A daily AI ad creative workflow starts with a current brand kit, useful reference ads, and clear requirements for each output. The agent uses that context to develop original concepts, generates and reviews each image, and saves the accepted work where the team can find it. A repeatable process makes creative production easier to inspect and improve.
In his September 25 walkthrough, Antonio Romero shows how Growthub approaches this process using Weekly Care as the demonstration. The lesson is the preparation behind the images: brand assets, messaging, examples, persona context, and explicit quality requirements all shape the result. Generating an attractive asset is one stage; campaign performance still depends on what happens when the team tests it.
TL;DR
- Start with the current brand kit and examples that explain what good creative looks like.
- Give each concept a purpose, audience, offer, and explicit copy requirements.
- Generate one image, inspect it, and correct material defects before moving on.
- Build meaningful variety across concepts while keeping brand details consistent.
- Save reviewed assets with enough context for the next person to use them.
- Use campaign evidence to guide subsequent iterations; production volume alone does not establish performance.
Watch the workflow behind the creative
Start at 0:34 for the brand-kit setup, 1:44 for disclaimer and persona requirements, and 2:37 for asset retrieval, generation, and review. The final section at 3:04 shows the emphasis on distinct creative variations.
1. Make the brand kit useful to the actual task
A logo and a color palette are necessary inputs, but they leave many creative decisions unresolved. The workflow also needs to know which audience it is addressing, which offer is current, how the brand speaks, and what a reviewer considers acceptable. Without that context, an image can look polished while missing the assignment.
In the walkthrough, the brand kit brings together tagged ad references, messaging, examples of quality, and explicit output requirements. That gives the agent something more concrete than a request to make a good ad. The operator can point to a specific example and explain why its composition, hierarchy, or approach matters.
For your own setup, assemble the approved logo, product imagery, visual guidance, current offer, audience description, and any exact text the asset must contain. Add a short explanation of what each reference contributes. A reference with a reason attached is easier to use than a large folder of unexplained screenshots.
2. Study references, then develop original concepts
Reference research helps a team see how other ads solve a communication problem. One example might make the offer immediately legible. Another might show an effective product close-up or a clear relationship between the headline and the image. Those observations can inform a new concept without carrying over another brand’s identity or claims.
The demonstration draws on competitor references and, for some workflows, existing account winners. Treat those as different evidence types. An ad visible in a public library is a creative reference; its presence does not prove profitable performance. An account asset described as a winner should be tied to the team’s actual measurement before it drives a performance conclusion.
A useful concept brief identifies the audience, communication goal, visual approach, offer, required wording, and reference principle. For example, a product-detail concept and a customer-situation concept can support the same offer while testing different ways to explain it. That is more informative variety than changing the background color on the same composition.
3. Turn requirements into a reviewable generation brief
At the prompt stage, separate creative choices from requirements that must remain exact. The composition can vary. Approved offer language, required disclaimers, product identity, and the intended audience need deliberate handling. Putting everything into one vague paragraph makes it harder to tell whether the output actually satisfies the task.
The video describes dynamic prompts that incorporate persona context and specific disclaimers. The practical takeaway is to supply those requirements before generation and inspect them afterward. A generator can omit a word, alter small text, or place essential copy where it is difficult to read. The presence of a disclaimer in the prompt does not establish that the final image contains it correctly.
Before running a batch, write down the expected number of distinct concepts, the format, the destination, and the conditions for accepting an image. Keep the brief specific enough that another reviewer could reach a similar decision. This turns quality from a general preference into something the team can discuss.
4. Generate one image and inspect the result
At 2:37, Romero describes retrieving the brand assets, generating images one by one, checking the result, and then continuing. He also shows that an operator can steer the work with further instructions. This sequence matters because a defect found early can improve the remaining concepts before the same mistake spreads through the batch.
Inspect the image at the size where it will be used. Check the product, logo, layout, spelling, required copy, and visual relationship to the brief. Then compare it with the other concepts. An individually attractive asset may still add little value if it repeats an idea the batch already contains.
- Brand fidelity: Does the image use the correct identity, product representation, visual style, and audience context?
- Exact copy: Is every required word present, accurate, legible, and consistent with the approved offer?
- Distinct ideas: Does the concept contribute a different visual or communication approach that the team can explain?
5. Save the reviewed work with a clear handoff
The demonstrated workflow saves outputs to Google Drive. The broader principle is to make the result available in the destination the team actually uses. A generated image still needs a stable home, a useful name, and enough context for someone to understand its intended use.
For a practical handoff, keep the asset alongside its concept, audience, offer, and review status. If a file needs a correction, make that visible. If a version has been accepted, preserve it so another run does not overwrite the work the team selected. These small details reduce the effort required to move from production to review and testing.
The video closes with a scheduled workflow example. Scheduling can make an established process recur, but each run still needs usable inputs, connected tools, and a saved result. Start by making one run dependable. Then decide what should repeat, when it should run, and which conditions require someone’s attention.
6. Let testing guide the next creative decision
Daily production is useful when it feeds a learning process. Record what each concept was designed to test before it enters a campaign. After the team has enough relevant evidence, connect the result back to the creative decision: audience framing, offer clarity, image approach, or another variable the concept deliberately changed.
The walkthrough mentions iterating from existing account winners. Apply that idea with the account’s actual results and operating context. Preserve the element you have a reason to keep, change a specific part, and document the hypothesis behind the next version. Avoid labeling a newly generated asset high-converting before campaign evidence supports that description.
Expert Q&A
What should be ready before the first batch?
Prepare the current brand kit, approved assets, offer, audience, useful references, exact copy requirements, and saving destination. Define what makes an image acceptable and what should differ across the concepts.
Why generate and review one image at a time?
Early review catches issues while they are still cheap to correct. It also gives the operator a chance to improve the next brief, rather than discovering at the end that every image repeats the same mistake.
Can competitor ads serve as proof that a concept will perform?
No. They can inform creative research, but visibility in an ad library does not establish profitability. Performance claims require relevant campaign evidence. Use references for ideas and your own testing to evaluate results.
Does saving a generated asset mean it is ready to launch?
No. Saving confirms that a file exists in the chosen destination. The team still needs to complete its creative review, check the offer and required wording, and decide whether to launch the asset through the appropriate campaign process.
Build the method behind your next batch
Explore the Growthub skills directory for reusable workflow resources. Start with one relevant skill, read its instructions and supporting references, supply your brand context, and use it on a bounded task. The purpose is to make a method reusable while keeping each brand’s inputs and requirements explicit.
For walkthroughs, resources, prompts, and community feedback, explore Growthub Agent OS Academy. If you want to map a recurring creative workflow around your company’s tools and knowledge, explore Growthub OS and book a mapping session.
Source: Antonio Romero’s September 25, 2026 video, linked above. The workflow is described from that demonstration; the implementation advice expands its operating principles. The article does not assert campaign results for the generated examples.