Over the past few weeks, I’ve been collecting ai prompt examples across three very different media — images, music, and video. I used ChatGPT, Gemini, Suno AI, Kling AI, and Luma Dream Machine, all on free tiers, running dozens of prompts and documenting every result.
The biggest surprise wasn’t any single tool or trick. It was realizing that the same fundamental principles keep showing up regardless of the medium. A vague image prompt and a vague music prompt fail in the exact same way — and for the exact same reason.
This guide pulls together the ai prompt examples and universal patterns I discovered across all three media. This is a synthesis hub, not a full review of each tool — I’m pulling out the cross-media principles that repeated across my image, music, and video tests. If you want to go deep on any specific medium, I’ve linked to my detailed guides throughout. But start here — because the principles you learn in one medium will make you better at all of them.
The One Rule That Works Across Every AI Tool
After testing prompts on five different AI tools across three media, one pattern emerged every single time: when you’re vague, AI gives you the safest, most generic result it can produce.
Here’s what that looked like in practice:
Images: I typed “a beautiful sunset over the ocean” into ChatGPT. The result looked like a stock photo — technically fine, but completely forgettable. No specific colors, no interesting composition, nothing that felt intentional. The AI picked the safest version of “sunset” it could find.
Music: I typed “a dreamy love song with soft piano and gentle female vocals” into Suno. What came back was a generic pop ballad — pleasant enough, but not dreamy at all. The vocals were clear and forward instead of gentle or breathy. Suno interpreted “dreamy” as “soft and romantic” rather than “atmospheric and otherworldly.”
Video: I gave Kling AI and Luma the same cinematic coffee scene prompt. Both produced watchable results, but the steam I specifically requested was barely visible, and the “dreamy atmosphere” defaulted to “clean and well-lit.”
The pattern is always the same: subjective words like “beautiful,” “dreamy,” or “cinematic” give AI too many possible interpretations, so it picks the most average one. The fix is also always the same — replace mood words with specific, measurable descriptors.
“Beautiful sunset” becomes “golden hour sunset with deep orange and purple gradient, low sun casting long shadows across wet sand, dramatic cloud formations.”
“Dreamy love song” becomes “dream pop, ethereal, 72 BPM, lush ambient pads with heavy reverb, breathy female vocals, intimate bedroom recording feel.”
This isn’t just a prompting tip — it’s the foundation that everything else in this guide builds on.
AI Prompt Examples for Images — What Worked
Image generation is where I started, and it’s where the principles became clearest. I tested across ChatGPT and Gemini, comparing how each tool interprets the same prompt.
The Key Discovery
In my testing, I identified five building blocks that consistently produced better image results: Subject, Style, Lighting, Composition, and Exclusions. When I included all five, the results were dramatically more specific. When I left any out, the AI filled in the gap with its own (usually generic) choices.
The most telling test was iterative: I tried to recreate a real photo of a macaron cake, starting with a 7-word prompt and refining it across four rounds until I reached 97 words. Each round brought the result closer to what I actually wanted — proving that image prompts reward specificity in a very measurable way.
What Surprised Me
ChatGPT and Gemini interpret the same prompt differently. In my tests, ChatGPT was more literal and polished — it did exactly what I asked, cleanly. Gemini was more creative — it added elements I didn’t request, sometimes improving the result, sometimes missing the mark. Neither was “better,” but knowing their tendencies helped me write prompts that played to each tool’s strengths.
For ChatGPT-specific image prompts with practical before/after examples, see my Best ChatGPT Image Prompts guide. For the universal 5 Building Blocks framework that works across tools, see my AI Image Prompt Examples guide.
AI Prompt Examples for Music — What Worked
Music prompts work fundamentally differently from image prompts, and this caught me off guard. With images, adding more detail almost always helps. With music, the relationship between your prompt and the result depends on which fields you fill in.
The Key Discovery
I ran the same concept — a dreamy love song — through three levels of control on Suno AI:
Level A (Simple Mode): One sentence describing what I wanted. The result was a generic pop ballad with no dreaminess at all.
Level B (Custom Mode — Style field only): I specified genre, mood, instruments, BPM, and vocal style. The atmosphere improved noticeably — but the vocals almost completely disappeared. Even though I explicitly wrote “female vocals” in the Style field, both tracks came out as essentially instrumental pieces.
