The biggest problem with AI-generated advertising is not quality. It is consistency. Faces change between shots, logos drift, and a product looks slightly different every time it appears. A recent tutorial, sponsored by OpenArt, shows a workflow aimed squarely at that problem: build every element of the commercial as a locked reference asset first, then generate the video from those assets.
As someone with an accounting and auditing background, I recognize the logic immediately. It is a controls approach. You standardize the inputs so the output is reliable.
Step 1: Character Sheets for Every Person
Each person in the ad gets a character sheet: one wide 16:9 image with three panels (full front, full back, and a tight face shot) on a plain gray background with a neutral expression. The creator used GPT Image 2.5 at 4K, attached a real photo of himself for the barista so only the face carried over, and built the other characters from text. Clothing was chosen on purpose: a green apron for the barista, a bright mustard jacket so one customer is easy to spot in a wide shot, and a rust cardigan close to, but not matching, the brand color.
Step 2: Build the Logo Once
This is the step the tutorial calls out as the one that ruins many AI ads. If you write the brand name into separate prompts, you get logos that do not match. The fix is to generate the logo a single time, as a three-panel sheet: the full lockup, the icon alone, and a reversed version. Generate a batch of four, pick the best, and attach that exact image wherever the brand appears.
Step 3: Products That Carry the Logo
With the logo sheet attached as an image reference, the creator generated two products: a burnt-orange paper coffee bag and a clear iced-coffee cup. Each got three panels, including a close-up of the spot where the camera will actually look, such as the bag label and the cup lid with the straw. That is where generated products usually go wrong, so it is where the reference matters most.
Step 4: The Set and the Rig
The shop interior was generated in a batch of four. The lighting prompt was specific: warm pendant lights creating visible pools of light, blue-gray walls in the gaps, and a blown-out window at the back. It also named the thing to avoid, generic even lighting. The creator noted one flaw he would watch for next time: a wall lit with no lamp pointing at it.
The brewing rig was built as its own asset because it has a job in the shot: standing in the center of the counter and blocking the view behind it. The lesson is that size must be built into the asset itself (about 60 cm tall, with real weight), not requested later at the video stage. Branding was left off it entirely, since any extra text is one more thing the model may change.
Step 5: One Generation, Written Like a Shooting Script
All nine assets were attached to a single video generation. The creator chose a model that produces a full 30 seconds with sound in one pass, because stitching many short clips is where consistency usually breaks. The prompt was organized into eight fixed blocks: scene context, asset list, geometry lock, optics, shot-by-shot timeline, performance, lighting, and sound.
The geometry lock is the most useful idea. Instead of describing positions in every shot, state the hard rules once: the cup never moves once it is on the counter, its label always faces the camera, the rig stays dead center, the queue order never changes, and the headphones stay around the barista’s neck. The settings were 16:9, 1080p, 30 seconds, audio on.
Why Reference Assets Beat Prompting Harder
Most beginners respond to inconsistent results by writing longer prompts. This tutorial takes the opposite view: the prompt should carry rules, and the images should carry appearance. A face, a logo, or a paper texture is far easier to show than to describe. Once an image is attached as a reference, the model has far less room to improvise, and every later generation inherits the same details.
There is also a practical benefit. When something goes wrong, you know where to look. If the logo is off, the problem is in the logo sheet. If a character changes, the character sheet needs work. Fixing one asset is much cheaper than regenerating a full 30-second video and hoping the next attempt lands better.
What Worked and What Did Not
By the creator’s own account, the result held together well. The close-up of the finished coffee looked clean, and the customers in the queue matched their reference sheets. One shot, where the coffee bag appears to pop up from the shelf on its own, looked unnatural. That is a useful reminder that even a well-controlled workflow produces flaws that need to be reviewed before delivery.
One more practical point: the tutorial shows each asset being reviewed before moving on, and that habit matters. The creator checks the paper texture, the condensation on the cup, and the lighting on the walls, and he names the flaws he sees. Treat every asset as a draft that needs approval. If you skip the review at this stage, any error is baked into all nine assets and every shot that uses them, and you will not notice until the final video is finished.
Cautions Before You Sell This as a Service
First, the video is sponsored by OpenArt, and the closing pitch is to sign up through the creator’s link. That does not make the workflow wrong, but it is relevant when weighing the praise.
Second, the tutorial offers no client results, conversion data, or earnings figures. Producing an ad and selling advertising services are separate challenges, and a good-looking commercial does not guarantee sales.
Third, the tutorial’s own example uses a fictional brand. For a real client, the actual logo and packaging must be supplied by the client and checked carefully against every generated frame. Rights to any photos, music, and trademarks also matter.
Takeaway
The most transferable lesson is not a particular tool. It is the discipline: lock the characters, the logo, the products, the set, and the props before generating anything in motion, then state your rules once at the top of the prompt. For photographers, retouchers, and digital artists considering AI advertising as a service, that process thinking is the real skill. Creative judgment decides what is worth building. The tools only speed up making it.
For a small business owner or independent creator, the practical version is simple. Before you open any video tool, make a short checklist of everything that must look the same in every shot: people, logo, product, location, and props. Build each one as its own reference image, review it closely, and only then combine them. It takes longer up front, but it is the same reason a good bookkeeper sets up the chart of accounts before recording a single transaction. Structure first, volume second.
Sources:
Transcript of a video tutorial on building a 30-second AI commercial, provided by the author. The video identifies OpenArt as its sponsor. Tools discussed in the video: OpenArt, GPT Image 2.5, and a 30-second video generation model.
Note: The transcript does not independently verify claims about advertising effectiveness, platform performance, or commercial earnings. All observations on what worked and what did not are the video creator’s own.
About the Author
Orlando Monteagudo combines analytical thinking with mixed media experimentation, Photoshop workflows, AI-assisted creativity, and practical digital refinement systems designed to help artists create more cohesive, polished, and sustainable creative work.
Disclaimer
This post is based on a sponsored video tutorial and reflects the creator’s own demonstration and opinions. I have not independently tested the tools or workflow described, and results may vary. The post makes no claims about advertising performance, sales results, or earnings. Any commercial use of AI-generated images or video, including logos, products, music, and likenesses, requires that you have the proper rights and permissions. Tool names, features, and pricing may change, so check each platform before relying on this information. This post is for general informational purposes and is not professional or legal advice.
Keywords:
AI commercial, AI advertising, OpenArt, GPT Image 2.5, AI video generation, character sheets, consistent AI characters, AI logo consistency, reference images, 30-second commercial, AI product ads, AI workflow, brand consistency, AI video prompt, geometry lock