prompt_h2501

Day 02 — Prompting for Music & Sound Generation

Duration: 2 h 30 min.
Focus: Crafting prompts for genre, mood, tempo, instrumentation, and structure; producing and evaluating short AI-generated tracks in a Eurovision-style contest (PromptVision).
Mode: Browser-only (web-based music models) — small teams of 2 (≈ 6).

Block	Activity	Duration
1	Foundations & Demos	20 min
2	Tools & Prompt Craft	20 min
	Break	10 min
3	PromptVision Creation Sprint	45 min
4	Listening Party & Voting	25 min
5	Wrap-Up & Takeaways	10 min
	Total	2 h 30 min

🧱️ Block 1 — Foundations & Demos

1. How Text-to-Music / Audio Models Work

• Key concepts: diffusion, autoregression, transformer sequence modeling.
• Modalities: text-to-music, text-to-sound (SFX), audio-to-audio (style/voice transfer).
• Audio pipeline: prompt → latent/audio tokens → decoding → waveform.

2. Prompt Parameters (Quick Demo)

• Structure: genre → mood → tempo (BPM) → instrumentation → structure (intro/verse/chorus/ending) → vocal style/lyrics (optional).
• References: mention artists/eras cautiously; better to describe features (e.g., “syncopated funk guitar”, “reverby 80s snare”).
• Iteration: change one variable per take to hear its effect.

🧱️ Block 2 — Tools & Prompt Craft

3. Tools Overview

• Text-to-Music: Suno.ai, Udio, MusicFX.
• Sound Design / Textures: Mubert, AudioLDM/AudioCraft.
• Editing/Finishing: BandLab, Soundtrap, Audacity (trim, normalize). 🔘 Run the same prompt on two tools → compare fidelity, vocals, and mix.

4. Effective Music Prompts (Mini Task)

• Start from a plain idea → refine with parameters:

  • “Ambient track” → “Ethereal ambient drone at 70 BPM with warm pad and field recordings, slow crescendo over 30 s.”
  • “Pop song” → “Uplifting synthpop at 118 BPM, side-chained bass, four-on-the-floor kick, female lead with vocoder doubles, verse–chorus–outro.”

• Students rewrite a prompt three ways (tempo/mood/instrument swap) and preview 1–2 takes.

— See teaching cues in Day 02 — Teaching Notes.

🧱️ Block 3 — PromptVision Creation Sprint (Contest)

Objective: Teams craft a 60+ s track that matches a shared theme and submit it with the final prompt text and tool used.

Structure:

• No fixed Theme.
• Each team gets ~45 min to iterate 2–3 times and export one final clip (normalize to consistent loudness).

Prompt checklist: genre, mood adjectives, tempo (BPM), instruments, structure/arc, vocal/lyrics (optional), mix notes (dry/wet, reverb).

👉 Full contest flow, judging rubric, and submission template: PromptVision Guide (to be added).

🧱️ Block 4 — Listening Party & Voting

• Play all tracks back-to-back (playlist or Padlet).
• Teams save read or project their prompt (or you read from instructor-only sheet).

• Voting:

  • 🏆 Best Song
  • 🥇 Audience Favorite
  • 🎧 Instructor Highlights

🧱️ Block 5 — Wrap-Up & Takeaways

• Discuss: Which prompt elements most affected structure and timbre?
• Note limitations (lyrics coherence, long-form structure).
• Ethical reminders for follow-up sessions (voice likeness, licensing, disclosure).

✅ Learning Outcomes

• Design prompts that control genre, mood, tempo, and instrumentation.
• Compare model outputs and iterate deliberately.
• Produce and evaluate a short track aligned to a theme and prompt design.

This site is open source. Improve this page.