Level C (Custom Mode — Style + Lyrics with performance directions): I revised the Style prompt based on what Level B taught me, wrote my own lyrics with section tags like [Verse] and [Chorus], and added performance cues like “(whispered, intimate)” and “(full voice, emotional peak).” The difference was night and day — vocals returned, the song had structure, and the dreamy atmosphere finally came through.
The bottom line: in my testing, Suno created “music” with Style alone, but only created “a song” when I added lyrics. As a practical rule of thumb, the Style field seemed to control the atmosphere, while the Lyrics field was what reliably turned a musical idea into a vocal song with structure.
What Surprised Me
I also tested five different genres (Pop, Lo-fi, Rock, Jazz, Electronic) and found that Suno handles structured genres like Pop and Electronic very consistently, but struggles with genres that depend on spontaneity — Jazz lacked improvisation feel, and Rock emotions felt over-the-top.
Want the full breakdown? See my complete guide: Suno AI Music Prompt Examples — including the 3-level comparison, genre-specific templates, and the Cover feature.
AI Prompt Examples for Video — What Worked
Video was the most challenging medium to test — partly because of tighter free-tier limits, and partly because video prompts need to control something images and music don’t: movement over time.
The Key Discovery
I tested the same prompts on Kling AI and Luma Dream Machine, and the same pattern appeared in all four tests:
Kling thinks like a director. It handles camera movement, action sequences, and dynamic motion well. When I wrote “slow camera push-in,” Kling actually pushed in. When I described an earbud floating out of its case, Kling animated the floating.
Luma thinks like a cinematographer. It excels at lighting, texture, and visual realism. The same coffee scene that looked slightly “CG” on Kling looked like actual footage on Luma — warm golden tones, convincing wood grain, natural ceramic texture. But the camera barely moved.
This pattern held across product demos, cinematic landscapes, and SNS content — four tests, four times the same result.
What Surprised Me
The biggest surprise was that aspect ratio instructions in the prompt text were ignored by both tools. I wrote “9:16 vertical format” in my SNS prompt, and both tools produced landscape videos. You need to set the aspect ratio in each tool’s UI settings — the prompt field only controls content, not format. This is a key difference from image generation, where aspect ratio prompts usually work.
Want the full breakdown? See my complete guide: AI Video Prompt Examples — including side-by-side comparisons, use-case recommendations, and free-tier credit realities.
The Universal AI Prompt Structure — Works for Any Tool
After testing across all three media, I noticed that every effective prompt follows the same underlying logic — even though the specific elements change.
The Common Framework
| Element | Images | Music | Video |
|---|---|---|---|
| What | Subject | Genre | Subject |
| How it feels | Style | Mood/Energy | Style |
| Sensory detail | Lighting | Instruments | Lighting |
| Structure | Composition | Song structure (Verse/Chorus) | Camera movement |
| Specifics | Exclusions | BPM, vocal style | Duration, format |
The order matters too. In my testing across all three media, front-loading the most important element produced more consistent results. For images, lead with the subject. For music, lead with the genre. For video, lead with the subject and its action.
The 3 Principles That Apply Everywhere
1. Specific beats subjective. “Warm golden tones, 72 BPM, shallow depth of field” works. “Beautiful, dreamy, cinematic” doesn’t — because those words mean different things to every AI model. Google’s prompt engineering guide makes the same point: clear instructions and context lead to more accurate outputs.
2. Structure gives AI a framework. In images, it’s composition (rule of thirds, centered). In music, it’s section tags ([Verse], [Chorus]). In video, it’s camera instructions (push-in, drone shot). Without structure, AI makes generic choices.
3. Iterate, don’t expect perfection. None of my best results came from a first attempt. My best image took four rounds of refinement. My best music track came after I revised the prompt based on a failed attempt. My best video happened after I learned that UI settings matter as much as the prompt text.
Choosing the Right Medium — At a Glance
| Images | Music | Video | |
|---|---|---|---|
| Best to learn first? | Yes — fastest feedback loop | Second — needs Style + Lyrics thinking | Last — tightest credits |
| Feedback speed | Seconds | ~30 seconds | 1–5 minutes |
| Biggest free-tier constraint | Daily generation limit | None significant (50 credits/day) | Very tight monthly/daily credits |
| Most common beginner mistake | Relying on mood words (“beautiful”) | Leaving Lyrics field empty | Writing aspect ratio in prompt instead of UI |
| What surprised me most | Gemini adds elements you didn’t ask for | Vocals disappear without lyrics | Camera movement only works on some tools |
Common Mistakes That Waste Your AI Prompts
These mistakes showed up in every medium I tested. If you fix just these, your results will improve immediately.
