AI Stories: Start from an idea or finished script for a multi-scene narrative, campaign, or explainer. Open AI Stories to create a project or continue one.
AI Short Drama Studio: For a series with episodes, character assets, and storyboards, use Short Drama Studio. It has its own project workflow.
Single Scene Clips: Use Single Scene Clips for one continuous generated shot.
Video Mixcut: Use Video Mixcut to edit existing Video Library footage together.

Create your first story

Open Create an AI Story. Choose an idea or an existing script, video type, language, and target duration. Start with 15–30 seconds.
Generate a script from your idea, or prepare your existing script for scenes. Read the result before continuing.
Upload at least one product, person, or logo reference. Choose a primary image in every category you use.
Select the image style, image model, audio mode, video model, resolution, and portrait or landscape format. Review the credit estimates.
Review every scene’s Image Prompt, Motion Prompt, and Audio Prompt. Check dialogue and supported scene durations, then start generation.
Open the project to inspect images and clips. Correct any scene that needs work, compose the complete video, preview it, and download.

Start from an idea or a script

Idea mode uses the selected script model to write the script, title, and scene prompts. Include who is involved, where the story happens, the action, and the ending. Existing-script mode prepares your text into scenes without asking an AI model to rewrite it.

Project settings: The target duration ranges from 15 to 300 seconds. Choose social, explainer, or storytelling/drama. Actual runtime depends on scene durations and generation results.
Dialogue and narration: Name the speaker and write the exact line. Mark narration explicitly. Action descriptions can remain silent; they do not need to become spoken narration.
Review the prepared script: Check scene boundaries, visual descriptions, and spoken lines. After changing the source or project settings, prepare or generate the script again when needed.

TITLE: The lost key

[00:00-00:05]
VISUAL: Mia stands outside a closed cafe at dusk, searching her bag.

[00:05-00:10]
VISUAL: A barista at the doorway holds up a small brass key.
DIALOGUE: Barista: Looking for this?

[00:10-00:15]
VISUAL: Mia smiles and reaches for the key.
DIALOGUE: Mia: You saved my evening.

Try this 15-second example in existing-script mode. Check the three prepared scenes and adjust their durations to the selected video model before generating.

Prepare reusable references

Full video generation requires at least one global reference image. References are grouped as product, person, and logo; you only need the categories relevant to your story.

Upload clear images in the appropriate categories. Use a name and note to explain which character or object the image represents.
Mark one image as primary in each non-empty category. Empty categories do not need a primary image.
Use consistent character names, clothing, and object descriptions in scene prompts. Reuse the references when continuing the project.

References guide appearance; inspect each generated image and video to check that identity, placement, and proportions match your story.

Direct each scene

Review these fields independently before rendering. An opening image establishes the scene; the motion prompt describes what changes during the shot.

Image Prompt: Describe subjects, appearance, position, orientation, environment, lighting, and the opening composition. This field is required.
Motion Prompt: Describe the action, camera angle or movement, and pacing. Keep the requested movement possible from the opening image. This field is required.
Audio Prompt: Write only the speech you want heard. Leave it empty for a silent action shot. Check the speaking character and exact dialogue.
Scene timing: Allow enough time for the action and spoken line. In native audio mode, select a supported duration for each scene; changing model or resolution may change the available durations.

Choose audio, voices, and text

Voice and video together: Native audio generates speech, action, ambience, and effects together. Only compatible video models are shown. Listen for complete dialogue, lip sync, and consistent character voices.
Character voice references: When the selected model supports them, assign saved voice references to speakers. Manage saved voices in AI Studio TTS. Some models use voice descriptions instead of saved samples.
Separate voice (TTS): Generate the picture and add separate speech. Choose the language, available TTS model, and a preset or saved voice.
Music and text: Enable or disable speech, background music, and text overlay as needed. Adjust music and voice/scene volume, and review text styling before generation.

Voice references help preserve timbre but do not guarantee a match. Review every new take, especially when changing models.

Models, credits, and generation

Choose the image model for opening frames and the video model for moving scenes. Available models and estimates are loaded in the form; use those current values when choosing.

Check image style, resolution, portrait/landscape, and crop or letterbox. Confirm the framing keeps faces, products, and text visible.
Wait for model lists and estimates to load. The script step charge is shown separately; the story estimate combines image and video estimates. Final credits are calculated at submission.
Start generation once. Follow project status and the completed image/clip counts in project history, then open the project for scene details.

An estimate is not a fixed price or delivery-time guarantee. Longer scenes, more scenes, and regenerations can change the cost. Check the displayed estimate before starting each generation.

Correct a scene and compose again

Preview the opening image and video. For positioning, movement, or a transition problem, open “Correct this scene” and describe the intended result.
Choose “Review correction” and inspect the proposed image and motion instructions. Apply the correction only after the review is ready; if the scene changes, review it again.
Applying a correction saves instructions. If a new opening image is required, generate it first, then generate the scene video. If the frame can be reused, regenerate the video.
For manual changes, edit scene prompts and choose the image/video generation controls. Check the new take and listen to dialogue.
Once every scene has a usable clip and none is waiting for regeneration, generate the whole clip again. Preview the new complete video before downloading.

Saving a correction or regenerating one scene does not update the final video by itself. Compose the whole video after the scene revisions are complete.

Save, resume, and download

Save progress: Use Save progress before leaving the creation form. A saved project keeps the input, script, references, and settings. Reopen it from AI Stories project history.
Draft or interrupted script: Open the existing project to restore its form. Retry a failed script or submission from that project, after checking the displayed error.
Interrupted video generation: Open the project and use Continue generation when offered. Review saved images and clips before starting any additional regeneration.
Final video: When the complete video is ready, preview and download it from the detail page. Generated videos are also discoverable in Generated Outputs.

Unsaved edits can be lost on refresh. Cancel stops the current generation request; it does not delete the saved project. Use project history to reopen it.

Troubleshooting

Generation is blocked: Check that a reference exists, every used category has a primary image, required image/motion prompts are filled, and the selected model supports all scene durations and the chosen format.
Models or prices do not load: Use Retry next to the failed model list. Wait for estimates to finish loading before submitting.
Unexpected speech or missing dialogue: Check Audio Prompt and speaker fields. Leave action-only scenes silent, mark narration explicitly, and shorten dialogue or increase the supported scene duration.
The final video shows an older take: Wait for the corrected scene video to finish, then generate the whole clip again. A prompt edit alone does not render new media.
A project seems missing: Confirm the account, refresh project history, and check whether you saved before leaving. If the list shows a loading error, retry it instead of assuming the history is empty.
Still blocked: Send the project ID, failed step, and displayed error to support.