Filip Ronaj

AI Creative Technologist

I make AI video for national TV and build the tools and agent systems that make it repeatable.

45+reviewed AI shot candidates for a national TV production
160K+followers on a TikTok account I built
ComfyUIspecialised in custom nodes and local generation
25+ hrssaved every week by the AI system I built (est.)

Selected work

My broadcast work for TV Markíza is confidential, so everything here was made new for this portfolio: original, rendered locally and documented from the first idea to the final frame.

Read the case studies ›

One Character, Twenty Worlds

One original character placed in twenty different scenes, with a face-similarity score for every frame. Most AI artists claim consistency. This piece measures it.

StatusPublished, clips next
PipelineComfyUI, from one reference image
Output20 stills and a drift report
ResultAll 20 scenes above the same-person line
ForBrands that need one face across a campaign
Drift reportMean 0.69 · lowest 0.45

Face similarity to the reference, scored with OpenCV SFace. 0.36 is its same-person line.

Thirty Seconds

A 30-second spot for a fictional product, made end to end: idea, script, storyboard, generated shots, edit, original music and sound design.

StatusIn development
RoleWriter, director, editor
ToolsComfyUI video models, YuE, DaVinci Resolve
ShowsDirection and taste, not only generation
Pencil storyboard frame: a hiker on a mountain ridge holding a lantern, with an arrow marking a camera push-in
Storyboard
Generated shot: the hiker seen from behind on the ridge at dusk, the lantern glowing in his hand
Generated shot
Final graded frame of the same shot, letterboxed
Final frame

ComfyUI nodes and apps

I build the nodes a pipeline needs, up to whole apps inside ComfyUI. Studio is one: create, edit, a gallery and a model manager in one window, with the workflows running underneath.

StatusStudio in testing on GPU
Built withPython, ComfyUI server API, JavaScript
Also builtShot Director, per-shot camera and lens controls
SolvesTeams run whole pipelines without opening the graph
Live demo · try it
Live demo of the real interface: press Generate, switch tabs or open the Gallery. Results are recorded, nothing renders here.

AIOS

The AI operating system I run my work and life on. Specialist routes do the work, research subagents gather evidence, and every correction becomes a rule the agents follow next time.

StatusRunning daily
Built onClaude Code, MCP, Python
Scale97 capability routes, 13,000+ indexed sources
Runs onMac and Windows, synced through Git
rules Routes97 specialist routes Researchparallel subagents Recordsrevision-checked state Lessonsfrom every correction Filip

Final Wish

An independent sci-fi short film I wrote, directed and shot.

StatusIMDb credit, 2022
RoleWriter, director, director of photography
GenreSci-fi short film
FestivalLift-Off Sessions 2022, Official Selection

About

I work where broadcast production meets agentic software. At TV Markíza I generate AI video and build the tooling that makes the pipeline repeatable. Outside work I design agent harnesses, including an AI operating system that runs my research, engineering and production and learns from every correction.

ExperienceTen years of making things people watch, the last three with AI at the centre.

  1. Apr 2026 – Present

    AI Video Specialist · TV Markíza (CME)

    AI video for broadcast productions at Slovakia’s largest commercial broadcaster, plus Cinema Studio, a custom ComfyUI node that turns the production workflow into one control surface.

  2. 2023 – Present

    Independent AI Systems Builder

    AIOS, my AI operating system, and automation shipped with tests: a sales-bot fix with 2,292 tests passing and an onboarding recovery flow merged and deployed to production.

  3. 2021 – Present

    Founder · Mediagy

    Social media growth agency: content strategy, audience targeting and analytics across multiple accounts, with AI automation in daily operations.

  4. 2016 – Present

    Content Creator & Filmmaker

    Wrote, directed and shot “Final Wish” (2022). Grew a TikTok account to 160,000+ followers and an Instagram account to 30,000.

Toolkit

Generative mediaComfyUI custom nodes, MiniMax, Seedance, Higgsfield, HeyGen, LoRA training, character consistency
Agents and LLMsClaude Code, Claude API, OpenAI API, MCP, multi-agent orchestration, RAG, evals
EngineeringPython, JavaScript, HTML/CSS, Git, pytest, REST APIs, Linux
ProductionPremiere Pro, DaVinci Resolve, Final Cut Pro, directing, camera
EducationFilm Direction, Private Secondary School of Animation Arts, Bratislava
LanguagesSlovak (native), English (professional), Czech (fluent)

One face. Every frame. Scored.

I build ComfyUI pipelines that keep an AI character on-model across outfits, poses, lighting and scenes, for creators, agencies and brands. Every delivery includes a drift report: twelve fixed test prompts, each with a pass, drifting or rebuild verdict, so weak spots are fixed before final delivery.

  1. Character testA drift report on your character concept and 10 on-model images.2 days
  2. Character pipelineA ComfyUI workflow tuned to your character, a trained LoRA when it needs one, 40 images and the drift report.4 days
  3. Content engineThe pipeline plus batch generation from a spreadsheet, image-to-video and setup on your GPU or cloud.7 days

Price by agreement. Pipelines come with the workflow files, the settings and a guide, with full usage rights for your project.

Let’s make something.

Open to creative technologist roles, and to AI video projects that need one consistent character, a directed sequence or a pipeline your team can run.

One portfolio, every place the work lives.

The services, AI video, the ComfyUI work, AIOS, the film on IMDb, the code on GitHub and my LinkedIn, each one click from here.