Tsiky

Start a new video. Step 1 (scenario) opens right away.

List existing projects, most recent first.

Takes no parameters.

Show where a project stands and what the next step is.

Lock in a step after the USER explicitly approved it. Unlocks the next step.

Store the full scenario, scene by scene (replaces any previous version). Returns a readable version to show the user.

Change one scene of the scenario without touching the others.

Readable version of the current scenario.

Build the cartoon character. Look at the user photo and describe ONLY what is visible: hair length/style/color, fringe, glasses and frame style/color/pattern, skin tone, top style/color, button color. Do not invent traits. Returns a character sheet image.

Change some traits (partial update, e.g. {"hair":{"style":"bob"}}) and get a new character sheet.

Render the current character sheet (5 expressions and poses).

Set the visual identity the whole video follows. Every field is optional; call it again to change only some fields. If the user shares a logo, a brand guide or their website colours, read the colours from it and fill them in. Bundled fonts: Bricolage Grotesque (sans-serif), Caveat (handwritten), DM Sans (sans-serif), DM Serif Display (serif), Fraunces (serif), Inter (sans-serif), Lora (serif), Montserrat (sans-serif), Nunito (sans-serif), Playfair Display (serif), Poppins (sans-serif), Space Grotesk (sans-serif). Or pass a font file path (woff2, woff, ttf, otf). Colours: bg = page background, surface/surface2 = shapes and panels, card = cards and bubbles, accent/accentDeep = highlights, ink = titles, text = body text, muted = small text.

Render the brand board: palette, fonts, logo and three frames of this video in that identity. Show it to the user before validate_stage "brand".

Render one frame of one scene, at any step (a placeholder character is used until the character is made). Use it to check a drawn decor or a layout before the storyboard.

Search the icon set (Lucide, about 2,000 line icons) by keyword, to use in scene "icons" or column "icon".

Render one key image per scene with the validated script and character, assembled into a storyboard board. Pass sceneIds to re-render only some scenes.

List the local synthetic voices.

Takes no parameters.

Synthesize the narration of every scene (or only sceneIds). Each scene is said in its voiceTone (pitch, pace, energy, pauses, breaths and timbre), with *stressed* words and ... hesitations from the narration. Returns a voice-only preview. To change how one line sounds: update_scene with voiceTone or markup, then generate_voice with sceneIds.

Replace the voice of one scene with a recording made by the user (any audio format). The mouth animation follows the recording.

Build the music bed and the full mix. With a voice-over, the music ducks under the voice. With voiceOver false, the video has background music only: no narration, the character does not speak and the narration stays on screen as captions. Moods: pop = Bright pop played on real recorded instruments (acoustic guitar, piano, bass, real drums); acoustic = Warm acoustic on real instruments (fingerpicked guitar, piano, shaker); lofi = Soft lo-fi (electric piano, round bass, light drums); piano = Calm piano, no drums; upbeat = Bright and light (nylon guitar, claps); halloween = Playful Halloween music box (celesta, A minor, no drums); none = No music, voice only.

Return the current voice + music mix.

Start rendering the final video in the background (needs steps 1 to 4 validated). Poll get_render_status.

Progress of the final render; returns the video path when done.