AI content production with editorial and directorial control

Where do we use AI?

In five stages of production, for precisely defined tasks. Development: treatment variants, scene lists, moodboard copy. Pre-production: concept sketches, storyboards, preliminary animatics and virtual location scouting with image and video models. Shooting: preparing LED wall backgrounds and virtual set elements. Post-production: rotoscoping and masking, object removal, generative extension, upscaling, dialogue clean-up, subtitling, dubbing and translation. Delivery: format variants (16:9, 9:16, 1:1) and localised versions.

Every AI step is paired with a human step: the director or the editor selects from the generated material, the colourist and the sound engineer finish it, and the project manager checks the labelling and the legal conditions.

Where do we not use AI?

In three areas, as a matter of principle: the client's face and voice, health claims, and legally sensitive imagery.

The client's face and voice. We do not generate or synthetically alter the likeness or voice of a real person, except with written consent and a documented purpose. Under Section 2:48 of the Hungarian Civil Code, creating and using a person's likeness or voice recording requires their consent, and no AI tool overrides that.

Health claims. For healthcare clients we do not create generated before-and-after images, body depictions or any visual claim that real footage does not support.

Legally sensitive imagery. We do not generate recognisable protected characters, brand logos, copies of an artist's style or real events in a form that looks authentic. In 2026 these are the subject of live lawsuits, and in advertising the risk always falls on the client too.

Careful: reference, control, post-production

Professional use rests on three pillars. Reference: consistency comes from references, not from text (Ingredients, Elements, ControlNet, IP-Adapter, a LoRA trained on the brand), with the product, the talent and the colour world locked shot by shot. Control: fixing the seed, step count and guidance, selecting from multiple generations (Kalshi's 2025 NBA commercial yielded 15 usable clips from between 300 and 400 generations), and tools built on top of each other: the second season of Amazon's "House of David" produced between 350 and 400 AI shots by chaining 10 to 15 tools. Post-production: colour grading, editing, sound mixing, upscaling, and finally provenance data and labelling. The market does the same: in August 2026 Runway released a dedicated HDR colour grading model, and Adobe's and Blackmagic's 2026 developments focused on colour and dialogue.

Netflix's partner guidelines (approval for key story elements and the likeness of real people, no training on production material, a closed enterprise environment), the SAG-AFTRA agreement ratified on 4 June 2026 (synthetic performers only where they add "significant value"), the 99th Academy Awards rules (May 2026: only roles played by humans with consent and scripts written by humans) and the mandatory AI declaration at Cannes Lions (in 2026, 40 percent of the 20,050 entries declared AI use) describe the same discipline at industry level.

What does a trained filmmaker add?

The model does not know the grammar of film. Composition, lens choice, lighting continuity, blocking, editing rhythm, colour science, sound design and story: the system does not make these decisions for us, and if they are not in the prompt, we get them at random. A trained filmmaker uses the generator the way a cinematographer uses a camera: they know which focal length, which lighting direction and which camera movement serves the scene, and they enforce it with references, with controls, and then at the editing desk.

The viewer can see the difference too: according to Kantar's effectiveness tests, more than 40 percent of ads using seamlessly integrated generative AI land in the top band for branded attention, while visibly AI-made visuals produce weaker branding and stronger negative emotion. That is why in our studio no AI scene goes out without an editor, a colourist and a sound engineer, and why real filming remains the backbone of brand content: the authenticity of the person, the product and the location is a form of trust capital that generated material cannot replace, only complement.

Why does it complement content production rather than replace it?

In an online environment, AI delivers three things: volume (format and message variants for A/B testing, 16:9, 9:16 and 1:1 cuts), speed (pre-visualisation that reduces the risk of the shoot, clean-up and upscaling in minutes instead of hours) and localisation (dubbing, subtitles and language versions at low additional cost). In the WFA's April 2026 study, 78 percent of global brands use AI-generated or AI-enhanced content, 87 percent of them for product images and 77 percent for background visuals; in Adobe's April 2026 survey, creative professionals use AI in more than 40 percent of their projects, while 84 percent of them worry about brand consistency.

What cannot be replaced: the trust of a real person, a real product and a real location (in the 2026 data from Canva and Harris, 74 percent of consumers are more likely to buy based on an ad made by people), legal certainty (likeness, voice, protected characters), platform algorithms that reward originality, and the brand's own unmistakable visual world. That is why at Oloren AI is built around the filmed footage: beforehand as a draft, afterwards as extensions and variants, never instead of the shoot. This model fits both the content production packages and project-based engagements.