Services
AI & Software
Solutions
Training
Tools
Insights
References
Contact
Free initial consultation Ask a first question by email
About Deutsch
← AI & Software

Lab

Tools beside the advisory work: image and video pipeline, content automation, AI web audit and my own website as an interface. With an honest status per tile.

in operation

Image pipeline on its own GPU

Restoration, upscaling, colourisation and face restoration run locally, without a cloud. Before-and-after images only from my own archive.

in operation

Video & voice

Explainer and product videos from my own pipeline, with a labelled voice and subtitles.

experiment

Portrait and motion experiments

Image-to-video and portrait animation with my own portraits. Labelling under Art. 50 of the AI Act.

in operation

Content pipeline

Watch sources, score them, write a script, publish. Runs for my own channels, not as a marketing service.

in operation

Website as an interface: pfisterer.digital

A website that works like a phone: lock screen, apps, lab. Own code, security audit, data protection documentation.

in operation

AI web audit with agents

Several specialised agents review a website in parallel: findability in AI search, security, usability, content.

These are the tools I build beside the advisory work. None of them is an advisory service: they are my own tools, and the status on each tile says what currently applies. I do not write “in use” when something is only built.

If you are looking for an image or video pipeline for your own business, the route leads through training videos or interfaces and modules. What counts as a service is listed under AI Automation.

Image pipeline on its own GPU

Restore old photos, remove damaged areas, recover faces, add colour, increase resolution. This runs locally on the GPU, without a cloud service. The models are named: Qwen3-VL for recognition, LaMa for removal, CodeFormer and GFPGAN for faces, Real-ESRGAN for resolution, DDColor for colour. A whitepaper describes the process and its limits.

Result images carry an “AI-edited” note, and I show before-and-after pairs only from my own archive or with permission.

Video & voice

The video pipeline builds explainer videos reproducibly: script, shot script, capture, voice, subtitles, mix, render, 15 modules with their own validation command. The voice is cloned, but only with written consent, and labelled in the video.

One product video has come out of it: myPlace Parking, with a labelled voice. It sits as a file on this server and is on the training videos page.

Portrait and motion experiments

Image-to-video and portrait animation are a test field, not an offer. I work only with my own portraits and label the results under Art. 50 of the AI Act. What emerges here stays in the lab until it has a real use case.

Content pipeline

Watch sources, score posts, write scripts, publish videos and posts. The pipeline runs for my own channels, with a local model and self-hosted automation. It is not a marketing offer; I could not back that claim.

Website as an interface: pfisterer.digital

A second website that works like a phone: lock screen, apps, lab, every AI image labelled individually. Own code, security audit, data protection documentation with retention periods. The site is publicly reachable.

On this page there is a phone frame with a still image. Live embedding is not enabled yet, because the Content-Security-Policy is missing on both sides. So it stays with the still image and the link “Open in a new tab”; before that, no request goes to the other server.

AI web audit with agents

Several specialised agents review a website in parallel: findability in AI search, security, usability, content. I ran the first pass on my own website. Client audits only anonymously or with approval, and without promises of a position in a search result.

What is not here

The lab is not the place for services. Process mapping, assistants, automation and interfaces are under AI Automation. 3D spare parts are a service of their own, with their own process and page, not a lab tile; they start soon.

The next step costs nothing.

30 to 60 minutes, no strings attached: where is the biggest friction, what is the most effective lever, and is a collaboration a fit?