In today’s world, data is everything. When we upload images, videos, or text to commercial AI apps, we rarely know what happens behind the scenes. Even legal protection for sensitive personal information is limited.
​
This workshop took a different approach. We worked with open-source tools like ComfyUI, which ran locally on consumer-grade computers. Where we controlled our creative process, our results, and our data. The goal was not just to generate content, but to understand how to negotiate with AI, using workflows that gave us agency, transparency, and freedom. 
​
The workshop was designed for beginners with no prior knowledge of ComfyUI. The aim was to produce precise results of an existing site image or a project image in the form of images, videos, and 3D models. The workshop began with an introduction to the fundamentals, checkpoint models, latent space, VAEs, and prompt structures, so that participants understood how ComfyUI operates beneath the interface.
​
From there, the workflow unfolded in three stages -
-  First, the image stage, where participants used ControlNet modules such as Depth and Canny to guide AI towards design iterations that responded to time, season, and material conditions. Flux tools were used for style transfer, inpainting, and upscaling to refine the outputs into coherent and detailed visuals.
​
- Second, the video stage, where still images were converted into moving sequences with WAN 2.2. Here, participants learnt prompt structures, explored model variations, and worked with settings like frame rate and camera motion to generate videos that added more details to their designs.
​
- Finally, the 3D stage, where images were prepared with Gemini 2.5 Flash Image (Nano Banana) for 3D interpretation, followed by conversion through Hunyuan 3D or Sparc 3D. The models were then analyzed, refined, and cleaned in Rhino3D (or a preferred method) to create fabrication-ready geometry.
​
By the end, participants had worked through the full cycle: text to image, image to image, image to video, and image to 3D model, in turn converting their design into a human - AI negotiated output with the help of AI’s latent capabilities.