Lesson 11. The AI assistant

Over the previous lessons you have learned to build light by hand. Now for the neural network as an assistant that takes the rough work off you: the AI assistant creates ready scenes, chases and presets from your description.

But this lesson is not about “which button is where”. The chat with the neural network has an intuitive interface. Let’s work out how to set the task so that the assistant produces something decent the first time. The AI is powerful, but it is a performer, not a mind reader. The quality of what it gives back depends directly on the quality of what you give it.

⚠️ The rule here is a banal one: rubbish in — rubbish out; precision in — precision out.

The set of fixtures and the way they are laid out is unique to almost every user. All of that has to be conveyed to the neural network properly. Happily, a great deal has been automated for you — what matters is supplying the source data correctly and in full.

⚠️ Imagine that you are the director and the assistant is a very fast lighting designer who follows the brief literally. Give a vague brief and it will fill in the gaps at random and miss by a mile. Give a precise one and it will hit the target first time.

Fig. 1 — “AI” mode: a wizard menu instead of an empty chat window

The three pillars of quality

We prepare the ground before turning to the neural network at all. Before composing a request, make sure of three things. Without them even a perfect command will not save anything.

Pillar 1. An accurate fixture profile

The assistant does not see your fixtures with its eyes — it works with their profile: which channels exist and what each of them does (dimmer, RGB, pan/tilt, gobo, strobe). And the key point: the assistant can do exactly what is written in the profile and not a drop more.

The pan and tilt channels are not marked as pan/tilt → it will not move the head.
The gobos are not written into the profile → it will not set a pattern.
The channels are nameless (“channel 5, channel 6”) → for the AI those capabilities simply do not exist.

💡 Tip: take profiles from the library or from a photo of the manual, then look into Patch and check that the channel types and the gobo names are set correctly.

Fig. 2 — A fixture profile in Patch: the channel types and gobos that the AI “sees”

With a relatively simple PAR everything is out in the open: the dimmer channel, the R-G-B channels — from a list like that it is immediately clear how it shines. With a complex fixture, though, bare DMX no longer tells you what it looks like in action: which of the “spider’s” beams is which, where each segment of the matrix points, what pattern hides behind a number on the gobo wheel.

The program — and the neural network with it — only sees “channel 7 equals 128”, but does not know that this means “the third beam has turned” or “the gobo is a star”. An extended profile adds what is missing: the geometry of the fixture (its shape, the arrangement of the beams) and the names of the gobo and colour wheel slots. Then the visualizer draws the fixture close to reality and the AI assistant understands what it is driving. This is not a quirk of the program but a shortcoming of DMX512 (although when it was created, fixtures like these simply did not exist yet).

⚠️ Extended profiles you build yourself in the editor (unless somebody has already created one and shared it with our community). Without an extended profile the neural network — and the visualizer — cannot conduct a complex fixture well, simply because it does not know its particulars! Which means it will either ignore those particulars or present them differently from how they really are. That is a shortcoming of the DMX512 protocol — when it was invented there were no fixtures as complex as today’s — so the extended profile editor is there to fill in the missing information.

Pillar 2. Accurate fixture placement

The assistant knows not only what a fixture can do but also where it stands and where it points: the height (floor / stand / truss / ceiling) and the direction. It relies on that when it builds spatial effects: “beams to the centre”, “a wave from left to right”, “a fan”, “brighter in the middle”.

⚠️ Careful: if the fixtures are all heaped in the default spot and the heads point anywhere at all, every spatial command turns to mush. That is especially critical for moving heads.

💡 Tip: before generating, lay the fixtures out on the 2D or 3D plan and set the heights and directions. Five minutes pay for themselves in the quality of every scene.

💡 The assistant reckons with your venue, not with a “typical room”. Along with the fixtures it receives the dimensions of the room and the actual hanging height of every fixture in metres — not “a truss in general” but your 6 metres or your 2.5. And when a scene says to aim the heads “into the room”, “at the stage” or “to the centre”, the program converts that into angles according to what stands on your plan: the podium, the crowd, the truss. Put the stage and the audience into the 3D view and the beams will land exactly there rather than at some notional point.

Fig. 3 — Right-clicking a fixture to set its placement and beam direction

Pillar 3. A precisely set task

And only now, when the profile is honest and the fixtures are laid out, does everything come down to what you ask for and how you word it.

Opening the AI assistant

“AI” mode — the button in the mode switch at the top, or Alt + 5. You are met by a clear wizard menu: there is no need to compose the perfect phrase from the doorstep — the assistant asks the main questions itself, step by step.

The “Create” section:

Button	What it does
🎨 Scenes (by mood)	Several static light pictures for the chosen atmosphere — your main tool
🔄 Chases from scratch	A running, dynamic sequence straight away
🔗 Chases from scenes	Builds a chase out of the scenes already in the project
🎛 Presets	Quick building blocks for a single parameter (wash in a colour, aim at the centre, rotation)

The “Analysis” section:

Button	What it does
🔍 Diagnostics	Looks through your scenes and points out where something is wrong (a fixture switched off, colours that fight, a forgotten dimmer)
💡 Explanation	Tells you how a particular feature of the program works

And you can also simply ask the assistant a question about any capability of SvetoMaster — it answers from the user manual. It is not only a generator but a built-in consultant.

Creating scenes with understanding

We press “Scenes by mood”. The assistant leads you step by step, like a wizard.

Step 1 — the mood. Twelve presets: Romance, Energy, Calm, Drive, Celebration, Anxiety, Minimalism, Multicolour, Cold, Warm, Neon, Monochrome. Behind each of them is a whole character — the palette, the brightness, the level of contrast. “Romance” means warm, muted tones; “Drive” means saturated colours and high contrast; “Minimalism” means restraint.

