brief-to-launch-video

Requirements

• Clueso MCP connected (create_project, add_clips, add_elements, voiceover_batch, export_project and friends must be available). Nothing else.

Inputs

Collect these before building. Ask for anything missing rather than inventing it:

1. Product name and one-line positioning.
2. Audience — who the video speaks to.
3. 3-5 key points — features, outcomes, or proof points, in priority order.
4. Call to action — what the viewer should do at the end.
5. Optional: brand colors / logo / product screenshots. If the user has real product footage or screenshots, use them — never fake their product with stock or generated imagery.

Workflow

1. Confirm the workspace

Call find(type='workspaces') and confirm with the user that the active workspace is the intended one before creating anything. Use switch_workspace if not.

2. Look for a clueprint first

Call find(type='clueprints', query='<product launch / announcement style the user wants>').

• Strong matches → show the best 2-3 (name, what it's for, why it fits) and ask whether to build from one. A clueprint's design — colors, typography, layout, motion — is the direction; follow it.
• Weak or no matches, or the user asked to start blank → call get_design_guide and follow it. Commit to a palette before building: brand colors if provided (applied at video intensity), otherwise one accent with real contrast and neutrals tinted toward it.

3. Write the script before touching the timeline

Draft the voiceover script as 5-7 beats in a classic launch arc:

Rules: conversational register, no feature-list monotone, every sentence earns its seconds. Call estimate_duration on the script; if it lands outside 30-60s, cut or split beats — don't speed up the voice to fit.

Show the user the script and get a nod before composing. This is the cheapest moment to change direction.

4. Create the project and build one clip per beat

• create_project with a name like " launch video".
• add_clips — one clip per beat, durations from the script estimate.
• Call get_element_schema before your first add_elements call and compose with real element parameters, not guessed ones.

Composition rules (from the design guide — read it, these are the load-bearing ones):

• Kinetic typography is the default for word-driven beats. Reveal lines in time with the voiceover using text animation_setting presets — slide, pop, masked_reveal, typewriter, with apply_to_unit: 'word' or 'line'. Swap old lines out as the script moves on; never pile lines into a wall of text.
• Keyframes are the main source of motion. Bars that grow, cards that slide and settle, highlights that travel — native text/rectangles/images with position + size keyframes look expensive and stay on-brand. Vary entrance direction and easing between elements.
• Nothing static. If a frame would hold unchanged for more than a beat, add a reveal, a swap, or a slow drift.
• Vary the technique beat to beat. Real screenshots for product beats (upload_file + check_uploads), stock only for generic atmosphere (find(type='videos'|'images', source='stock', query='…') — one type per call), generate_media(kind='animation') only when the motion is genuinely beyond keyframed elements. First ask: can this beat be a few rectangles and text with keyframes? If yes, build it that way.
• Size type for video, not for a web page.

5. Voiceover and sync

• Pick a voice with set_voice (list options with find if the user has a preference).
• Generate narration for all beats in one voiceover_batch call.
• Align visuals to the narration: auto_sync first, then add_sync_point for any reveal that must land on a specific word. Each on-screen line should appear as it's spoken (start_time on that VO beat), hold while relevant, then exit.

6. Verify before export

For each clip, render a mid-beat frame with get_clip(render=true, timestamp=<middle>) and check:

• Text is legible at video scale and inside safe margins.
• Generated animations sit inside their boxes.
• Palette is consistent — every color traces back to the committed palette.
• Total duration is within 30-60s.

Fix issues with update_elements / update_clips / update_audio, re-render, then export_project. Report the export link to the user.

Fallbacks

• No clueprint match → design guide + committed palette (step 2).
• User has no screenshots for a product beat → ask for them; if none exist, use an abstract keyframed composition for that beat — do not fake their UI.
• estimate_duration says a beat is too long → cut words, not pace.
• Upload stuck processing → poll check_uploads; if it fails, continue with the remaining beats and tell the user which beat needs their asset re-uploaded.
• A generated animation renders off-box or illegible → re-render the frame after tuning animation_duration / box size via update_elements; if still wrong, replace the beat with keyframed native elements.