PTENES
Explainer videos · INEMA

Turn content into explanation

The process used for Astra and OSWork Quick, organized to produce new explainers with an avatar, illustrations, and captions.

Nei Maldaner, avatar from the HeyGen template used in explainer videos
What it is

A production you can resume

Scripting, generation, assembly, and delivery each have their own statuses and files.

▶ Complete content

Each topic points to its source and scenes. OSWork covers all 48 topics, examples, and exercises.

▶ Nei explains

The approved HeyGen template preserves the avatar and voice. Illustrations follow the measured speech timings.

▶ Verifiable delivery

Audio, duration, captions, and playback are checked before the MP4, player, and notice in bot v3.

How it works

From source to player

Scripts handle repetitive steps. Editorial and visual reviews happen before scaling up.

Source and script→ Avatar and illustrations→ Verification and delivery
Prerequisites

Production environment

This version uses the INEMA environment. For another machine, adjust the configuration and adapter paths.

Tools and access

Python, Node, FFmpeg, GitHub CLI, systemd, an authenticated HeyGen profile, and runtime-loaded lookup/transcription credentials.

# command
python3 --version
node --version
ffmpeg -version
gh auth status
User guide · step by step

Produce with one setup

The OSWork example is already prepared. If a run has started, check its status instead of generating another manifest.

1

Prepare and track

Copy the configuration to a new project and review the source, destination, and language. Each step preserves its results.

python3 explica.py --config examples/oswork.json status  # shows generation, downloads, render, and publishing
2

Prepare

Only before starting generation.

python3 explica.py --config examples/oswork.json prepare
3

Review

Audio-free preview to review the diagrams.

python3 explica.py preview
python3 engine/check_layouts.py
4

Run

Five persistent services continue running after you close the terminal.

python3 explica.py start
5

Follow along

States and logs show what actually happened.

python3 explica.py status
journalctl --user -u explica-oswork-completo-submit -n 20 --no-pager
V2 format · animated explainer

The visuals explain what Nei is saying

In v2, each scene has a visual script with spoken cues: when the narration says "input, model, output," the flow appears; when it says "there is no best model," the podium crumbles and the criteria radar identifies the right model. The avatar and audio are reused, with no new generation in HeyGen.

V2 scene: OSWork system with components connected to the core

21 animated primitives

hub, pipeline, workflow, radar, podium, trails, comparison, chat, terminal, folder tree, profile, lab, quiz, timeline, and more, in engine/v2/runtime.

V2 scene: models separated by function across tracks

Spoken cues, never seconds

The script uses verbatim speech segments; the build resolves each cue against the actual transcript and flags gaps over 10 s. Captions use the spelling from the script.

v1 preserved

The v1 engine (explica.py, engine/*.py) stays the same and remains available for productions and improvements. The previous version of OSWork is in the release video-v1.0.0.

# visual script: <output>/visual-v2/pt-bNN.json (see engine/v2/AUTHORING.md)
export EXPLICAVIDEOS_CONFIG=examples/oswork-v2.json
python3 engine/v2/setup_output.py          # reuses avatar, transcript, and timings from v1
python3 engine/v2/build_block.py 1 --strict # cues → timings; 0 warnings
engine/v2/run_lane.sh 1 2 3                 # check + render verified per block
python3 engine/assemble_languages.py && python3 engine/publish_finished.py
Formats · 16:9 and 9:16

Four formats, one avatar recording

The same speech from Nei becomes a horizontal or vertical video, in a lesson or viral profile. Pick by subject: lesson for courses and projects; viral for news, numbers, weather and reels. The format comes only from the v2 config.

16:9 viral format: full-screen image, big number and avatar in a box at the bottom right

16:9 viral

When to use: news, numbers, weather, YouTube videos that need to hook the viewer.

What it looks like: frame 0 is the cover, with a striking image and a big title already in the artwork. Then full-screen Codex images changing every 1.5–3 s, a giant number (bignum) and full-screen charts (ranking, barra, mapa). The avatar sits in a box with an amber frame at the bottom right and grows to full screen in rosto shots. Large captions of 1–3 words, with the spoken word in amber.

"profile_visual": "viral"

Rules and QA (qa_viral.py): engine/v2/PADRAO-VIRAL.md.

16:9 lesson format: dark background, cards on the left and a small avatar at the top right

16:9 lesson

When to use: courses, projects, step-by-step explanations.

What it looks like: dark background, usable area on the left (x 90–1420) with cards and diagrams built from the 21 primitives (flow, radar, terminal, comparison…). The avatar is small at the top right, with CENA n/N and the TAKEAWAY card below. Captions follow the speech.

# v2 default: no profile_visual
"engine": "v2"

Visual script: engine/v2/AUTHORING.md.

Simple 9:16 format: lesson scene on top and avatar in the bottom strip

Simple 9:16

When to use: vertical version of a lesson or an informative short (30–40 s).

What it looks like: scenes are still drawn at 1920×1080 and scaled to the top. The cropped 16:9 avatar fills the bottom strip. The right column (CENA n/N, TAKEAWAY) does not appear: use "takeaway": false. 46 px captions in 2 lines inside the safe zone.

"aspect": "9:16"
9:16 viral reel format: full-screen cover on frame 0, then image on top and avatar on the bottom half

9:16 reel

When to use: Reels, Shorts and TikTok.

What it looks like: frame 0 is the cover (hook shot at "@start", visible at t = 0). Captions of up to 3 words at 70 px, no header on top, rendered at 30 fps and −14 LUFS. With the viral profile, the image fills the top half and the avatar the bottom half; rosto takes the whole screen. Publishing requires APROVADO_REEL with the MP4 sha.

"aspect": "9:16",
"reel_profile": "divulgacao" | "tutorial" | "mini-aula",
"profile_visual": "viral"   # optional
Examples

The origin of the method

Everything Explicavideos has produced so far: 8 productions and over 90 assembled videos, with Nei's avatar and voice, chapters, and captions.

LOOP-R

5 tracks in PT, EN, and ES, ~17 min each, explanatory animation (v2).

Watch

OSWork v6.2

8 modules in PT, EN, and ES, ~16 min each, in v1 and v2.

Watch

Full OSWork

The whole course in one 66-min video, with chapters.

Watch

OSWork Quick

Quick version in PT, EN, and ES, ~30 min each.

Watch

Astra Básico

Complete videos in three languages, ~17 min each.

View Astra

IA Cultivada

10-min video in Portuguese, in v1 and v2.

Watch

Codex × Claude

6-min comparison in Portuguese, in v1 and v2.

Watch

Event shorts

9 vertical 9:16 shorts of 30–40 s: AGI Ready, JEV, RSI, WebMCP, and more.

Watch
Roadmap

Production in progress

Status on 2026-10-05. Each video is published as soon as it passes the checks.

Done
Complete LOOP-R15 videos published: 5 tracks in Portuguese, English, and Spanish, about 4h13 in total.
Now
OSWork v6.221 of 24 videos ready; modules 3 and 6 in Portuguese and 6 in English remain.
Done
Full OSWork, Quick, Astra, IA Cultivada, Codex × Claude, and shortsGenerated, synchronized, verified, and published, with MP4 and SRT captions.