AI Creative Production Pipeline

Can AI generate production-ready campaign prompts while maintaining a brand's creative identity and product accuracy?

Traditional AI image generation often produces inconsistent results. Prompts become repetitive, products lose visual accuracy and maintaining a consistent brand aesthetic across an entire campaign requires significant manual effort. I wanted to explore whether a structured workflow could automate prompt generation while preserving both creative direction and garment fidelity.

Rather than writing prompts manually for every image, I designed a modular node-based workflow in Figma Weave. The system combines product photography, visual references and brand direction before passing them through multiple LLM stages to generate structured prompts for Higgsfield. By using the garment image as a visual reference rather than describing it in text, the workflow maintains product accuracy while allowing the prompts to focus on scene composition, lighting and atmosphere.

Reference Images - Visual Guide Generator - Brand Style Guide - Prompt Generator - 10 Campaign Prompts - Higgsfield - Campaign Images

1. 01

   Understanding AI Image Generation - Research how prompting affects consistency, product accuracy and brand identity across image generation models.

2. 02

   Designing the Workflow - Map a modular pipeline that separates brand analysis, visual style generation and prompt creation into reusable stages.

3. 03

   Building the Pipeline - Develop a node-based workflow in Figma Weave that combines product imagery, reference photography and creative direction into structured prompts.

4. 04

   Testing Outputs - Experiment with different reference images, prompt structures and visual guide formats to improve consistency across generated campaigns.

5. 05

   Refinement - Iterate on prompt structure and workflow logic to maximise garment fidelity while giving AI enough freedom to generate varied campaign imagery.

• —

  Maintaining Product Accuracy - Most AI image generators distort products when they are described entirely through text. I redesigned the workflow to use the uploaded garment image as the primary visual reference, allowing prompts to focus on creative direction rather than re-describing the product.

• —

  Brand Consistency - Creating individual prompts manually often resulted in inconsistent visual styles. Introducing a dedicated brand style guide stage ensured every generated prompt inherited the same creative direction.

• —

  Prompt Architecture - Finding the right balance between structured prompts and creative flexibility required extensive experimentation. The workflow evolved through multiple iterations to produce prompts that remained consistent while avoiding repetitive outputs.

• —

  Workflow Scalability - Rather than creating prompts one at a time, the workflow was designed to generate multiple production-ready prompts from a single creative brief, making the system suitable for larger content production.

I developed a reusable AI production pipeline capable of transforming product photography, brand direction and creative references into structured, production-ready prompts for campaign imagery. Beyond improving workflow efficiency, the project deepened my understanding of prompt architecture, creative systems design and how AI can support—not replace—creative direction within production workflows.

Building this workflow completely changed how I think about AI image generation. I realised that consistent results come less from writing better prompts and more from designing better systems. Separating brand understanding, product analysis and prompt generation into distinct stages produced significantly more reliable outputs than relying on a single prompt alone.

• →

  Generate both image and video prompts from single creative brief

• →

  Expand workflow to support multiple industries besides retail alone

• →

  Support multiple brand style libraries for different clients