The project
A client needed a 30-second commercial for a car brand. Budget was mid-tier — enough for proper post-production, not enough for a full live shoot. The brief called for cinematic exterior shots: wet asphalt, dramatic lighting, motion blur, the usual.
My pitch: generate the base footage with AI, do all the post in After Effects, deliver something that looks like it cost three times the budget.
I used Kling 2.0 as the primary generator. Here's the honest breakdown.
What Kling 2.0 does well
Cinematic motion. The camera movement quality in Kling 2.0 is genuinely impressive for a static-to-video generator. Slow push-ins, tracking shots, parallax depth — when it works, it looks like real camera operation.
Lighting consistency. Within a single generation, lighting holds up well. I got consistent golden-hour looks across multiple clips from the same prompt family.
Wet surface rendering. Reflections on wet asphalt are surprisingly convincing. This was a big win for the car project — I didn't have to fake rain in post nearly as much as I expected.
Where it falls apart
Hands and interiors. Any time the camera moved toward a car interior or showed hands on a wheel, the generations became unusable. Kling still struggles with complex geometry that requires physical accuracy.
Consistency across clips. This is the core production problem with current AI video: each generation is its own universe. The car in clip 3 doesn't look exactly like the car in clip 1, even with identical prompts. I spent significant time in After Effects colour-grading each clip toward a common look, and doing subtle warp/transform work to align shapes.
Prompt sensitivity. Small changes in prompts produce wildly different results. "Cinematic car driving on wet road at sunset" and "cinematic car driving on wet road, golden hour" look like they came from different directors. You need to iterate fast and accept waste.
My actual workflow
- Generate 30–40 clips from 5–6 base prompts, varying seed and small prompt tweaks
- Cull aggressively — I kept about 25% of generations
- Build the edit first in Premiere, cut to music before touching colour
- Into After Effects for colour unification (Lumetri base, then custom expressions to link grades across clips)
- Motion graphics layer — titles, supers, the brand package — this is where AE shines
The AI generates raw material. You're still the director, editor, and colourist.
Would I use it again?
Yes, for the right brief. Kling 2.0 works when the deliverable allows for stylization and the brief doesn't require shot-to-shot character consistency. Car exteriors, landscape/environment shots, abstract brand moments — strong use cases.
It doesn't replace live shooting. It adds a new tier of production that didn't exist 18 months ago.
The production value I delivered on that project was above what the budget should have been able to buy. That's the real value proposition.