AI in the Creative Industry

From Idea to Result

How I actually use AI in creative work — hands-on, not theory

Obai Sukar — Co-Founder at NHJ

TechTown × aiSalon — Sunday, August 2, 2026 — 5:00 PM — Damascus

ASK THE ROOMHands up — who has already released something made with AI?

Good evening, and welcome to TechTown. I want us to start with a simple agreement: this session won't be a lecture defining artificial intelligence, it won't be a list of tool names, and I won't show you polished results I prepared in advance and then claim we made them together. We're going to do something harder and more useful. We'll start with a completely new idea from this room, and try to turn it into a short project with image, motion, sound and music. We'll see what works, what fails, how we tell the difference, and how we fix the result when it's weak. I've worked in audio and production for more than twenty-five years, and today I run my creative work with a mix of human experience and AI tools. So I won't talk about the tool as a miracle, and I won't treat it as an enemy. I'll talk about it the way I actually use it: a powerful tool inside a working system — not a replacement for the person holding the vision.

?

What has actually changed?

Four Real Shifts

Reaching the first version

What used to take days can now start within hours.

Team size at the experiment stage

A small team can test an idea that used to need more specialties.

The cost of trying

Prototyping and testing alternatives has become far cheaper.

The value of creative judgment

The faster execution gets, the more selection, review and rejection are worth.

ASK THE ROOMWhich of these four actually changed how you work?

I see the real change in four places. First, the time needed to reach a first version collapsed. I don't mean the final version — I mean the first thing we can see or hear and discuss. That matters, because an idea in our heads never reveals its flaws; a first version does. Second, a small team can now test ideas that used to require more specialties. A writer can see a visual concept, a designer can hear a sound concept, and a director can build a quick model before spending a production budget. Third, the cost of trying dropped. We can now test three angles or five styles before choosing. But cheap attempts can also produce a huge pile of mediocre material if we don't know when to stop. Which brings us to the fourth shift: the value of creative judgment went up. When execution is available to everyone, the advantage is no longer pressing the button — it's knowing which result deserves to stay, which must be rejected, and why.

?

Does AI create the idea?

The Tool Executes What You Can Describe — and Judge

Vision

What do you want the audience to feel?

Craft knowledge

Which decisions produce that result?

Judgment

What worked, what failed — and why?

Execution & iteration

This is where AI expands your reach and speed.

AI doesn't read your mind; it works on the decisions and references you give it.

ASK THE ROOMYour last prompt: were you deciding — or just hoping?

My answer: the tool doesn't execute what we imagine inside — it executes what we can turn into decisions, and then judge. Vision comes before the prompt. Before I describe a lens or a light or a style of music, I have to know what I want the audience to feel. Tension? Calm? Should the image feel human and close — or engineered and decisive? Then comes craft knowledge. In music, it's not enough to say "I want something sad." I need to know how rhythm, arrangement, harmony, space and performance create that feeling. In image, "cinematic" isn't enough; I have to understand composition, lens, light direction, color, and camera movement. Then judgment. The tool may give me a beautiful result that doesn't serve the idea — or a less flashy one that's more honest. The tool cannot make that call for me unless I can see the difference myself. Execution and iteration is where these tools genuinely excel: generating possibilities fast, compressing repetitive work, getting me from idea to reviewable material. That's why I say: without a visual, sonic or narrative vocabulary of your own, you'll mostly get surprises. Some of them beautiful — but not necessarily extensions of a vision that existed before the result.

Creativity Is a Loop, Not a Flash

Idea

→

Questions

→

Prototype

→

Review

→

Release

→

Learn ⟲

Then we return to the start — as a better version of ourselves.

ASK THE ROOMWhere does your process usually stall?

The Council: A System of Roles, Not One Answer

The Maestro

Sets the plan and splits the task.

The Executor

Builds the file, the code, or the model.

The Critics

Test the idea from opposing angles.

The Verifier

Checks the details before anything ships.

ASK THE ROOMWho — or what — is the critic when you make something?

I don't treat AI as one person I ask, who then hands me the truth. I treat it as a set of roles. There's the Maestro: understands the goal, splits the work, decides what happens first. The Executor: writes the code, builds the file, makes the model. The Critics: one hunts for gaps, another argues the opposite position, a third asks whether the idea is even clear to an audience that doesn't share its context. Then the Verifier — who doesn't care how beautiful the idea is, only that the numbers are right, the files work, and the claims don't outrun the evidence. These roles might live inside one tool, or be spread across several. What matters is that the voice producing the work is never the same voice approving it without resistance. This is how good production houses work: the writer doesn't review alone, the editor doesn't lock the first cut, and the sound engineer doesn't mix and deliver without checking on more than one listening system.

?

How many questions do I ask before accepting a result?

AI Is a Drilling Tool, Not an Oracle

What is this idea missing?