Relying on mood words alone
“Dreamy,” “cinematic,” “epic,” “beautiful” — these words feel descriptive but they’re actually ambiguous. Every AI model interprets them differently. In my image tests, “dreamy” produced a soft-focus portrait. In my music tests, the same word produced a generic pop ballad. In my video tests, it produced a cleanly lit but static scene. The word meant something different to every tool.
The fix: Use mood words as a starting point, then back them up with technical specifics. “Dreamy” + “heavy reverb, wide stereo space, 72 BPM, breathy vocals” is much clearer than “dreamy” alone.
Confusing the prompt with the settings
This hit me hardest in video testing. I wrote “9:16 vertical format” in my prompt and expected a vertical video. What I got was a landscape video — because Kling and Luma both require format settings to be configured in the UI, not the prompt text.
But it applies to other media too. In my Suno tests, the “Style” field and “Lyrics” field served different functions — vocals became much more reliable once the Lyrics field had actual content, regardless of what I wrote in Style.
The fix: Before writing your prompt, check what the tool’s UI settings control versus what the prompt text controls. They’re not always the same. This aligns with what OpenAI’s prompting guide calls “understanding model parameters” — the prompt is only one part of what shapes the output.
Cramming too many elements into one prompt
I’ve seen prompts that try to specify ten elements at once. In my experience, both image and video AI struggle when overloaded — the result either ignores some instructions or produces a confused compromise.
The fix: Prioritize. Pick the 4-5 most important elements and be very specific about those. Let the AI handle the rest. You can always refine in a follow-up.
Which AI Medium Should You Start With?
If you’re new to AI prompting and wondering where to begin, here’s my honest recommendation based on testing all three:
Start with images. The feedback loop is the fastest — you type a prompt, see a result in seconds, and can immediately identify what worked and what didn’t. Free tiers on ChatGPT and Gemini are generous enough for plenty of experimentation. The principles you learn here (specificity, structure, iteration) transfer directly to music and video.
Move to music when you’re comfortable with structured prompts. Suno’s Custom Mode requires you to think about your prompt in two layers (Style + Lyrics), which is a good stepping stone to more complex prompting. The free tier gives you about 10 songs per day — plenty of room to experiment.
Try video last. Not because it’s the hardest conceptually, but because free-tier credits are the tightest. During my April 2026 tests, Kling’s free account gave me roughly 4 videos per day, and Luma’s monthly allowance worked out to about 8 videos total. These numbers may change, but the general pattern holds — video credits are scarcer than image or music credits, so every prompt needs to count.
FAQ
What’s the best way to learn AI prompting?
Start by testing. Reading ai prompt examples (including this guide) gives you the principles, but the real learning happens when you type a prompt, see the result, figure out what went wrong, and try again. I’d recommend starting with image generation on ChatGPT — it’s free, the feedback is instant, and you’ll develop an instinct for specificity that applies to every AI tool. Once you’re comfortable, move to music (Suno) and then video (Kling/Luma).
Do I need to pay for AI tools to write good prompts?
No. Every tool I used for this guide had a usable free tier. ChatGPT’s free tier generates images. Suno’s free plan offers 50 credits per day (~10 songs). Kling and Luma both offer free tiers with limited daily or monthly credits — enough for a handful of test videos each day or month. All free tiers have limitations — watermarks, lower resolution, no commercial use — but they’re more than enough to learn prompting and test ideas. I wrote all four of my detailed guides using only free tiers. Free-tier limits change frequently, so check each tool’s pricing page for the latest.
Final Thoughts
The most valuable thing I learned from testing ai prompt examples across images, music, and video isn’t a specific technique — it’s that the same thinking process works everywhere. Be specific instead of vague. Structure your prompt with the most important elements first. Understand what the prompt controls versus what the tool’s settings control. And expect to iterate.
If you take only one thing from this guide, let it be this: a prompt is not a wish — it’s a set of instructions. The more precise your instructions, the closer the result will be to what you actually imagined. That’s true whether you’re generating a sunset photograph, a jazz instrumental, or a product demo video.
Based on testing conducted in March–April 2026 using free tiers of ChatGPT, Gemini, Suno AI, Kling AI (v2.5 Turbo), and Luma Dream Machine. All observations reflect my personal experience — results may vary depending on tool updates and specific prompts.