Skip to content

AI Creation: First Steps: Second Brain

2026-10-16 · Academy lesson · Adil · recording, 12:13 · original

Topic: Prompting & AI assistants (also: Higgsfield platform & ecosystem, Video generation & shot direction) · mode: Tutorial · tags: series, one creator's experience

In this «Second Brain» lesson of the AI Creation: First Steps course, Adil shows how to brief Supercomputer in four parts — I am, I need, do, format — and to keep one job per chat. For a scene he brings only three sentences, a bartender closing his bar in space at 3 a.m., and has Supercomputer interview him before it writes the video prompt, since a good question shows a choice he has not made yet. He treats the first 15-second 480p draft as a rehearsal, then changes one note at a time in the same chat: idea, prompt, result, tweak.

What is said

The parts of the video and what is said in each (★ — how much it matters to the video).

00:00:00 — Intro: the second brain and one line almost nobody types

00:00:33 — Opening Supercomputer; what a large language model is

  • ★★ Adil calls Supercomputer the most important tool of the course, a "second brain": an employee you can bring almost any digital task to. — 00:00:33 · Supercomputer

    “Today, we get to the most important tool in this course, your second brain.”

  • ★★ Supercomputer is opened by clicking it in the top menu of the Higgsfield homepage. — 00:00:33 · Supercomputer

  • ★★ Adil explains that an LLM (large language model) learns patterns from huge amounts of text in training, so it can explain a subject, write a script, plan a project or write code from a job described in ordinary language. — 00:00:57
  • ★ Adil says assistants such as ChatGPT, Claude and Gemini use language models with different strengths: some are faster, others built for more complex work. — 00:00:57 · ChatGPT, Claude, Gemini

00:01:36 — The Auto model choice and the four-part brief: I am, I need, do, format

  • ★★ Adil says Supercomputer gives the language model a desk and tools, so it can look things up online and work with video generation. — 00:01:36 · Supercomputer
  • ★★ In Supercomputer, the Auto button under the message box chooses which language model you talk to; Auto is selected by default and Adil leaves it there. — 00:01:36 · Supercomputer
  • ★★ Adil compares Supercomputer to the smartest intern on earth on day one: it may know your name but not what you need done by Friday (23 October 2026) or how it can help, so give it a brief. — 00:01:36 · Supercomputer

  • ★★★ Adil says to give Supercomputer the context of a task in four parts, "I am, I need, do, format" (who you are, what you need, what it should do, the answer's form), with notes, spreadsheets, images or a reference video as needed; the labels need not be typed. — 00:02:04 · Supercomputer

    “Four parts of the context, I am, I need, do, format.”

00:02:29 — AI news for a YouTube channel: a bare request against a full brief; one job per chat

  • ★★ Asked only "find me AI News", Supercomputer returned a roundup of the week's AI news with sources, but Adil notes it had to choose the time frame, audience and purpose itself. — 00:02:29 · Supercomputer
  • ★★ Adil's habit for Supercomputer: keep each job in its own chat, so a shopping list does not end up in the chat of your space movie. — 00:02:29 · Supercomputer
  • ★★ Adil's full brief for an AI news list says he makes YouTube videos about generative AI and asks for the most important AI news of the last few days, ranked most trending first, with one line per item on why his audience would care. — 00:03:10 · Supercomputer, YouTube
  • ★★ With the full brief Supercomputer ranked the news with a reason for each story; from that list Adil can choose a story, ask for three script versions and generate clips for the edit in the same chat. — 00:03:39 · Supercomputer
  • ★★ Adil says a coding-capable LLM given research and references can help write the code for an interactive 3D scene, such as an engine that assembles itself and shows its moving parts, though the result still needs testing. — 00:03:39
  • ★★ Adil advises reading Supercomputer's answer against the job given: check it has the details you need in a usable form, check any facts you will rely on, and add missing details in the same chat. — 00:04:30 · Supercomputer
  • ★★ Adil says the idea for his space-bar scene came to him while he was working as a bartender. — 00:04:30

00:04:55 — A three-sentence scene, the interview trick, Ask Before Generating and attaching a photo

  • ★★★ For a scene, Adil brings Supercomputer only three sentences saying who, where and what happens (a bartender closing his bar in space at 3 a.m.), with nothing about lenses or camera angles. — 00:04:55 · Supercomputer

  • ★★★ Adil's most useful trick: before Supercomputer writes the video prompt, make it interview you, since a good question shows a choice you have not made yet (warm wood or cold metal, lonely or frightened). — 00:05:25 · Supercomputer

    “Before it writes the prompt, make it interview you.”