Where could it fail?

What's the opposite angle?

What's the evidence — and how certain is it?

The first answer is material for review — not a final verdict.

ASK THE ROOMBefore you trusted the AI — what did you ask it last?

There is no sacred first answer. I use the tool to drill deeper into an idea, not to stop at the first clever-sounding phrasing. I ask it: what's missing here? Then I ask it to imagine the ways this could fail. Then I demand the completely opposite angle — because these tools tend to agree with however you framed the question. And when the topic involves a fact, a number, or a technical judgment, I ask for the evidence and the degree of certainty, then verify with a proper source. There's an important distinction between error and uncertainty. An execution error means I asked for something specific and it wasn't delivered as asked — that needs fixing. Declared uncertainty means the tool doesn't have enough evidence and says so clearly — that's not a scandal; it's a signal that the decision belongs to a researcher or a specialist. The more important the result, the more it needs a second or third voice reviewing it. Not because AI is bad — but because review is a core part of any professional work, machine or no machine.

?

What separates describing a mood from a musical decision?

Clear Words, Missing Decisions

The generic prompt

A sad song, but not typical —
no clichéd nostalgia.

What didn't we specify?

The key or mode.
Meter and tempo.
The chord progression.
Instrumentation and arrangement.
Performance style and spatial texture.

The result: the machine will make most of the decisions for you.

ASK THE ROOMWhat's the first thing you'd put back into this prompt?

This sentence sounds specific: we want a sad song, but not a typical or cheaply sentimental one. Yet it contains not one musical decision that can be executed precisely. We didn't specify a key or a mode. We didn't set a meter or a tempo. We didn't say what the progression is, how the notes are voiced inside the chords, what the instruments are, what the space feels like, or how it should be performed. The tool will fill those gaps with its own expectations. We might like the result — it might even be more beautiful than what we imagined — but let's be honest: most of the decisions in that case were not ours. That doesn't make generic prompts wrong. They're fine for exploration, for discovering an unexpected direction. But there's a difference between exploring by chance and steering the work toward a vision whose elements you actually know.

The Knowledge That Makes the Difference

The key

D Major

The progression

Gmaj7 → Dmaj7
IVmaj7 → Imaj7

The voicing

A low root on beat one, then the upper chord on beats two and three. F# is the shared anchor tone.

The rhythm

3/4, very slow. One chord per bar.

The progression does reach the tonic — but Dmaj7 keeps the landing open and translucent.

Now we turn the mood into decisions. The piece is in D Major. The progression repeats between Gmaj7 and Dmaj7 — the fourth degree to the first, both as major-seventh chords. Let's be precise: this motion does arrive harmonically at the tonic chord, so it's wrong to say it never resolves. But it never gives a hard, triumphant cadence either, because Dmaj7 contains the note C#, which keeps the color open and translucent. The voicing starts with a solitary low root on beat one: G, then D. The upper chord follows on beats two and three: B–D–F# over G, then A–C#–F# over D. F# stays on top as the shared tone, binding the two chords and softening the transition. The meter is 3/4, very slow, one chord per bar. And the space in time between the bass note and the upper chord is part of the expression — not a gap that needs filling. This is what craft knowledge means: we're no longer saying "sad and beautiful." We now have something we can attempt, evaluate by ear, and correct.

Turning Knowledge into Executable Instructions

[Intro] [D major] [3/4, very slow]
[Instrumentation: sparse cello and piano]
[Harmony: Gmaj7 to Dmaj7, one chord per bar]
[Voicing: low root on beat 1, upper chord on beats 2–3]
[Mood: melancholic, suspended, understated]
[Vocals: distant, wordless choir]

A precise prompt increases your steering power — it doesn't guarantee literal musical execution.

With these instructions, the tool now holds far more decisions: key, meter, tempo, instruments, progression, harmonic rhythm, voicing, mood, and the nature of the human voice. But let's not oversell it. A music generator is not notation software, and not a DAW executing MIDI literally. It may honor the mood and the instruments, then change the progression or the voicing. So I listen to the result as an attempt — not as proof that the prompt was executed to the letter. If I need a specific harmony that must not change, the right path is to build it in the DAW or via MIDI, and use AI in the spaces that welcome chance: color, performance, an initial arrangement, exploration. The point isn't that longer prompts are always better. The point is that every detail should represent a decision I can explain. That was sound. Now let's test the same principle on the still image, then on motion — the rule stays the same; only the craft changes.

LIVE BREAK

Now — try it in Suno

VPN on. Open suno.com. Paste the prompt.
Hit generate. Two minutes.

What did it keep — and what did it rewrite without asking?

?

What separates describing a mood from a visual decision?

Clear Words, Missing Decisions

The generic prompt

A sad, cinematic photo of a person standing in the street — nothing clichéd.

