From a spark of inspiration
to a video of your own.

AI video creation starts with a reference. Upload a video, paste its URL, or select an inspiration. That gives the creative a starting point: the opening hook, delivery, scene progression and overall rhythm you want to adapt.

The source picker is the same component used in BYO. In this preview, a file you select stays in your browser tab. You can also choose the included example and move straight to the next screen.

Mark whether the reference shows a real person or an AI presenter. For a human subject, the intended avatar path remixes a still from the video into a new fictional presenter. For AI inspiration, it starts from remixed TikTok UGC. The aim is a presenter of your own, with the reference supplying creative direction.

The script and avatar sit together because they shape the same performance. Edit the lines, paste a new script, split it into scenes, and reorder the beats. Read the hook, claims, offer and final call to action before committing to generation.

The avatar screen offers two candidate paths: Nano Banana Pro and GPT Image 2.5 Sunburst. Choose one identity and add edit direction for clothing, expression or setting. The selected original image becomes the identity reference for subsequent scenes; scene edits should derive from that original.

Choose the aspect ratio here too: vertical 9:16 for TikTok, Reels and Shorts, or landscape 16:9. These decisions stay together in the draft. The preview’s avatar artwork is illustrative, so you can inspect the selection and edit controls without generating candidates.

Set the language, then choose how to cast the voice: match the inspiration, apply saved presets, or skip a separate voice audition. A useful match depends on delivery and language as well as the presenter.

The voice presets store a choice for male and female presenters in each age group: 18–29, 30–44, 45–59 and 60+. They make repeat casting easier; they still need a listening check for pronunciation, character and fit.

The requested voiceover path uses Gemini 3.8 Flash TTS through FAL. Scene direction is prepared behind the scenes from the video breakdown, alongside voiceover work where the selected video path supports it. The preview stores your choices without producing speech or substituting a browser voice.

Each generated scene belongs in a horizontally scrollable timeline. Review the clip beside its script, compare it with the inspiration, and keep the scenes that work. A redo belongs to the individual scene, so a weak shot does not require replacing the whole video.

Generated clips need review for identity, source-actor leakage, unwanted text, visual artefacts, exact dialogue, lip sync and complete audio. Pending or incomplete checks stay pending. An automated score cannot replace your playback review or silently trigger another paid attempt.

Choose the caption style on this screen: TikTok native, clean and simple, bold highlight, condensed impact or editorial serif. The preview shows a sample overlay immediately. Final captions belong in assembly, as one reviewed caption layer.

Assembly joins the approved scenes in order, maintains narration over silent cutaways and applies captions. The FAL-only architecture uses FAL auto-subtitle for the first captioning path, with the uncaptioned master retained separately.

Finishing is a final creative check: the words are correct, captions sit safely, the picture holds together, and the beginning and end of every spoken phrase survive the cuts. Keep the raw master, captioned master and finished output separate so a caption change does not require regenerating the scenes.

The sample player lets you inspect the output screen now. Its video is pre-recorded; editing the preview does not render a new export. The new live workspace currently saves drafts and preferences while generation and cloud assembly are being integrated.

Saving a workflow is optional. It keeps the decisions you want to repeat: the avatar option and direction, caption style, aspect ratio, language and voice presets across gender and age groups. Your next inspiration and script can change while those preferences stay consistent.

This is the foundation for future multi-generation: select multiple inspirations or submit multiple scripts, apply a workflow, and make several videos together. Batch submission is a future capability, rather than a job silently queued when you save.

Try saving a sample workflow in the preview, then load it again from another step. Preview workflows stay in this browser. In your workspace, draft and workflow saving use private project routes.