@azerogluemin/ai-bootstrap 0.2.1

package/templates/skills/suno-prompt-engineer/knowledge/stable-audio.md

@@ -0,0 +1,102 @@
+# Stable Audio — Model Specification
+
+> Stability AI's audio generation model. Best for short loops, sound design, and background music.
+
+## Model identity
+- **Provider**: Stability AI (stableaudio.com)
+- **Generates**: Instrumental audio (vocals weak/limited)
+- **Output length**: Up to 3 minutes (Stable Audio 2.0)
+- **Audio format**: WAV / MP3
+- **Cost basis**: Subscription + commercial license tiers
+- **Released**: 2023, v2.0 in 2024
+
+## Strengths
+- Strong instrumental quality
+- Good for loops / samples (loop-perfect option)
+- Sound design / SFX capable
+- Background music / underscore quality
+- Audio-to-audio style transfer
+- Commercial license clarity (paid tier)
+- Stable, predictable output
+- Open weights (Stable Audio Open) — can run locally
+
+## Weaknesses
+- Vocals weak (basically not for songs with lyrics)
+- Less "musical structure" than Suno/Udio
+- Better for atmosphere than hooks
+- Multilingual not relevant (instrumental focus)
+- Less coherent for long-form pop song format
+
+## Prompt format
+
+### Text prompt
+Natural language description of desired audio.
+
+### Example prompts
+
+**Background music**:
+```
+ambient cinematic underscore, soft piano with sustained string pads, 
+slow tempo BPM 60, melancholy contemplative atmosphere, 2 minutes 
+duration
+```
+
+**Loop / sample**:
+```
+hip-hop boom bap drum loop, 90 BPM, dusty vinyl crackle, 4-bar loop 
+seamless, Dilla-style swing
+```
+
+**Sound design**:
+```
+sci-fi spaceship interior ambient hum, deep low frequency drone, 
+occasional electronic beep, 30 seconds
+```
+
+**Underscore for video**:
+```
+emotional cinematic underscore for documentary, slow piano motif over 
+sustained orchestral strings, building gradually, 3 minutes, no vocals
+```
+
+## Best uses
+- **Background music** for video
+- **Sound design** for film/games
+- **Loops** for sample packs (loop mode)
+- **Stems** for further production
+- **Atmospheric pads** and drones
+- **Royalty-free music** for commercial use
+
+## Avoid for
+- Songs with vocals (use Suno/Udio instead)
+- Hook-driven pop
+- Lyric-based work
+- Multilingual songs
+
+## Audio-to-audio feature
+Stable Audio 2.0 lets you upload reference audio + prompt to transform style. Useful for:
+- Variations on existing tracks
+- Style transfers
+- Remixing approaches
+
+## Best practices
+- Be SPECIFIC about instruments (it honors instrument requests well)
+- Use BPM tag (it honors well)
+- Use genre tag for production style
+- Specify duration explicitly
+- For loops, request "loop" or "seamless" in prompt
+- For SFX, describe environment + sound character
+
+## When to use Stable Audio vs Suno/Udio
+- **Stable Audio**: instrumental background, loops, SFX, commercial use clarity
+- **Suno**: full songs with vocals, multilingual
+- **Udio**: hip-hop / electronic instrumental beats, sample-quality
+
+## Commercial use note
+Stable Audio's paid tier provides clearer commercial licensing than competitors. For commercial projects (ads, films, products), this matters significantly.
+
+## Anti-patterns
+- Don't use for vocal-heavy work
+- Don't expect Suno-style "song" output
+- Don't ignore the loop mode for samplework (huge feature)
+- Don't expect non-instrumental success

