Sora and other AI video generators fail in marketing for predictable reasons: teams treat them as a replacement for a brief, ignore brand and rights rules, and publish clips without human review. The tools themselves, including OpenAI’s Sora, Google’s Veo, Runway and Kling, can produce striking footage. The mistakes happen in planning, editing and governance, and each one has a practical fix.
What can AI video tools do well right now?
At the time of writing, text-to-video and image-to-video models are good at short atmospheric shots, stylised scenes, product mood footage, b-roll and concept visuals for pitches. They let a team test five creative directions before paying for a shoot.
They are weaker at precise continuity across many shots, legible on-screen text, accurate hands and faces in close-up, exact product geometry and consistent characters across a campaign. Clip length and resolution limits vary by tool and plan, so check the current documentation before promising a deliverable.
That split shapes every decision below. Agencies that get good results plan around the strengths and keep real footage for the parts where accuracy carries legal or commercial weight. Teams that get poor results ask the model to do everything, then spend more time fixing artefacts than a conventional shoot would have taken. Knowing the limits early is the cheapest quality control there is.
Mistake 1: Starting without a creative brief
Typing “make a cool ad for our sneakers” produces generic footage that could belong to any brand. The model fills gaps with averages from its training data.
The fix is the same brief you would give a director: audience, single message, shot list, camera movement, lighting, colour palette and length. Write prompts per shot, not per video. “Low-angle tracking shot, white running shoe on wet asphalt at dusk, orange street light reflections, slow motion” gives the model something specific to render.
Can one generation give you a finished ad?
AI clips are raw material. Brands that post the first output usually get odd motion, mismatched shots and no sound design.
- Generate several variations per shot and pick the strongest.
- Edit in a normal editor such as Premiere Pro, DaVinci Resolve or CapCut.
- Add licensed music, voiceover and real typography in post rather than asking the model to render text.
- Colour-grade all clips together so they look like one piece.
- Insert real product shots where accuracy matters, such as packaging or labels.
Why do product details go wrong?
Generators approximate objects. A logo may warp, a bottle cap may change shape between frames, and a phone may gain an extra camera lens. For regulated or technical products, that is more than cosmetic: it can misrepresent what the customer will receive.
Use AI for environments and motion around the product, then composite real product photography or 3D renders on top. Image-to-video workflows that start from an approved product still also reduce drift.
Mistake 3: Ignoring rights, likeness and disclosure
This is the mistake with legal consequences. Prompting a model with a celebrity’s name, a competitor’s character or a famous film style can create content you cannot safely publish. Real people’s likenesses need consent, and many tools block such prompts anyway.
- Read each tool’s terms on commercial use and output ownership before client work.
- Never generate identifiable real people without written permission.
- Follow platform rules: YouTube, TikTok and Meta ask creators to label realistic synthetic media in certain cases.
- Keep provenance data where available; Sora and several Adobe and Google tools attach C2PA Content Credentials or watermarks.
- Log prompts, tool versions and dates for every published asset.
How do you choose the right tool for each shot?
Different generators suit different tasks. Some are stronger at realistic physics, some at stylised animation, some at image-to-video from a still, and some integrate directly into editing suites. Picking one tool for everything usually means compromises.
Run a short bake-off: the same three prompts in two or three tools, judged by the creative director on brand fit, motion quality and how much editing each clip needs. Repeat it every few months because the field changes quickly.
A quick pre-publish checklist
- Does every shot match the brief and brand guidelines?
- Are product details, logos and any text accurate?
- Are there visual artefacts in hands, faces or backgrounds?
- Is the clip labelled where the platform requires it?
- Is the audio licensed and the voiceover approved?
Mistake 5: Measuring novelty instead of results
A clip that gets compliments in the office is not automatically a better ad. AI footage can look impressive and still lose attention after the first second because the message arrives too late.
Judge AI-assisted videos by the same numbers as any other creative: three-second view rate, average watch time, click-through rate and cost per conversion. Test them against your current best performer with equal budget, and keep the ones that win on those numbers rather than on how new they feel.
How Alien Road applies AI video generation
In a typical engagement we begin with discovery: the brand’s guidelines, product constraints and the platforms where the video will run. We decide early which shots are suitable for generation and which need real footage, so the client knows the mix before production starts.
Setup means a shot-by-shot prompt sheet, tool selection based on a small test, and an editing pass with real product assets, music and captions. Every asset goes through human review against the checklist above and is logged with prompts and tool details. Measurement compares the AI-assisted variants with existing creatives on the same placements, using hook rate, hold rate and cost per result, and reporting shows which shots and styles earned their place.
Where to start this week
Pick one upcoming campaign and identify three b-roll shots you would normally license from a stock library. Write a detailed prompt for each, generate them in two tools, and edit the best versions into an existing ad. Compare cost and turnaround with stock, and you will have a grounded view of where AI video fits in your production line.
Keep reading: Video Production · AI Marketing