The Season 2 title sequence for Menna Wa Fena went to air on Almashhad in 2024. Gold Arabic calligraphy on black marble. Every frame, every texture, every light reflection generated with AI. No Blender. No traditional compositing chain. I want to be precise about that, because “AI-assisted” usually means somebody generated a background plate and called it a pipeline.
It worked. It also broke in specific, predictable places, and those places are more useful to talk about than the parts that went well.
Why I tried it at all
The brief was material-led: marble, gold, depth, weight. Nothing about it needed precise geometry or a camera move I had to control to the frame. In a traditional pipeline that shot list is a lot of render time for a look that a path tracer is merely adequate at — subsurface on stone, caustics on metal, soft specular falloff. Those are exactly the things generative models are unreasonably good at.
So the question wasn't “can AI make a nice frame.” Everyone knows it can. The question was whether it could hold one consistent look across a complete broadcast title sequence, on a transmission deadline, with a client who expects the same thing they approved.
What it was genuinely better at
Material and light
Marble, gold, caustics, the way light rolls off a polished edge. Getting this out of a renderer is hours of lookdev and then hours of render, and the result is often clean in a way that reads as CG. The generated frames had the incidental imperfection that sells a real surface — and I didn't have to build it.
Atmosphere
Dust, haze, depth falloff, the soft bloom around a highlight. In a normal comp that's four or five layers I'm hand-balancing. Here it arrived inside the plate, already integrated, already correct relative to the light.
Iteration speed
This is the one that actually changes how you work. A directional note — warmer, heavier, more negative space — became a re-prompt instead of a re-render. When a change costs minutes instead of a night, you try things you would otherwise talk yourself out of. That's a real creative difference, not a productivity one.
Where it broke
Arabic letterforms — by far the biggest problem
Generative models mangle Arabic script. Connections between letters break. Diacritics appear that shouldn't exist. A model will produce something that has the texture of Arabic calligraphy and is, as language, nonsense. If you don't read Arabic you will not catch it, and if you do, you'll catch it immediately — and so will every viewer.
This is not a prompt problem. You cannot phrase your way out of it. The type has to be controlled deliberately and brought into the sequence as a known, correct element rather than trusted to the model. Anyone telling you generative video is ready for Arabic typography has not shipped Arabic typography.
Continuity
One marble surface and one gold treatment across a whole sequence is harder than one beautiful frame by an order of magnitude. It takes reference discipline and a large pile of rejected output. Budget for the rejects — they are the actual cost of the method, and nobody puts them in the case study.
Timing
Models produce clips. They don't produce edits. The rhythm of a title sequence — how long you hold, when you cut, what lands on the show name — is still a human decision made in the cut, and it is the decision that separates a sequence from a moodboard. Every weak AI title sequence I've seen online fails here, not on image quality.
Brand geometry
Related to the type problem but worth separating: a logo has to be exactly right. Generative output is approximately right. In branding, approximately right is wrong.
The hybrid is usually the correct answer
The counter-example from around the same period is a cinematic Arabic title open I built with Blender 3D for the metallic typography and structure, and Runway ML supplying the smoke and embers around it. Structure and type where precision matters. Generative where texture and atmosphere matter.
That split is the working method now. Choosing correctly per shot is the actual skill — not prompting, not tool knowledge. Knowing which shots you can hand over and which you absolutely cannot.
The honest verdict
Full generative production is viable today for sequences that are texture-led, atmosphere-led and non-typographic — and only if someone in the chain has traditional motion design instincts about pacing and restraint. It is not viable when precise brand geometry or Arabic type has to be exactly right.
The part people get wrong is assuming the tools removed the need for craft. They moved where the craft sits. It used to sit in execution. Now it sits in judgment: what to generate, what to build, what to throw away, and when to hold the shot two beats longer than feels comfortable.
See the sequence and full project credits →
More on how I use AI video in a broadcast pipeline →
Questions people ask
Can AI make a broadcast title sequence?
Yes — the Season 2 title sequence for Menna Wa Fena on Almashhad was produced end-to-end with generative AI, with no traditional 3D or compositing. It works best for texture-led and atmosphere-led sequences. It is not reliable where precise brand geometry or Arabic typography has to be exactly correct.
Why does AI struggle with Arabic text?
Arabic is a connected script. Generative models break the joins between letters and invent diacritics, producing output that has the texture of Arabic calligraphy but is not readable language. It cannot be fixed by prompting; the type has to be controlled deliberately and brought in as a known-correct element.
Is AI replacing motion designers?
It moves where the craft sits rather than removing it. Generation handles material, light and atmosphere well. Pacing, edit rhythm, typographic accuracy and knowing which shots can be handed to a model are still human decisions, and they are what separate a finished sequence from a moodboard.
Let’s make something move.
Open to full-time motion design and 3D roles in Dubai, the wider Gulf and remote. Reels, breakdowns and CV available on request.
Email me CV & availability Watch the showreel