Generative image and video work, and the tools behind it: production software for ministudio.ai, AI-driven commercials and Sakura Leaf, a narrative image series made between Midjourney, Stable Diffusion and Photoshop.
Full walkthrough of an internal desktop application built for ministudio.ai in January-February 2026. The tool is designed as the first module of a complete AI animation suite, covering consistent character, object and location generation, asset library management, automatic image classification, and auto-naming.
Built in Python, runs on Windows and Mac.
This is a fictional ad I created using Flux, ComfyUI, and AI-driven image-to-video tools. This project explores the evolving possibilities of AI-generated storytelling.
A Dynacord ad-teaser I have contributed to. Made for Laterna Magica.
“Sakura Leaf” tells a story about a Yakuza clan set in the 60’s and 70’s Japan. Images in the series are created with the help of A.I. tools such as midjourney and stablediffusion and further refined and scrupulously composited in photoshop. It’s a narrative series inspired by books and movies that explores the future of storytelling in light of new technological developments. Each image was constructed with a particular theme in mind that covers different aspects related to the life of Yakuza and their influence.
I worked on this project from December 2022 to December 2023. A lot has happened in the world of generative AI during this period: general quality improvements, ControlNet, custom stable diffusion models… Midjourney advanced from version 3 to version 6, representing a three-generation leap in just one year. Therefore, the final quality and realism of different images may vary as they were created at different times throughout the year.
The general workflow for the project was Midjourney -> Photoshop -> Stable Diffusion -> Photoshop. Initially, I had a specific concept in mind for each image, aiming to portray an aspect of the Yakuza. I utilized Midjourney to establish the general composition and atmosphere of the intended image. In certain instances, I used a source image along with prompts to guide the generation of the desired image. At the outset, the images generated by Midjourney v3 lacked realism, requiring adjustments in Photoshop alongside Stable Diffusion to enhance realism and remove artefacts.
Overall, this project served as a valuable learning experience, introducing me to the world of generative AI. Images below show the before and after of the initial image generated by Midjourney and the final composite.
