London ad director completes commercial with Wan3.0, skips color grading

Wan3.0 is the video generation model Alibaba released in public beta on August 6. As covered in Alibaba launches Wan3.0 public beta with native 30-second video generation, the model's defining feature is its ability to take a variety of inputs — text, images, documents — and produce up to 30 seconds of video in a single generation pass. On August 14, Alibaba followed up with an "Agent Skills" feature that lets the community share workflows built for the model.

Put simply, what Tasker was testing was whether Wan3.0's native 30-second generation, launched in public beta on August 6, could actually hold up to commercial production standards. AI video tools have mostly been used so far to string together short clips, and inconsistent lighting and color between scenes has been a persistent problem.

A chicken riding a unicycle, "Let's give chicken a break"

The ad Tasker made is for Plant Beast, a plant-based nugget brand. The concept shows a chicken that no longer has to live in a cage, riding a unicycle around freely, built around the line "Let's give chicken a break." He said he drew the idea from the growing public interest in factory farming, spurred in part by documentaries like Netflix's "Big Chicken," but that he didn't want to make an ad that lectures viewers about giving up meat.

Filling 30 seconds from a single base image

Here's how Tasker describes the workflow. Instead of feeding the model separate character and prop references, he first used an image generation model to create one "base image" that already contained the chicken, the unicycle, the lighting, and the color palette all in a single frame. Feeding that image into Wan3.0 along with a prompt let him generate shots from multiple angles while preserving that same lighting and color. A typical 30-second ad runs about 17 shots; for this project, he says he wrote prompts for 10 shots, including variations.

Item	Industry-average 30-second ad	Tasker's project
Shot count	About 17	10, including variations
Length per shot	2-3 seconds	2-3 seconds
Post-production color grading	Done separately	Skipped
Share of time on planning/prompting	—	More than half of total time

"Prompt fidelity is the single most important factor for a director," he said. Because lighting and color held steady across shots instead of drifting, he says he also needed fewer regenerations than expected to land on a result he was happy with.

What changed from the previous version

Tasker had also worked with the earlier Wan2.7, and he says the biggest improvements in Wan3.0 are cinematic realism and prompt fidelity. He added that Wan3.0 is also cheaper in credit costs than other video generation models, which makes it more viable for projects without a big-budget client behind them. His studio is currently using Wan for a pharmaceutical 3D animation project as well, one that visualizes the inside of cells using multiple reference materials.

Advice for creators just getting started

Tasker's advice to creators looking to make ads or brand videos with Wan3.0 is to nail down the idea, story, and prompts before ever opening the model. More than half of the actual working time, he says, goes into planning before the model is even turned on. He also recommends building a single base image that captures the whole mood, rather than feeding in multiple separate character references, and prompting several shots in a batch at once rather than working through them one at a time in sequence.

Editor's take

What makes this case interesting isn't just that color grading got skipped — it's that the skip wasn't a lucky accident, it came out of how the workflow was designed. Instead of pushing multiple character sheets into the model, Tasker locked in a single image with lighting and color already baked in and used it as the anchor point. Given how long AI video tools have struggled with lighting drifting from scene to scene, this approach secures consistency through upfront preparation rather than relying on the tool's own capabilities.

The generational comparison makes the shift concrete. Through the Wan2.7 era, getting an exact desired composition out of a single prompt wasn't easy, so directors typically shot first and matched tone afterward through grading. Tasker's experience with Wan3.0 is that higher prompt fidelity made that after-the-fact correction step unnecessary altogether. Given how much of an ad's budget typically goes into post-production, this isn't just a quality improvement — it means an entire stage of the production pipeline disappears.

There are two practical takeaways for advertising and video production teams here. First, locking in a genuinely usable base image with an image model — not just a moodboard — up front is what determines the consistency of the final output. Second, as Tasker put it, more than half of total time needs to go into planning and prompt writing. Bolting on an AI video tool doesn't make copywriting and storyboarding skills less important — if anything, those skills need to get sharper for the output to actually hold up.

Expect to see more real-world cases like this surface over the coming weeks.