💡 The secret of quality: under the buttons there is a “Your own mood description” field — and it almost always gives a better result than a ready button. The button is a general direction; your description is a precise hit.

Fig. 4 — The “mood” step: 12 presets and the field for your own description

Step 2 — how many scenes to create. Step 3 — for which fixtures: for all of them or only for a group (“Heads”, for instance). Generating group by group gives you the control of building by hand. Step 4 — the character: dynamic, static or smooth.

⚠️ A common mistake that ruins the impression is a mismatch between the character and the music. For a slow ballad take “static” or “smooth” (otherwise the light fusses where quiet is needed); for a dance drop take “dynamic” (otherwise the light is limp where drive is needed). Choose honestly, for the real moment of the show.

The assistant thinks for a few seconds (generation happens on the server) and delivers finished scenes straight into the project — the result is visible in the visualizer at once.

The main skill: how to describe the task

This is where quality is born or dies. The assistant does not read minds. Everything you have not said explicitly it will fill in itself — that is, choose at its own discretion, essentially at random. And every such coincidence is a lottery that can spoil the scene:

you asked for “blue light” but said nothing about movement → the heads stood still, although you wanted them sweeping the room;
you said nothing about the strobe → it turned up at the wrong moment (or was missing when it was needed);
you did not say which fixtures to use → all of them came on, although only the backdrop was wanted.

⚠️ The more you leave “to the AI’s discretion”, the more of the result is down to chance. Describe it in full.

A good description answers several questions at once:

Question	An example of what to state
Palette	specific colours, or “warm / cold”
Brightness and contrast	“soft and even” or “sharp accents”
Movement	the heads stand still / sweep smoothly / dart about
Strobe	needed or not
Fixtures	which groups take part and which keep quiet
Tempo	“breathing slowly” or “living in time”

Compare:

❌ A bad brief: “make it beautiful” / “energetic”. There is nothing to grab hold of — it will be a lottery.
✅ A good brief: “cold, uneasy light, blue and white; the PARs wash evenly and dimly; the moving heads sweep their beams slowly across the room; no strobe; the mood is dark and tense”.

In the second case you left no room for chance — the assistant will build exactly what is in your head.

💡 The rule: one coherent task at a time, but described in full. Do not pile five ideas into one phrase — better to get a solid foundation from one clear brief and catch up on the details with refinement.

Refinement

This may matter more than the very first generation. The result almost hit the mark but you want it different? Do not rewrite the request from scratch. Under the answer there are prompt buttons:

🔥 Make it brighter · 🌙 Make it softer · ❌ Remove the strobe · 🎨 Different colours · 🎭 More contrast · 🎯 Focus on the centre · 🔇 Quieter, calmer · ⚡ Add movement · 🔁 Redo it completely.

You press the one you need and the assistant edits in exactly that direction, remembering the previous version. It is a dialogue, not a one-off order: you shape the result step by step, like a sculptor. Too bright → “softer”; dull → “more contrast”; the strobe keeps creeping in → “remove the strobe”. Three or four touches and the scene is finished.

“Create another” — a noticeably different version that leaves the accepted one alone: the assistant knows what you already have and deliberately does not repeat itself. Handy for generating several and picking the best.
Editing one scene in particular — right-click the scene → “Refine with AI” → describe it in words (with the same fullness): “make it colder and take the movement out”, “aim the heads into the room”.

Chases, presets, analysis — the same logic

Chases. You can start not “from scratch” but through a combination: use the assistant to make several good scenes, pick the successful ones, and build a sequence from them with the “Chases from scenes” command. That way you control every frame. “Chases from scratch” are good for quick dynamics (there you will be asked for the format: a short effect of 2–4 steps or a full chase of 5–10).

Fig. 10 — “Chases from scenes”: choosing ready scenes for the sequence

Presets — 12 ready building blocks for a single device (red wash, strobe pulse, movement left/right, cold/warm wash, rainbow, centre focus, dim out, rotation, UV, blackout flash). The bricks from which it is convenient to shape scenes by hand.

Diagnostics — set the assistant on the project: it will honestly point out the weak places and suggest improvements.
Explanation and questions — forgotten how something works? Ask in words and get an answer from the manual without leaving the program.

A working order for a consistently good result

Prepare the ground — honest fixture profiles and a layout with heights and directions.
(Preferably, though not always) scenes first, then a chase from them. That said, “chases from scratch” with a well-set task also come out quite decent.
Describe the task in full, leaving the AI no room for chance, one coherent idea at a time.
Do not rewrite — refine, with the prompt buttons and “Create another”.
The result is a strong draft: the assistant gives you up to 90% in seconds, and you add the final touches by taste, by hand.

💡 This is exactly how professionals work with AI: not “do it all for me” but “make the foundation — I will finish it”.

Limitations

Generation happens on the server, so the AI assistant is a feature of the PRO add-on and it needs the internet while it works. But everything created stays yours for ever and works without the internet — these are ordinary scenes and chases.

In conclusion

The AI assistant is not a magic “make a masterpiece” button but a very fast and sensible helper that needs to be directed. Quality rests on three pillars: an accurate fixture profile, an accurate layout, a precisely set task. Give it good ground and a full brief and in seconds it will build what would have taken hours by hand. Describe things in detail, refine step by step, finish by taste.

Guide the neural network and it will repay you in quality, saving you a great deal of time.

With that, the basic SvetoMaster course can be considered complete: from the first fixture to network universes and working with the AI helper. Lessons on individual topics will follow — keep an eye on this section.

Good luck, and may the light always obey you!