Hook, desire, fear, ease, discovery, benefit and comment — each step lands on the right word of the speech, with its own image, headline and panel. It ends with a "COMMENT <WORD>".
reel7etapas is a template for writing short videos (Reels, Shorts, TikTok) and a tool that assembles those videos on its own. The script always follows the same 7-step order, designed to hold someone who is scrolling and to end with a simple ask: comment one word. It is for creators who want a repeatable path instead of starting from scratch with every video. To just use the template, reading the 7-step table is enough; to generate the video, the tool uses Nei's avatar in HeyGen, images generated by Codex and a Python script.
Show the result before explaining the technology. People grasp the benefit in 2 seconds and stay to learn how to get it.
Seven steps, always in the same sequence. You change the topic, the structure stays.
Talking avatar, one image per step with camera movement, headline and highlight that land on the exact word, and a short caption, all inside the 9:16 safe zone.
The mapa command generates a sheet with the 7 steps side by side: the video frame, the trigger and the why.
Desire → fear of wasting money → ease → discovery → benefit → comment, opened by a visual hook. The examples come from the reel "Your own JARVIS" (Seu próprio JARVIS).
The result anyone wants, in 2 seconds, in one huge sentence. "VOCÊ DORME. ELE TRABALHA." (YOU SLEEP. HE WORKS.) and the spoken line "Você sabia que o seu computador já pode trabalhar enquanto você dorme?" (Did you know your computer can already work while you sleep?)
The person pictures themselves with what they want. "Já pensou em ter um funcionário dentro do seu computador?" (Ever thought about having an employee inside your computer?)
The risk of losing money holds people more than the promise of gain. "Tem gente pagando uma fortuna por coisas que você já pode montar em casa." (Some people are paying a fortune for things you can already build at home.)
Knock down the objection that makes people give up before trying. "E o melhor: não precisa ser programador!" (And the best part: you don't need to be a programmer!)
Name the solution and say what it does, in simple verbs. "Você pode criar o seu próprio JARVIS… conversa, lembra, pesquisa e executa." (You can create your own JARVIS… it talks, remembers, researches and executes.)
Close the picture of how life looks afterwards. "Um funcionário digital só seu, trabalhando do jeito que você ensinar." (A digital employee that's all yours, working the way you teach it.)
One small, easy action: comment a word. "No inema.club tem vídeos e material… Comenta JARVIS que eu te mostro o caminho." (On inema.club there are videos and material… Comment JARVIS and I'll show you the way.)
Dark background, golden light, the computer front and center, yellow and white lettering. No futuristic robots, no sci-fi look.
Jargon (AI, agent, model) before step 5, two theses in the same video, and a long CTA. One word to comment is enough.
To use only the template, nothing: it is the script. To generate the video with the tool:
The engine reuses reelcomenta's transcription, face crop, HeyGen studio and Telegram delivery, and makeshorts' qa_short.py quality gate.
git clone https://github.com/inematds/reelcomenta ~/projetos/reelcomentaThe images come from Codex's image generator (image_gen), through the subscription. If it fails, the local FLUX.2 klein server steps in as a fallback.
codex --versionPillow, OpenCV, PyYAML and ffmpeg. The avatar comes out of the HeyGen studio through the subscription, with a recorded approval.
pip install pillow opencv-python pyyamlEverything runs through reel7etapas.py. Each edition lives in edicoes/<date>-<topic>/. No step asks for approval: the engine has gates that fail and explain what went wrong.
Copies the JARVIS example script into a new folder. Rewrite the speech (550–700 characters, about 40 s), ending with "Tá tudo no inema ponto club. Ou comenta <PALAVRA> que eu te mando o link." (It's all on inema dot club. Or comment <WORD> and I'll send you the link.)
python3 reel7etapas.py novo meu-assunto # edicoes/2026-10-11-meu-assunto/roteiro.yaml
Each step has the ancora (anchor: the word in the speech that makes the step come in), the headline, an optional destaque (highlight: a card next to the avatar), the camera and the image cena (scene). gatilho (trigger) and por_que (why) only show up on the map. The cta closes with the word to comment.
- id: medo
ancora: gente
linha1: "TEM GENTE PAGANDO"
linha2: "UMA FORTUNA"
destaque: {titulo: "CARO", texto: "por algo que\nvocê monta\nem casa"}
camera: pan_dir
cena: "Mesa escura com notas de real, faturas e uma calculadora..."
Generates the 7 images in Codex, records the avatar if it does not exist yet, assembles, validates and makes the map. An image that is too dark, the wrong orientation or too similar to another is redone automatically.
python3 reel7etapas.py fazer edicoes/<pasta>
Nothing outside the Reels/Shorts safe zone (top 10%, bottom 20%, button column), captions of at most 3 words and a font of 64 px or more, an image change at least every 3.5 s, speech identical to the script, −14 LUFS and makeshorts' qa_short.py passed.
python3 reel7etapas.py validar edicoes/<pasta> # + build/revisao/folha-*.jpg at full size
Each stage also runs on its own. imagens accepts the step numbers to redo only those.
python3 reel7etapas.py imagens edicoes/<pasta> 3 7 # redoes images 3 and 7 python3 reel7etapas.py avatar edicoes/<pasta> # requires avatar/APROVADO_HEYGEN python3 reel7etapas.py montar edicoes/<pasta> python3 reel7etapas.py mapa edicoes/<pasta> # mapa.jpg + capa.jpg
Sends the video through the Telegram bot and publishes it as a YouTube Short through yt-pubx (the script's channel; default lives1). It only runs if validar passed after the last assembly.
python3 reel7etapas.py publicar edicoes/<pasta>
The first reel made with the template. Full script at edicoes/2026-10-11-jarvis/roteiro.yaml. ▶ Watch on YouTube
What exists today and what's coming.
