Addis Pulse Studio — Silver Spring, Maryland

A studio that buildsits own tools.

Films, presenters, brand identity and print — for small businesses that need to look like bigger ones. Produced on systems I write, run and publish as open source. Nothing is outsourced, and nothing is a wrapper around somebody else’s API.

Start a project
See what I make

The work

Every frame was made here.

No stock, no licensed footage, no subcontractors. Generated, directed, graded and finished on the studio's own hardware — which is the only reason a small business can afford it.

Two ways in

Hire the studio

You need the work made.

Restaurants, clinics, trades, shops, creators — anyone whose materials are holding them back.

  • Identity
  • Presence
  • Film

Own the engine

You need the system.

Agencies, legal and medical practices, and teams whose files cannot go to a cloud.

  • The Engine

Four outcomes

Named for what you end up with.

01

Identity

Be taken seriously in the first three seconds.

Your work is good and your materials say otherwise. A logo someone made in Canva, a menu in three different fonts, a website that looks like it closed in 2019. Customers price you on what they can see.

What you get

  • Logo and full brand system
  • Typography, colour, and the rules that keep them consistent
  • Packaging, menus, signage, business cards — print-ready
  • A website that loads fast and reads well on a phone

Why it is different

I design for paper and screen in the same pass, because the thing that breaks a small brand is the gap between them — a beautiful site and a photocopied flyer. Everything ships in the languages your customers actually speak.

Fix how I look →

One conversation, then a first direction in about a week. You react, I refine, you own the files outright.

Brand identity carried onto paper — where most small brands fall apart
02

Presence

Show up every week without it becoming your job.

You know you should be posting. You are also running the business. So the account goes quiet for six weeks, then you post three times in a day and it reads like panic.

What you get

  • A month of social content produced in one sprint
  • Long-form cut into clips people actually finish
  • Search and listings kept current
  • A calendar you approve once, then forget about

Why it is different

Produced on my own hardware, which is why a month of content costs what a week does elsewhere. No per-minute API meter deciding how much you get.

Keep me visible →

A single production sprint each month. You approve a plan; the work arrives scheduled and ready.

Podcast clips — hours of conversation cut into moments people watch
03

Film

Say it in motion, in your customer’s language.

Video is the one thing that makes a small business look like a big one, and it is the first thing that gets cut — because a crew, a studio day and an edit is five figures before anyone has spoken.

What you get

  • Brand films and launch promos
  • AI presenters who deliver your script to camera
  • Motion graphics, titles, and product animation
  • The same film in Amharic, Spanish, French or Arabic

Why it is different

No crew, no studio day, no reshoot fee. A script change is a re-render, not a second shoot — which is why the fourth language costs almost nothing and why most studios will not offer it.

Put us on camera →

You send a script or a rough idea. First cut in days. Revisions are cheap because nothing was filmed.

AI spokesperson — any script, any language, no reshoots
04

The Engine

Run the AI yourself, on hardware you own.

Your files are confidential, your volume is high, or both — and every cloud tool wants you to upload the thing you cannot upload, then bill you per minute for the privilege.

What you get

  • OMNI installed on your hardware or private cloud
  • ComfyUI pipelines built for your actual work
  • A plain internal web app, so nobody has to touch a node graph
  • Automation for the repetitive parts — intake, replies, paperwork

Why it is different

I am not reselling an API, and I am not handing you someone else’s node graph. The pipelines are built on tooling I wrote and published under Apache-2.0 — you can read it before you buy it, and keep running it if I disappear.

Scope an install →

A scoping call, a fixed install, then training your team on the interface — not on ComfyUI.

Output from the engine — generated, graded and finished in-studio

Not an add-on

Everything ships in your customer’s language.

The same film, the same menu, the same campaign — delivered again in another language with a presenter who sounds like they grew up speaking it. Because nothing was filmed, the fourth language costs a fraction of the first. Most studios cannot quote this at all.

  • English

    Hello, welcome in.

  • አማርኛ

    ሰላም፣ እንኳን ደህና መጡ።

  • Español

    Hola, bienvenido.

  • Français

    Bonjour, bienvenue.

  • العربية

    مرحباً بكم.

Proof

It is running right now.

Not a diagram of a system I could build. A console you can open, with workflows you can run on the GPU sitting in my studio. Scroll to clear the node graph away — that is the before and after of an install.

