Free workflow

One-photo product ad workflow

The exact local pipeline behind Test 001: cut a product out of one photo, put it on AI-generated sets, relight it, and animate it into commercial shots.

Scripts: MIT. Models: their own licenses (Apache-2.0 at time of testing).

A still from the test this workflow came from

What's inside

  • ComfyUI API graphs: set generation, masked relight, Wan 2.2 image-to-video
  • composite.py (cut-out, grade, shadow, reflection, change map) + finish_still.py
  • Example configs from the HALVEN test
  • README with setup steps, settings and known limitations

You need

  • NVIDIA GPU with ~24 GB+ VRAM (we used an RTX 5090)
  • ComfyUI + Wan 2.2 I2V A14B + Z-Image-Turbo (free, Apache-2.0)
  • Python 3 with Pillow + NumPy

Where it breaks

  • Camera orbits around the product break the label (see Test 001)
  • The cut-out uses measured geometry: you trace your product's outline once

Get the improved version

One email when we update this workflow.