package/templates/skills/suno-prompt-engineer/knowledge/suno-v4.md

@@ -0,0 +1,125 @@
+# Suno v4 — Model Specification
+
+> Suno's flagship model (pre-v5). Strong vocal coherence + lyric adherence. Best for structured songs.
+
+## Model identity
+- **Provider**: Suno AI (suno.com)
+- **Version**: v4 (often called "Chirp" internally)
+- **Released**: Late 2024
+- **Generates**: Full songs with vocals or instrumentals
+- **Output length**: Up to ~4 minutes per generation
+- **Audio format**: MP3 / WAV
+- **Cost basis**: Credit-based subscription
+
+## Strengths
+- Excellent vocal coherence (lyrics actually sung, not garbled)
+- Lyric adherence (sings what you write — mostly)
+- Strong genre adherence
+- Good structure handling ([Verse], [Chorus], [Bridge])
+- Multilingual (Azerbaijani, Russian, Turkish, English all good)
+- Section transitions handled
+- Production polish (commercial-sounding)
+
+## Weaknesses
+- Sometimes pronounces non-English oddly
+- Long-form coherence drops past 3 min
+- Outro fades sometimes abrupt
+- Specific instrument requests sometimes ignored
+- Vocal gender sometimes ambiguous if not specified
+- BPM not always exact
+
+## Prompt format
+
+### Two main fields with HARD CHARACTER LIMITS
+1. **Style of Music** — **MAX 800 characters (Suno UI hard limit)**
+2. **Lyrics** — **MAX 4500 characters (Suno UI hard limit)**
+
+> ⚠️ **HARD RULE:** Suno UI rejects prompts exceeding these limits. Always count characters BEFORE submitting. Trim aggressively if over — do not assume tolerance.
+
+### Style field example
+```
+indie folk with fingerpicked acoustic guitar, female vocal soft 
+intimate, melancholy atmosphere, BPM 80, key D minor
+```
+
+### Lyrics field with tags
+```
+[Intro]
+(soft guitar fingerpicking)
+
+[Verse 1]
+Walking through the empty streets tonight
+Stars above are burning out of sight
+
+[Pre-Chorus]
+And I think of you again
+Wondering where you've been
+
+[Chorus]
+Come back to me, come back to me
+The morning won't break without your light
+
+[Verse 2]
+...
+
+[Bridge]
+[soaring strings enter]
+
+[Final Chorus]
+[full instrumentation, harmonies stacked]
+
+[Outro]
+(fade with guitar)
+```
+
+## Recognized structure tags
+- `[Intro]`, `[Outro]`
+- `[Verse 1]`, `[Verse 2]`, etc.
+- `[Pre-Chorus]`
+- `[Chorus]`
+- `[Bridge]`
+- `[Hook]`
+- `[Refrain]`
+- `[Drop]` (for EDM)
+- `[Instrumental]`, `[Solo]`
+- `[Spoken Word]`
+
+## Inline production directives (parentheses)
+```
+(soft piano enters)
+(drums kick in)
+(harmony joins)
+(reverb-heavy guitar solo)
+(fade out)
+```
+
+## Best practices for v4
+- Keep Style field tight, specific (not vague)
+- Use ALL structure tags for clearer sections
+- Specify vocal gender explicitly ("male tenor", "female alto")
+- BPM in style helps but not always honored
+- Key signatures help mood (D minor = melancholy default)
+- For non-English, be patient — may need 2-3 tries
+- Avoid contradictions (e.g., "soft loud aggressive gentle")
+
+## Multilingual notes
+- Azerbaijani: works well, occasional mispronunciation of ə/ı
+- Russian: very good
+- Turkish: very good
+- English: native quality
+
+## When to use Suno v4 vs v5+
+- **v4**: production-ready stable, predictable
+- **v5+**: more experimental, longer form, but more variance
+
+## Suno-specific quirks
+- Cover art auto-generated (ignore for prompt purposes)
+- Stems available for paid tiers
+- "Persona" feature lets you save voice for reuse
+- "Cover" feature reinterprets existing audio
+
+## Anti-patterns
+- Don't put ALL details in Style field (use Lyrics structure)
+- Don't request impossible specifics ("solo by John Coltrane")
+- Don't expect exact BPM — give range tolerance
+- Don't expect lyric mistakes corrected (write phonetically clear)