  • Runs on hardware you own, or your private cloud
  • Open-source node packs — Apache-2.0, readable before you buy
  • One model resident at a time, sized to the machine
  • A plain internal web app, not a node graph
  • Zero egress — nothing leaves your premises
Open the console →

Open source

The tools are public.

Two Apache-2.0 node packs for ComfyUI, written for work that had to ship. You can read every line before you hire me, run them without me, and keep running them if I vanish. That is a different kind of claim from a case study.

01

ComfyUI-HyperFlow

Apache-2.0Python

Runs Video Rebirth’s 8-step HyperFlow adapter for MiniMax-H3 on ComfyUI’s native H3 implementation, with ComfyUI’s own samplers.

Why it exists

Upstream ships only a diffusers loader. Dropped into ComfyUI as an ordinary LoRA it silently loads nothing — the key names do not match, and HyperFlow is not only a LoRA: it conditions each step on the interval it integrates.

Native ComfyUI integration

Converts diffusers/PEFT key names onto all 314 native targets, or stops with an error

8-step HyperFlow

Eight model calls in place of forty-nine

t/r endpoint conditioning

Patches the time embedder so every row gets the right endpoint per step

T2V · FL2V · Ref2VA

Text, first/last-frame, and reference-driven video with audio

Audio-guided AV

External audio drives synchronized video and generated sound

Live GPU verification

All four workflows run and documented with frame checks and timings

In the audio-guided and Ref2VA runs, the generated soundtrack follows the reference voice’s loudness envelope at correlation 0.97–0.98, 0 ms lag, no drift.

Addis-Pulse-Studio/ComfyUI-HyperFlow
02

Pulse Studio

Apache-2.0Python

A production toolkit for MiniMax H3: direct a render from a shot timeline and an asset bin, on one node, without ever typing a reference tag number.

Why it exists

Reference tags are positional, so adding a character three shots in renumbers everything after it and the film quietly renders the wrong faces. The failure is silent, which is the worst kind.

Pulse Slate

The shot timeline — the window where the film is written

Named asset references

You write @Mimi; the ordinal is computed from live bin order at compile time

Pulse Shot

Per-shot control inside the timeline

Pulse Render

Long-form rendering across many windows, with a segment cache

Retake

Patch a bad three seconds without re-rendering fifteen

Still Mode

Make reference images in the same window that uses them

Pulse Voice · Lip-Sync

One narration across every window, with a report when the recording does not cover the film

No dependencies beyond what a working ComfyUI already has. Five example graphs, each opening with no third-party pack installed.

Addis-Pulse-Studio/comfyui-pulse-studio

Why trust a studio you have not heard of

I publish what I learn.

Seventy-two breakdowns of open-weight model releases — what the licence actually permits, where the benchmark is being gamed, what it costs to run on one GPU. Written before the work, not as marketing after it.

You can read it and decide for yourself whether I know what I am talking about. That is a harder claim to fake than a testimonial.

Daily RadarOpen-source model drops, tracked daily — what shipped, what licence it carries, what it runs on.Read the research →

How it goes

Three steps. No account manager.

  1. 01

    One conversation

    Thirty minutes. I learn the business, you learn what it costs. No deck.

  2. 02

    I make the thing

    Design, film, build. You see work, not status updates.

  3. 03

    You own it

    Files, source, accounts — handed over outright. I stay on only if you want me to.

Behailu Weldeyohannes

Who you will be working with

Behailu Weldeyohannes

Founder and engineer — Silver Spring, Maryland

Software engineering and global health law, which between them taught me that most systems fail at the handover — when something built by specialists reaches people who were never meant to operate it.

So I build the infrastructure, run it myself, and hand over work you own outright. You talk to me, not to an account manager, and the person who answers the phone is the person who made the thing.

Contact

Start with one conversation.

Three questions, thirty seconds. It comes to me directly — there is no one else here to route it to.

What’s your business called?
What do you need most right now?
Best email to reach you?

Straight to my inbox. I reply within one business day, usually sooner.

Prefer email? contact@addispulsestudio.com

Addis Pulse Studio

Private AI production — Silver Spring, Maryland

contact@addispulsestudio.com

Services

IdentityPresenceFilmThe Engine

© 2026 Addis Pulse Studio. All rights reserved.

TermsPrivacy