What didn't we specify?

Camera angle and height.
Focal length: compress the background, or open it up?
Light direction and temperature.
Depth of field.
Composition: centered, or leaving negative space?
The reference color palette.

The result: the machine decides for you — and the image is "cinematic" only in the generic sense.

The Knowledge That Makes the Difference

Composition

The figure on a third of the frame, negative space opposite the gaze — the emptiness itself carries the feeling.

Lens

85mm, shallow depth of field — separating the figure from the background without fully isolating it.

Light

A cool rim light drawing only the edge, no fill — the mystery is part of the feeling, not a defect.

Color grade

Low contrast, desaturated broken tones; not full black-and-white, not vivid color.

ASK THE ROOMLook at this image: what did we still leave to chance?

This is craft knowledge in the image — the exact counterpart of what we did in music. Composition is a decision, not an outcome: I place the figure on a third of the frame and leave the empty space opposite the direction of the gaze, because in this kind of picture the emptiness carries as much feeling as the figure. I choose an 85mm lens with shallow depth of field, so the figure separates from the background without being cut out of it. I decide the light is a cool rim from behind, drawing only the edge: what the image doesn't show is part of what it means. And I choose a desaturated, low-contrast grade — because untheatrical sadness doesn't shout in color, and doesn't collapse into ready-made black-and-white either. Every one of these items can be reviewed, criticized and changed. That is the difference between describing a mood and making a visual decision.

Turning Knowledge into an Executable Prompt

[Subject: single figure, back three-quarter view, street at dusk]
[Composition: rule-of-thirds, negative space toward gaze direction]
[Lens: 85mm equivalent, shallow depth of field]
[Lighting: cool rim light only, no fill, low-key]
[Color: desaturated, broken tones, low contrast]
[Mood: melancholic, restrained — no visible tears, no rain cliché]

A precise prompt increases your steering power — it doesn't guarantee literal visual execution.

LIVE BREAK

Now — make the image

Freepik, Ideogram, or whatever you use.
Paste. Generate. Two minutes.

Which choice did the model steal from you?

?

What separates describing a mood from a motion decision?

Clear Words, Missing Decisions

The generic prompt

Animate the image of the person standing in the street — make it cinematic and emotional.

What didn't we specify?

The type of move: push in, pull out, slide — or hold perfectly still?
Speed and duration.
Which element stays locked as the anchor?
The easing curve.
Does the move serve meaning — or is it just decoration?

The result: the camera moves — but the meaning doesn't necessarily change.

The Knowledge That Makes the Difference

Move type & function

A very slow push-in builds quiet tension; total stillness is sometimes stronger than any move.

The anchor

One element in the frame stays visually locked for the whole shot.

The motion curve

Soft ease-in and ease-out; robotic linear motion is what exposes a generated shot.

Shot duration

Measured in seconds — not in vague description.

Turning Knowledge into Executable Instructions

[Shot: slow push-in, static wide to medium]
[Anchor: figure remains centered, background drifts subtly]
[Duration: 4 seconds, ease-in-ease-out, no linear motion]
[Camera: no pan, no rotation — single axis only]
[Function: builds quiet tension, not spectacle]

A precise prompt increases your steering power — it doesn't guarantee literal motion execution.

LIVE BREAK

Now — animate it

Kling, Luma, or your tool. Still + prompt.
Generate — we'll talk while it renders.

Did the move carry the idea — or just dress it up?

?

When does the result become an extension of my vision?

More Decisions, Less Chance

Direct

Define what you want — and why.

Compare

Test several options against a clear standard.

Reject & revise

Don't let liking a result erase your vision.

This is a standard for creative contribution — not a legal ruling on ownership or copyright.

ASK THE ROOMWhich decisions should always stay yours?

For me, my relationship to a result isn't measured by how many words I typed into the prompt — it's measured by how many conscious decisions I made along the way. Did I define the emotional goal? Did I choose the reference? Did I know why I rejected the first version? Did I compare results against a standard, or just pick the flashiest one? Did I revise until the result approached a vision that existed before it appeared? The more of those decisions there are, the less the outcome owes to chance — and the closer it comes to being the extension of a clear human vision. But this is a conversation about creative contribution, not law. Commercial ownership, terms of use, and copyright eligibility are separate questions that vary by tool, paid plan, country, and the share of human contribution. We shouldn't confuse the two. And I'm not saying simple prompts are wrong — they're great for exploration. I'm only saying: let's be precise about the role we played, and the role we left to the machine and to chance.

?

What did my mistakes teach me?

Four Mistakes That Built My System

I built the same system six times

The fix: one central registry, and formal archiving for every retired system.

I believed the first answer

The fix: independent reviews, and a clear standard for error vs. uncertainty.

The character changed from scene to scene

The fix: a locked reference, locked specs, and every consistency tool available.