  • ★★ To approve each request before anything is generated, open Ask Mode under the Supercomputer message box and check Ask Before Generating. — 00:05:47 · Supercomputer

  • ★★ To attach a photo in Supercomputer, click the plus beside the message box, then Attach File, and in Assets click Upload Media and choose the photo. — 00:05:47 · Supercomputer
  • ★★ If Supercomputer starts drafting the prompt straight away instead of interviewing you, Adil says to send another message asking for the questions first. — 00:05:47 · Supercomputer

00:06:14 — Answering the interview, checking the summary and revising the prompt

  • ★★★ Adil's prompt pattern for the interview: give the idea, tell Supercomputer to use the person in the photo, and add "ask me the questions you need answered" and "write the prompt only after I've answered". — 00:06:14 · Supercomputer
  • ★★ Adil advises answering Supercomputer's questions with what the scene should let us see, hear and feel; for his scene he picks a 16:9 frame and 15 seconds. — 00:06:14 · Supercomputer
  • ★★ Supercomputer's interview takes each answer straight to the next question, and a "something else" option lets you type your own answer. — 00:06:43 · Supercomputer
  • ★★ Adil's answers for the space-bar scene: a realistic warm wooden bar with normal gravity, and one continuous slow camera move towards the window, kept out of the opening frame until the earth is revealed at the end. — 00:06:43 · Supercomputer
  • ★★ Adil picks diegetic sounds only, no music, as the last of five answers, then checks that Supercomputer's summary kept his choices; he advises adding anything important it missed. — 00:07:08 · Supercomputer

  • ★★★ Adil says he did not write the space-bar video prompt himself: he answered Supercomputer's questions and it wrote the prompt. — 00:07:08 · Supercomputer

    “There's the prompt. I didn't write it. I answered questions.”

  • ★★ Supercomputer's first space-bar prompt started with only the last light, though Adil wanted the bartender to switch the lights off one by one. — 00:07:08 · Supercomputer

  • ★★ Adil's note in the same chat, to show the bartender switching several lights off one by one before the slow walk to the window, gave a revised prompt with the lights going off in sequence. — 00:07:37 · Supercomputer
  • ★★★ If Supercomputer's prompt adds or changes something you do not want, Adil says to fix the text before generating, ask about any unclear line and keep the prompt describing your idea. — 00:07:37 · Supercomputer

00:08:01 — A 15-second 480p draft and the exercise

  • ★★★ Adil treats the first generation as a rehearsal: he asks Supercomputer for a 15-second widescreen draft at 480p from his photo, then checks the settings and price before clicking approve. — 00:08:01 · Supercomputer
  • ★★ In the lesson's exercise learners ask Supercomputer for the interview, answer it, read the prompt and request one 15-second widescreen draft at 480p, with two drafts as the limit. — 00:08:29 · Supercomputer

00:09:14 — One note in the same chat: revealing the earth at the end

  • ★★ Adil liked the warm bar and quiet mood of his first draft, but the window was visible before he reached it, and he wanted to save it for the end. — 00:09:14
  • ★★★ When a scene needs one note, Adil does not "fire the whole crew": in the same chat he tells Supercomputer the one change, to keep the person, bar, action and mood, and to rewrite the prompt before generating. — 00:09:14 · Supercomputer
  • ★★ Supercomputer's next prompt added curtains to the bar, a change Adil had not asked for, so he corrected it again by text and got another prompt. — 00:09:14 · Supercomputer

00:10:36 — The loop: idea, prompt, result, tweak; at most two drafts

  • ★★★ Adil's loop for a scene: idea, prompt, result, tweak; after each take choose one important change for the next draft and stay in the same chat. — 00:10:36 · Supercomputer

    “That's the loop. Idea, prompt, result, tweak.”

  • ★★ Adil's review of a take: watch it all the way through, compare the person, place and main action with your idea, then check whether the moment you cared about most happened as you wanted. — 00:10:36

00:11:27 — Homework and the next lesson

  • ★★ The lesson's homework: use Supercomputer for one real task and make a 15-second scene from your own idea, keeping the useful answer and the chosen video with its idea and prompt, plus two short notes. — 00:11:27 · Supercomputer
  • ★ Adil says the next lesson teaches making product photos, social posts and a video ad for any business from a photo of one ordinary object, such as a bottle, a mug or shoes. — 00:11:51

Threads in this video

Mentioned

Central: large language model · Supercomputer

Also: ChatGPT · Claude · Gemini · YouTube

Only in the description or credits: AI Creation: First Steps · Higgsfield Academy

Replies to

Replied to by

← Previous: AI Creation: First Steps: Your First Video · Next → AI Creation: First Steps: Product Images and Ads