package/templates/skills/suno-prompt-engineer/knowledge/suno-v5.md

@@ -0,0 +1,130 @@
+# Suno v5+ — Model Specification
+
+> Suno's next-generation model. Longer form, richer arrangements, more expressive — with more variance.
+
+## Model identity
+- **Provider**: Suno AI (suno.com)
+- **Version**: v5 and beyond (rolling updates)
+- **Released**: 2025+
+- **Generates**: Full songs with vocals or instrumentals
+- **Output length**: Up to ~8 minutes per generation (longer than v4)
+- **Audio format**: MP3 / WAV / stems
+- **Cost basis**: Premium credit tier required
+
+## Strengths over v4
+- Longer coherent generations (5-8 min realistic)
+- Richer arrangements (more instruments per song)
+- Better vocal expressiveness (emotion, dynamics)
+- Improved multilingual (Azerbaijani notably better)
+- Better section transitions
+- Higher fidelity audio (closer to studio quality)
+- Better instrument-specific requests honored
+- Better adherence to BPM / key
+- Stems output (separate vocal, drums, bass, etc.)
+
+## Weaknesses
+- More variance per generation (less predictable)
+- Sometimes too ambitious (overproduces simple briefs)
+- Costs more credits per generation
+- Can be slower to generate
+- Occasional "creative" departures from prompt
+- Stems quality varies
+
+## Prompt format
+
+### Two fields with HARD CHARACTER LIMITS
+1. **Style of Music** — **MAX 800 characters (Suno UI hard limit)**
+2. **Lyrics** — **MAX 4500 characters (Suno UI hard limit)**
+
+> ⚠️ **HARD RULE:** Suno UI rejects prompts exceeding these limits. Always count characters BEFORE submitting. Even v5+ does NOT increase these limits despite richer vocabulary tolerance.
+
+### Same structure as v4 (Style + Lyrics)
+But with richer accepted vocabulary and more honored directives within the same character budget.
+
+### Style field (richer accepted)
+```
+indie folk in Bon Iver tradition, fingerpicked acoustic guitar with 
+reverb-drenched ambience, female vocal soft intimate building to 
+falsetto in chorus, layered self-harmonies, melancholy autumnal 
+atmosphere, BPM 78, key D minor, 4-minute song form with cathartic 
+bridge climax
+```
+
+### Lyrics field — same tag system
+```
+[Intro - 8 bars]
+(fingerpicked guitar establishes)
+
+[Verse 1]
+...
+
+[Pre-Chorus - building]
+...
+
+[Chorus - full]
+...
+
+[Verse 2 - drums enter]
+...
+
+[Bridge - cathartic, harmonies stacked]
+...
+
+[Final Chorus - even bigger]
+...
+
+[Outro - guitar alone again]
+```
+
+## New v5+ features
+- **Bar count hints**: `[Intro - 8 bars]` honored more
+- **Dynamic directives**: `(building)`, `(soft)`, `(loud)`, `(climax)`
+- **Instrument-specific entries**: `(strings enter)`, `(drums kick in)` work better
+- **Persona/voice cloning**: stronger, more consistent
+- **Stems output**: separate vocal/drums/bass/other tracks
+- **Style references**: can point to artist styles more reliably (still risky)
+
+## Inline directive vocabulary (expanded)
+```
+(soft acoustic intro)
+(drums enter with brushes)
+(strings swell)
+(modulate up)
+(key change)
+(double-time feel)
+(half-time feel)
+(harmonies stack)
+(falsetto break)
+(spoken word interlude)
+(instrumental break - 16 bars)
+(big finish)
+(fade outro)
+(sudden stop)
+(quiet ending)
+```
+
+## Best practices for v5+
+- Use richer Style descriptions (it can handle complexity)
+- Use bar counts for important sections
+- Specify dynamic arc explicitly
+- Use multiple Verse / Chorus iterations to show evolution
+- Stems output for editable workflow (DAW import)
+- Run multiple generations — variance means best take varies
+- Save successful "Persona" for series/album work
+
+## Multilingual notes
+- Azerbaijani: notably improved over v4 (ə, ı, ş, ç pronunciation)
+- Russian: excellent
+- Turkish: excellent
+- English: native quality
+- Mixed-language verses: work better in v5+
+
+## When to use v5+ vs v4
+- **v5+**: longer songs, complex arrangements, multilingual deep
+- **v4**: simple structured pop, predictable output, lower cost
+
+## Anti-patterns
+- Don't expect 100% prompt adherence — variance is feature
+- Don't underdescribe (v5+ rewards rich Style field)
+- Don't ignore stems for serious projects (huge editability win)
+- Don't compare 1-take to v4 — generate 4-5, pick best

