Video summary

Я купил почти все ИИ для музыки! Жесткие выводы

Main summary

Key takeaways

Technology

Summary (Tech + Product/Review/Guide Takeaways)

The speaker reviews AI tools for music production he has been subscribing to for roughly two years (about $300–$400/month). He tests tools, cancels and re-subscribes, and aims to provide an “objective” view of what AI helps with versus what it turns into new routine.

He frames everything around ROI (cost vs. return) and repeatedly argues that while AI can generate components quickly, it often fails at the core professional tasks—especially where sound quality and musical logic matter.

Product/Feature Evaluations by Workflow Stage

1) Lyrics / Text Generation

  • Lyrics are mainly generated using GPT Chat or Suno.
  • He finds the Suno text workflow “complete trash” (especially in Russian), because it produces formulaic nonsense.
  • With GPT Chat, he doesn’t accept raw phrases. Instead, he starts from his own idea and uses what he calls “creative ping-pong”:
    1. Define dramaturgy/history and rhyming fragments
    2. Iterate prompts
    3. Reject most output (~80% garbage), keeping only ~10–15% as inspiration
  • Key criticism: delegating authorship/crucial creative thinking to the machine creates a crisis of meaning/self-identification for him.

2) Melody Generation (Suno-Related Critique)

  • He does not use Suno as a melody generator “from scratch” (i.e., input text → press button → get melody).
  • Instead, he uses Suno more like an arranger/vocal performer: he provides a melody and Suno performs/sings what he created.

Core argument against “melody-on-demand”:

  • Melody is the uniquely identifying authorial element and reflects the composer’s emotional state and “logic.”
  • The machine doesn’t understand context, cultural codes, and character/world logic, so it generates melodies that don’t match the intended meaning.
  • He criticizes a model limitation: Suno treats lyrics as just letters, lacking narrative/semantic understanding.
  • He also notes mismatches like AI choosing Western melodies in a Russian context that don’t work.

3) Arrangement (Where Suno Performs Best—Still Not Enough)

He says Suno shines at arrangement, because you can upload your melody and it can make “everyone plays, everyone sings.” In simple or average cases, it can replace an arranger.

However, he reports a downside:

  • Repeated use makes songs sound the same across outputs, creating a loss of individuality and yielding mostly “average” results.

Additional notes:

  • Even with updates like “advanced splits” that divide into stems, he still dislikes how the stems sound.

Analogy / implication:

  • He compares sound-generation inconsistency to smartphone improvements: post-processing gets better, but the core “camera principle” doesn’t fundamentally change—implying sound quality hasn’t fundamentally improved.

Bottom line for arrangement: He argues AI output is not truly an arranger tool. Professionals still need to do the heavy work; AI is useful only for occasional “interesting bits,” because otherwise it ruins the mix.

4) Mixing / Sound Engineering Assistants

Mixing is described as the most disappointing area:

  • Mixing assistants make things worse—cuts when they should raise, raises when they should cut.
  • The result “kills the sound,” especially for larger projects (he cites 70+ tracks as typical for pro projects).

He concludes these tools are essentially advertising with inflated expectations and can’t handle complex projects reliably.

5) Mastering (Online Mastering Critique + “Volume Scam” Claim)

  • He initially liked the UI/slider (before vs. after) and paid about $30.
  • Then he discovered:
    • The mastered version was quieter by ~1 dB
    • His tracks were almost 1.5× quieter than others
  • After loudness alignment, the sonic difference became almost indistinguishable, suggesting the “magic” is mostly EQ/processing that isn’t worth it.

He further claims:

  • The website misled users using before/after comparisons that depended on volume differences.

Conclusion:

  • He calls it banal deception based on loudness, unsubscribed, and doubts he’ll recover the money.

6) Vocal Generation (Positive Standout)

  • He actively uses vocal generators (mentions “Audio mi”).
  • Only a few models sing Russian acceptably, but overall he finds it helpful.
  • Use cases include:
    • Producers who don’t want to hire singers
    • Generating vocals “on the fly,” including backing vocals (thirds/choruses)

He calls it a real breakthrough because vocal layering is typically labor/time-expensive.

Overall Conclusion (ROI + “Why AI Doesn’t Replace Professionals”)

  • He argues AI adoption should be judged by ROI, not hype.
  • Example of sunk cost: generating a good intro/verse/chorus/bass often requires many generations (20–30), and combining “the best parts” still leads to heavy manual editing:
    • Stems drift out of sync in the DAW (Logic/Studio One)
    • He must cut/align every few measures
    • Groove disruption accumulates

He compares rebuilding with AI stems to patching low-quality audio instead of using real professionally mixed assets (e.g., splice/loops/sample packs).

Main philosophical/industry claim:

  • An arranger’s real tools are taste, ear, experience, and thinking—not just sound selection.
  • AI can produce material, but it doesn’t deliver the creative/professional decision-making that makes music work over time.

He closes with the sentiment that “miracles haven’t happened yet,” and that buttons/loops are not the profession.

Main Speakers / Sources

  • Main speaker: the video creator/reviewer (single host; mentions being an author and having students).
  • External sources mentioned: The Atlantic, including a journalism investigation about what Suno was trained on.

Original video