Most people talk to AI music tools the way they talk to search bars: “lo‑fi for studying”, “chill background”, “cinematic ambient”. The model does its best, pulls something from the giant cloud of “lo‑fi” and “cinematic”, and you end up with a track that’s technically on‑label but emotionally generic. It sounds like the playlist everyone else is using because, at the prompt level, you literally asked for the same thing. The way out is not to learn “secret prompt hacks”; it’s to stop writing prompts as tags and start writing them as small, specific snapshots of your life, your channel, and this particular moment. When you describe “this exact evening” instead of “lo‑fi”, the model has a chance to sound like you instead of a preset. imagine
Modern text‑to‑music systems respond much better when a prompt reads like a short creative brief instead of a vague query. In practice, that means answering four questions in one or two sentences: what the music is for, what it should roughly sound like, what it should make people feel, and what must be avoided. When you add your own context on top of that — time of day, how tired or wired you are, what your audience expects from you — the AI stops guessing and starts following. This post is about turning “lo‑fi for studying” into “this exact evening in your head”, and using that shift to make AI music that actually feels like your channel.
Part 1: Why generic prompts always sound like someone else
If you scroll through any prompt guide in 2026, you’ll see the same advice on repeat: add genre, mood, instruments, maybe tempo; avoid one‑word prompts; describe feelings in plain language. That’s good for getting something usable, but it still tends to produce tracks that could belong to anyone. “Soft piano with warm pads, peaceful background for focus” is a nice sentence, but there’s nothing in it about you, your audience, or your reality.
The reason is simple: tags describe categories, not identities. “Lo‑fi for studying” captures a vibe thousands of other people share. AI models trained on huge datasets of background music will map that onto whatever the average “study beats” track looks like: dusty drums, soft keys, a bit of vinyl noise. If that’s all you say, you get “average lo‑fi” back by design. To make the output feel personal, the text has to move away from category labels and into specific context. Instead of “chill sunset vibe”, you talk about “Friday evening after a brutal week, windows open, you finally closing your laptop and cleaning your Notion board”. That’s not poetry for the model; it’s free metadata about pacing, density, brightness, and mood changes. imagine
The good news: you don’t need to write novels. You just need to add one or two details that no generic playlist prompt would ever contain, and anchor them to a clear use case. Modern guides explicitly recommend starting with purpose (“what is this for?”) and then layering in mood, genre, instruments, and constraints; the identity piece is your extra layer on top.
Part 2: Turning “lo‑fi for studying” into “this exact evening”
Let’s take the classic prompt: “lo‑fi hip‑hop for studying”. It’s fine as a starting point, but it doesn’t know:
- who is studying (you, your audience, your character)
- when this is happening (morning, late night, last sprint before exam)
- what the mental state is (calm, anxious, exhausted but wired)
- what the track is actually doing (under voiceover, standalone playlist, loop for a stream)
A more personal, still compact prompt might look like:
“Slow, warm lo‑fi track for a Friday night deep work session after a long week. Soft, dusty drums and gentle keys, no vocals, no big melodies, almost like a quiet kitchen radio in another room. Designed to sit under my late‑night coding stream without stealing attention, steady and slightly sleepy, this exact evening.”
Notice a few things. Purpose is explicit (“deep work session”, “under my coding stream”). The emotional state is anchored in time and body (“Friday night after a long week”, “slightly sleepy”). Hard constraints (“no vocals, no big melodies”) are in the same breath as imagery (“quiet kitchen radio in another room”). This isn’t just “lo‑fi”; it’s your night, with a job description.
This style of prompt works especially well in brief‑first tools like SonGo, because their whole interface assumes you’ll paste real sentences rather than just keywords. Instead of filling a “mood” dropdown, you drop in that one paragraph and let the model interpret it into tempo, instrumentation, and arrangement. If it’s too busy, you tweak a line (“even fewer notes”, “barely‑there melody”) and regenerate. You’re not chasing the perfect combination of sliders; you’re refining the story of this track. If you want a low‑risk way to try that flow, you can use the Dev‑specific trial link SonGo free for 3 days and dedicate it to turning your usual generic prompts into “this exact evening” versions for your main content formats.
Part 3: A simple four‑question template for “sounding like you”
If writing prompts from scratch feels heavy, it helps to turn the idea into a template you can reuse. Drawing from current best‑practice guides, an effective AI music prompt usually covers genre context, mood, instrumentation, purpose, and constraints. To make it “you‑shaped”, add two extra elements: personal context and one thing you absolutely don’t want.
Here’s a compact four‑question template you can keep in a doc:
- What is this for? (exact use case: “background for a YouTube tutorial about X”, “intro for my podcast”, “loop for a 2‑hour study stream”)
- What does this moment feel like? (time of day, energy level, emotional state in human words)
- Roughly what should it sound like? (genre, key instruments, density; 1–2 clear cues)
- What must never happen in this track? (vocals, big drops, certain clichés you hate)
Combine the answers into one or two sentences. For example:
“Soft, slightly melancholic piano and light electronic textures for a Sunday evening reflection vlog about burnout. Feels like cleaning your desk with a cup of tea after a draining week, calm but not numb. No vocals, no dramatic rises, no epic trailer drums, just gentle movement underneath my voice.”
or:
“Upbeat but not cheesy electronic groove for a short product launch reel. Think confident walking tempo, muted bass, subtle claps, something you’d hear in a modern hardware store demo rather than a festival. No huge drops, no cheesy synth leads, no whistle hooks.”
This way of writing lines up perfectly with how text‑to‑music engines are now tuned: they expect a clear description of use case, mood, and sonic texture, plus at least one constraint. You get to sneak your personality in through the images (“cleaning your desk with a cup of tea”) and the specific clichés you refuse to accept.
Again, a brief‑centric generator like SonGo is a natural place to paste these paragraphs. You can even keep a small library of them — one per recurring format — and then adjust only one detail per generation (e.g., time of day or instrumentation). Over time, this becomes your private prompt bank, and the sound of your channel grows out of your own sentences instead of out‑of‑the‑box presets. Keeping that doc near your SonGo entry point — for example, opening it alongside SonGo free for 3 days — makes it easier to stick to this pattern instead of falling back to “lo‑fi for studying” at the last minute.
Part 4: Iterate like a human, not a search engine
Even with better prompts, the first generation won’t always be perfect. The difference between “sounds like me eventually” and “generic forever” is how you react when the model misses. A lot of creators just hit “random” again or add more adjectives at random. The teams that actually win with AI music treat this as a feedback loop: generate, diagnose, change one variable, regenerate.
The key move is to react in sentences, not slider tweaks. If the track feels too bright and plastic, don’t just type “warmer”; say “less high‑end sparkle, more muted tones, like an older recording in a small room”. If the chorus comes in too big, add “no big chorus section, keep intensity almost flat, like a loop”. If the groove is right but the lead is annoying, say “keep rhythm and tempo, remove lead melody, replace with soft pads only”. This style of revision matches how prompt‑driven systems are documented to respond best: constraints and specific changes instead of vague “make it better”.
Brief‑first tools like SonGo shine in this phase because they encourage you to keep everything in text. You start with a “this exact evening” prompt, listen, then tweak one or two lines based on what you disliked, and hit generate again. Two or three iterations per track is usually enough. If you do this during a focused window — say, a few evenings using SonGo free for 3 days — you won’t just get a handful of tracks done; you’ll learn which parts of your language the model picks up on and which need to be more concrete. After that, each new prompt you write will be more “you” from the start.
Top comments (0)