package/templates/skills/suno-prompt-engineer/knowledge/udio.md

@@ -0,0 +1,103 @@
+# Udio — Model Specification
+
+> Suno's main competitor. Stronger on instrumental detail, sample-quality output. Tag-based prompting.
+
+## Model identity
+- **Provider**: Udio (udio.com)
+- **Generates**: Full songs with vocals or instrumentals
+- **Output length**: 32 seconds per generation, extendable
+- **Audio format**: MP3 / WAV
+- **Cost basis**: Credit-based subscription
+- **Released**: 2024
+
+## Strengths
+- Tag-based prompting (very specific control)
+- High-fidelity instrumental detail
+- Good for sample-style output (loop-quality)
+- Strong genre adherence (with tags)
+- Can extend tracks indefinitely (32s + 32s + ...)
+- Strong on hip-hop, electronic genres
+- Audio quality often more "produced" sounding
+- Good for instrumental beat production
+
+## Weaknesses
+- Vocal coherence weaker than Suno (lyrics sometimes garbled)
+- Multilingual weaker than Suno (English-dominant)
+- Short initial generation (32s — must extend)
+- Extension can drift from original mood
+- Less intuitive for full-song writing
+- Lyrics adherence less consistent
+- Fewer structure tags supported
+
+## Prompt format
+
+### Tag-based system
+Tags entered like searchable keywords, comma-separated.
+
+### Example prompt
+```
+hip-hop, boom bap, jazz samples, vinyl crackle, BPM 90, key A minor, 
+introspective male vocal, 90s east coast aesthetic, MF DOOM influence
+```
+
+### Lyrics input (optional)
+- Can input lyrics for vocal versions
+- Less reliable adherence than Suno
+- Better for instrumental-only
+
+## Supported tag categories
+- **Genre tags**: "indie folk", "trap", "house", "ambient"
+- **Mood tags**: "melancholy", "uplifting", "aggressive"
+- **Instrument tags**: "acoustic guitar", "synth lead", "808 drums"
+- **Production tags**: "lo-fi", "polished", "vinyl crackle"
+- **Era tags**: "90s", "80s synth-wave", "vintage"
+- **Vocal tags**: "male vocal", "female vocal", "rap"
+- **Tempo tags**: "BPM 120", "fast", "slow"
+- **Key tags**: "key C major", "minor key"
+- **Reference tags**: "MF DOOM influence", "Burial style"
+
+## Best practices for Udio
+- 8-15 tags optimal (more = confusion)
+- Specific over vague ("808 drums" not just "drums")
+- Use BPM tag for tempo control
+- Use instrumental tags for beat production
+- Extend in 32s chunks for full song (re-prompt each extend if drifting)
+- Best for instrumental — vocal less reliable
+
+## Extension workflow
+1. Generate initial 32 seconds
+2. Click "Extend" → adds 32 more seconds
+3. Can adjust prompt for each extension
+4. Build to 3-4 minute full song
+5. Each extension = chance for drift
+
+## Multilingual notes
+- English: native quality
+- Spanish: decent
+- Other languages (Azerbaijani, Russian, Turkish): weaker
+- Best for English-language output
+
+## When to use Udio vs Suno
+- **Udio**: instrumental beats, hip-hop, electronic, sample-quality output
+- **Suno**: full songs with vocals, multilingual, structure-heavy
+- **Both**: try both for important projects, pick better take
+
+## Udio-specific features
+- Extension system (key feature)
+- Inpainting (regenerate specific sections)
+- Stems (paid tier)
+- Style transfer between generations
+
+## Best fit genres
+- Hip-hop / boom bap / lo-fi hip-hop
+- Electronic / house / techno
+- Ambient / drone
+- Jazz instrumental
+- Sample-based production
+- Beat production for rappers
+
+## Anti-patterns
+- Don't expect coherent multi-minute song from one prompt
+- Don't expect non-English vocals to work well
+- Don't use vague tags ("good music", "nice")
+- Don't ignore extension drift — re-prompt actively