Campaign OS
Services
Generative AI Design
Art Direction
Location
London
Year
2026
Credits
ComfyUI
Qwen Image Edit 2509
Info
Built a custom local GenAI workflow using the Qwen Image custom node focused on fast, controllable campaign generation for fashion, retail, sportswear, and marketing applications. Using a consistent model image plus outfit, product, pose, or background references, the system can quickly generate scalable campaign visuals while maintaining character consistency across outputs. The workflow supports outfit changes, product placement, environment swaps, pose adaptation, and editable campaign typography, creating a flexible pipeline for editorial campaigns, social assets, virtual try-ons, and creative marketing production without requiring a large traditional shoot pipeline.

CONTEXT
Exploring:
How consistent I could keep a model across multiple outputs
How quickly campaign visuals could be generated from minimal inputs
How modular the workflow could become for future fashion/retail systems
Simple workflow:
One consistent model image + one outfit/product reference + optional pose/object/background references.
Runs fully locally and free once set up. I specifically chose to run the workflow locally rather than relying on external APIs or hosted tools because I wanted more control over the generation pipeline, faster iteration, lower long-term costs, and the ability to experiment modularly with different nodes, models, and conditioning systems. Running locally also makes it easier to prototype scalable production workflows without API limits, usage costs, or dependency on closed platforms.
Blueprint for a controllable visual production pipeline that can rapidly generate scalable campaign assets without rebuilding everything from scratch each time.
Disclaimer:
All source imagery used here is pulled from publicly available internet references purely for workflow demonstration and research purposes. I do not own the original brand imagery, products, or photography shown. These examples are non-commercial and intended only to demonstrate the functionality of the workflow.




EXPLORATIONS
Had the most fun putting random things together that made no sense


