Title: GitHub - DreamLab-AI/THG-world-record-attempt

On February 25, 2026, DreamLab AI demonstrated a practical application of autonomous agents in creative work: the IRIS system generated a complete Topshop fashion campaign (138 assets) in 96 minutes from voice prompts alone. A human creative director simply spoke 21 natural language instructions—no interface work, no manual workflow setup. The system autonomously decomposed tasks, selected appropriate AI tools, generated assets, and handled quality checks.

The speed advantage is striking. Traditional experts need 4 hours to construct workflows and generate 36 images at 15 minutes each. IRIS produced 138 assets (including videos, composites, and variations) in 96 minutes—approximately 9.4x faster. The system eliminated the 4-hour pipeline construction phase entirely, requiring only ~5 minutes of human voice direction across the entire project.

The technical approach is noteworthy for how it divides labor: local GPU-based generation for base images (faster, more controllable) combined with cloud APIs for refinement and specialization. Multiple agent swarms worked in parallel—one for scene variations, another for garment reskinning, others for quality assurance and documentation. All ComfyUI workflows were generated by agents, not humans.

Limitations exist: garment fidelity required one human correction, AI text rendering failed entirely (solved with programmatic overlays), and all assets use digital mannequins rather than human models. QA covered only 36 of 133 assets formally. Despite these constraints, the project demonstrates that agentic systems can handle the orchestration, decision-making, and quality management of complex creative pipelines—shifting human roles from execution to high-level artistic direction.