Stop Posting AI Slop (Build Your Own Style Instead)

Summary

The video presents a workflow for creating distinctive, on-brand AI imagery instead of publishing generic images that may weaken a business’s social presence. Its central argument is that brand context and human curation matter more than increasingly elaborate prompts: images should reflect the audience’s pains and desires while conveying an intentional emotional style.

Three-step workflow

1. Provide brand context. Use a brandbrain.md file with information about the brand, target audience, offers, and audience psychographics. The creator says the workflow can also work without the file, with the skill asking for any missing information.
2. Generate audience-relevant scenes. Install the Brand Scene skill in Claude and run it with the brand context. It proposes scenes based on audience pain points and desires, along with Midjourney style-reference instructions.
3. Explore, test, and lock a visual style.
   • Use Midjourney’s sref random with repeat 4 to explore styles for the same scenes.
   • Select promising results using human judgment and taste, then test the style across multiple prompts and image formats.
   • Tune style weight and stylization. You can also combine reference codes or use a reference image.
   • Once the style works across different scenes, record the settings in the brand documentation and generate a larger image bank.

Practical guidance and examples

The creator contrasts generic AI imagery—such as images used by local restaurants and estate agents—with a curated style intended to feel distinctive and emotionally resonant.

Test the chosen style on varied scenes before producing a large batch. The goal is consistency across multiple uses, not just one attractive image. Keep results that fit the intended brand world and discard those that do not. If a prompt clashes with the style reference and produces inconsistent results, move on.

Midjourney’s curly-brace syntax can combine prompt variations, generating multiple scene variants from a single prompt instead of requiring each prompt to be entered separately.

The resulting image bank can be used across social posts, websites, posters, cards, carousels, and ads. The creator describes using an image bank in an automated process that turns articles into carousels, with branded designs going out weekly.

Metrics and settings mentioned

• Midjourney is described as having over one billion possible style-reference combinations.
• The initial random-style exploration uses four generations per run (repeat 4).
• The creator tests a style weight of 400, then 1,000; style weight is described as ranging from 0 to 1,000.
• The demonstrated stylization setting is 1,000.
• One generated bank is described as containing 310 images; the creator later downloads 140 images from that day’s work.
• The creator says the workflow can support a bank of 100–200 or more image variants.

These figures are examples from the demonstration, not performance or engagement results. The video does not report changes in reach, engagement, or sales.

Sources and tools

Presenter: Luke Carter Tools and materials: Claude, the Brand Scene skill, Midjourney, and a brandbrain.md file.