The voice became so perfect it died

The fix: deliberate variance, natural pauses, and a human take when needed.

ASK THE ROOMWhich of these four have you already done?

Mistake one: I built the same work-organization system six times, each time under a new name, each time convinced it was final. The problem was never the tool — it was that I loved the moment of building more than the discipline of using. The fix is one central registry showing what's active, what's paused, and what's archived. An old system doesn't linger as an open ghost, and doesn't get deleted on a whim; it's retired by an explicit decision. Mistake two: I believed the first answer. It looked tidy and confident, so I skimmed past it — and discovered precise errors later, at a much higher price. So now I separate who executes from who reviews, and I distinguish an execution error that must be fixed from a declared uncertainty that needs a source or an expert. Mistake three: my first multi-shot generated video — the character looked different in every scene. Similar wording was not enough. I needed a locked reference image, exact unchanging specs, and Reference or Seed or character-consistency features wherever the platform supports them. And even then, visual review stays mandatory. Mistake four: I used a generated voice that sounded perfect — clean diction, even rhythm, zero errors. It sounded dead. Life in a performance isn't chaos, but it isn't machine symmetry either. I added uneven pauses, small shifts in intensity and tone, and sometimes recorded a human take and blended it with the generated material. The lesson: a mistake isn't the opposite of a good system. Very often, it's the raw material a good system is built from.

The Real Production Pipeline

Idea

↓

Script

↓

Storyboard

↓

Design & animation

↓

Sound & music

↓

Edit & review

Every stage has a right tool — and every tool has a failure you should know before delivery.

ASK THE ROOMHonestly — which stage do you usually skip?

?

Can we build a new project in front of you — right now?

Yes — from zero, with no hidden pre-made result.

The Live Project

1	Pick the idea & write the creative brief	5 min
2	Build a 3–4 scene storyboard	10 min
3	Produce & animate the key shot	10 min
4	Design the sound & music	10 min
5	Fix the weak version	10 min
6	Edit & screen it	5 min

ASK THE ROOMCall it now: how much of what we're about to make will be AI — and how much us?

We'll split the work into six stages, each with a time limit. The goal is not to produce a broadcast-ready ad in fifty minutes. The goal is to watch an idea being translated into decisions — and to watch each stage expose a new problem. Stage 1 — idea and brief (5 min): I'll take only three words or moods from the room. Then we write, in front of everyone: "We're making a 15–20-second visual piece about [the idea], aimed at [the audience], that must leave the feeling of [the intended effect], through 3–4 connected scenes, without long dialogue." That sentence is our standard: if we produce a beautiful shot that doesn't serve it, we reject the shot — even if we love it. Stage 2 — storyboard (10 min): I give the tool the duration, audience, effect, scene count and constraints, and ask for three different structures with a comparison. I choose the one with the clearest transformation — not the prettiest images. Any scene that doesn't change the meaning gets cut. Stage 3 — key shot and animation (10 min): one reference image defines the character, wardrobe, color, lens and light; every other shot uses the same reference and the same locked specs. I separate content (changes per scene) from style (never changes). For motion, I define what moves, what stays, the camera direction and speed, the start and end. Stage 4 — sound and music (10 min): the music's function first — does it drive rhythm, build tension, leave room for sound design? Then tempo, meter, instrumentation, density, and the shape of the beginning and end. I listen critically, and I don't claim the generator executed anything literally. Stage 5 — fixing the weak version (10 min): we name the problems before touching any tool — at most three, ordered by impact. There's no good repair without a clear diagnosis. Stage 6 — edit and screening (5 min): a simple honest cut; we delete anything that adds no meaning.

What Will We Fix When the Version Fails?

A generic idea

No clear shift, no specific angle.

Beautiful frames, no story

Every shot works alone — none work together.

Weak visual consistency

The character, color, lens or place keeps changing.

Motion without function

The camera moves; the meaning doesn't.

Music bigger than the project

Huge trailer music smothering the idea.

A perfect, lifeless voice

Rhythm too even, diction too polished.

ASK THE ROOMWho's had an AI piece rejected by a client — and why?

Six Ideas I Want You to Keep

Creativity is a system, not a flash of inspiration.

The first answer is the start of review, not the end of it.

Craft knowledge turns mood into decisions.

Consistency comes from reference and discipline — not luck.

Mistakes are raw material for a better system.

AI speeds up execution — it doesn't excuse the human from judgment.

ASK THE ROOMWhich ONE idea from tonight hits your work on Monday morning?

AI Speeds Up the Work

Vision and signature stay human

Thank you — let's open the floor for questions

Obai Sukar — Co-Founder at NHJ · obaisukar.com

techtown. × aiSalon — Sunday, August 2, 2026 — Damascus

BEFORE YOU GOShow your favorite result from tonight to someone near you — and say what you'd change next.