How to Make an AI Commercial Film: PAM's Real Production Pipeline

Most 'AI commercial film' tutorials stop at prompt-in, video-out. Real production doesn't work that way. We break down the Midjourney → Kling → Sora → ElevenLabs → DaVinci pipeline we use on actual client projects — including the production decisions that generic tutorials skip.

• AI commercial film takes a pipeline, not a single tool — each stage uses a different model.
• PAM's pipeline: Midjourney v7 (storyboard + visual language) → Kling 2.5 / Runway Gen-4 (shots) → Sora 2 (complex scenes) → ElevenLabs (audio) → DaVinci Resolve (edit).
• What separates good AI film from bad isn't the tools — it's production decisions: light, composition, continuity, rhythm. Those come from set experience.
• A well-built pipeline delivers in hours, not days, at a fraction of traditional shoot costs.
• The most common mistake: expecting one model to do everything. Each model is a tool. You're still the director.

Most 'how to make an AI commercial film' guides look like this: write a prompt, get a video, done. Real production doesn't work that way. A commercial is storyboard, visual language, shot continuity, sound design, edit rhythm — dozens of decisions stacked on top of each other. AI doesn't make those decisions for you. It just lets you execute them faster. We didn't build this pipeline as an experiment. We've been on set for 20 years and developed this as a production discipline. Here's what each model is actually for — and the decisions that most tutorials skip.

Why AI commercial film needs a pipeline, not a single tool

On a traditional shoot, one camera doesn't do everything. Director, DP, sound, editor — separate disciplines. AI production works the same way. No single video model can deliver consistent visual language, character continuity, and professional sound at the same time. So we use the strongest model for each stage. Each model hands off what it can't handle to the next one. When people say a film 'doesn't look like AI,' the reason is usually not the tools — it's how well the transitions between them were managed.

Step 1 — Storyboard and visual language: Midjourney v7

Everything starts on paper before the camera rolls. Same with AI. Jumping straight to video generation is like showing up to set without a shot list. We start with Midjourney v7 to build the storyboard and establish visual language. Not to make pretty images — to lock in the film's tone, lighting, color palette, and compositional approach. Get one frame's lighting right here and every video step that follows aligns to it. This is where 20 years of set experience shows up concretely: 'cinematic lighting' in a prompt isn't enough. You need to write like a DP — 'low-key, single source side light, warm practicals, light haze.'

• Generate 3-6 key frames first: opening, product moment, close. These are the film's spine.
• Keep the same prompt skeleton across frames, changing only the scene — light and palette need to stay consistent.
• If you need character continuity, use a reference image and a stable character description. Faces that shift between frames break the film.
• Render high-resolution with clean composition — these go into the next step as image-to-video starting frames.

Step 2 — Shot generation: Kling 2.5 and Runway Gen-4

Once the Midjourney frames are locked, things start moving. We feed those frames into Kling 2.5 and Runway Gen-4 as image-to-video. Short and controlled shots are more convincing — same rule as real production. The most common mistake in AI video is expecting one prompt to generate a long, complex shot. We break the film into 3-6 second shots instead, the same way you'd shoot each setup separately and cut them together. Kling handles physically consistent, controlled movement better. Runway Gen-4 is stronger on dynamic camera movement. For critical shots we generate both and pick whichever cuts best.

• Keep each shot to 3-6 seconds. Longer shots multiply the risk of AI drift and morphing artifacts.
• Describe camera movement explicitly in the prompt: 'slow push-in', 'locked tripod', 'light handheld' — camera language shapes the feel of the ad.
• Generate 2-3 variations per shot and choose in the edit. Don't commit to the first output.
• For product shots, precise focus and correct scale matter — don't let AI 'improve' the product's appearance.

Step 3 — Complex scenes: Sora 2

Some shots are too complex for the image-to-video workflow: multiple moving elements, crowd scenes, physical interaction. That's where Sora 2 comes in. We don't use it for everything — only for the 'expensive' scenes that other pipeline steps struggle with. The scenes that would eat the most budget on a traditional shoot. One thing that matters: Sora's lighting and color palette have to match the visual language established in Midjourney. If they don't, you get a seam in the film. Coherence matters more than any individual shot's quality, so Sora outputs always go through a color and tone pass.

Step 4 — Sound design and voiceover: ElevenLabs

Half of a commercial is sound — and it's the most neglected part of AI production. No matter how good the visuals are, an artificial-sounding voiceover collapses the whole film. With ElevenLabs we build both the voiceover and a sound identity that fits the brand's tone. Which word gets emphasis, where the sentence pauses, how the voice rhythm tracks the visual rhythm — these are director-level decisions made by ear. We design the sound alongside the edit, not pasted on afterward.

• Match voiceover tone to brand identity — an energetic sports brand and a quiet luxury brand can't share the same voice.
• Don't skip music and ambience. Silence is a design decision too.
• Align audio transitions with cut points — the edit should breathe at the same moment the sound does.
• Always mix on both speakers and headphones. Most people will watch the ad on their phone.

Step 5 — Edit, color, and final: DaVinci Resolve

When all shots and audio are ready, everything comes together at the edit. In DaVinci Resolve we assemble the shots, build the rhythm, and most importantly establish color coherence. Shots from different models carry different color characters. Pulling them into a single visual language is the last defense against the film looking like it was assembled from parts. Edit is also where rhythm gets built — cut tempo is one of the biggest drivers of a commercial's impact. AI gave us the raw material in hours. Turning that material into a film is still a production decision.

3 mistakes people make with AI commercial film

• Expecting one model to do everything: no model handles storyboard, shots, sound, and edit on its own. The work is in managing the pipeline.
• Trying to generate long shots: three controlled 3-second shots are always more convincing than one 10-second AI shot.
• Treating audio as an afterthought: spending hours on visuals and leaving voiceover to default settings is the most expensive amateur mistake.

Should you do this yourself or work with a production team?

You can learn this pipeline and we'd encourage trying it. But what makes a brand commercial work isn't access to the tools — it's the production decisions behind them. Right light, right rhythm, voice that fits the brand, continuity across shots. Those come from years of accumulated set knowledge. We run this pipeline on real client projects every day, using AI not as a shortcut but as a layer that accelerates production quality. If you're thinking about an AI commercial for your brand, take a look at our [AI Video Production](/en/video) page or [get in touch](/en/contact).

Frequently asked questions

Commercial licensing is something we cover in a separate post — each tool has different terms and they keep changing, so a quick summary here would do more harm than good. Short answer: yes, commercial use is possible, but each tool needs to be checked individually.

On speed: with the right pipeline, a production that would take days compresses to hours. But the real difference isn't on shoot day — it's that location scouting, set build, and crew coordination disappear entirely.

There's no single 'best tool' answer. Midjourney leads on visual language. Kling and Runway on shots. Sora on complex scenes. ElevenLabs on audio. Which one is 'best' depends entirely on which stage of production you're in.

Blog · Get a quote