ZukunftBilden
Deep Dive archive

05.10.2026 · Retrospective and resources

Videos, Audio Dramas and Apps from One Prompt

Retrospective of 5 October 2026: with Opus 5.5 and Sol 6.1, videos, audio dramas and apps come from a few prompts at an affordable cost. Includes chapters, transcript, the examples shown and a practical pack on subscriptions, voices, Dots and local models.

The written retrospective covers the talk and the open Q&A. The transcript was generated automatically and may contain errors. The recording is embedded below. Sources checked on 5 October 2026.

What I took away

Retrospective of 5 October 2026

In September the competence was there, but the price was not yet right. Astra used up the weekly limit so quickly that a single “hello” became a joke. In October both come together: with Opus 5.5 and Sol 6.1, videos, games, websites and apps can often be built in a few prompts, at a cost you can carry alongside a regular job. My core message of the evening was therefore: now is the moment to start your own project. This assessment comes from my own work and the examples shown.

3D, motion graphics and sound from code

The creator examples range from a boat ride through Japanese landscapes to a space game with more than 80 planets, built from one prompt with Opus 5.5 and three.js according to the creator. Sound effects and soundscapes are now generated directly in code as well. For explainer videos in the style of 3Blue1Brown, the model now uses its animation library cleanly in my experience. Opus 5.5 built the evening’s presentation and the page with all the posts shown in about 15 to 20 minutes each.

A website with a sales video in an hour and a half

On the Saturday before the Deep Dive, the young people at Colearning and Fabian described what their site should show. Website, soundscape and sales video came from the same prompt, followed only by small corrections. After about an hour and a half, everything was finished. I estimate the first run used three to four percent of a weekly limit.

Sol 6.1 executes, Opus 5.5 designs

Sol 6.1 comes close to Astra, works reliably, checks its own work and is very good at computer use. It is my default for automation and, by my calculation, about five times cheaper than Astra. Opus 5.5 has better taste. If I could choose only one tool, it would currently be Claude Code for 200 euros a month. This recommendation may change again with the next model.

An audio drama from one prompt

Chapter 1 of my novel The Human Trial is available as an audio drama with several voices, breathing, music and sound effects. Opus 5.5 directed the script, casting and music. It received an API key from Google AI Studio and used Gemini 3.8 Flash TTS for the voices and Lyria for the music.

Jev and decisions as probabilities

Jev from TypeSafe AI answers typed questions with values and probabilities between 0 and 100 percent. According to the provider, input is very cheap and output is free. I used it to turn an old idea into a matching app for web, iOS and Android, backend included, in one prompt.

Dots: an agent that works around the clock

OpenAI’s Dots runs continuously, reaches out proactively and works on its own cloud computer with Blender, GIMP, Godot, CAD tools and a terminal, and on your own machine when needed. In the EU, Dots is currently only enabled on the Business plan, probably for data protection reasons. Its biggest advantage for me: Dots runs on Astra without a token limit.

AGI, research and robotics

Alan Thompson’s conservative AGI countdown stands at 99 percent, and the latest percentage point came from robotics. We talked about the deciphered Napoleon cipher, newly solved maths problems and robots that fold towels and make beds in homes they have never seen. They still move slowly, but at a pace that can already be useful.

Two levers

In the short term, this is a time for arbitrage: anyone who masters these tools can build, animate and program what almost any company needs on a computer. In the long term, I am investing in growth relationships in which people see, value and challenge one another. This is my strategic view, not a forecast with a fixed deadline.

The resource page brings together chapters, transcript, the posts shown, a practical pack and invitations to upcoming dates.

Recording · 5 October

Watch the session in the Google Drive player or open the recording directly in Drive.

Open in Google Drive

Chapters from the session

Approximate timestamps from the transcript, provided for orientation rather than as video links.

  1. 00:00:00Competence is solved, compute becomes affordable
  2. 00:05:283D worlds and sound effects from code
  3. 00:14:23Explainer videos and motion graphics
  4. 00:15:43Colearning: website and sales video in an hour and a half
  5. 00:17:00Which subscription pays off
  6. 00:23:58Sol 6.1 as a reliable executor
  7. 00:25:15An audio drama from one prompt
  8. 00:31:27Jev and an app for three platforms
  9. 00:39:54Rethinking old projects and the arbitrage window
  10. 00:44:03Dots: an agent that works around the clock
  11. 00:53:11The AGI countdown at 99 percent
  12. 00:56:42Robots in unfamiliar homes
  13. 01:01:39Community projects, two levers and invitations
  14. 01:11:30Computer use, projects and mods in Claude Code
  15. 01:22:36Music, images and video: Suno, Lyria, FLUX 3 and Seedance 2.5
  16. 01:31:36The moment to get started
  17. 01:34:33Q&A: Hermes, Paperclip and local models

Practice from the session

Subscriptions, prompts and setups as I use them in October 2026. Prices, availability and terms change quickly, so check them before you buy or connect anything.

01Which subscription pays off

If you choose only one tool, choose Claude Code on the 200-euro Max plan for now. For everything at full speed, I budget about 500 euros a month: 200 euros for Claude, 200 euros for Codex and the rest as API credit, for example in Google AI Studio for voices, music and video. Sol 6.1 is my reliable executor with computer use and, by my calculation, about five times cheaper than Astra. Opus 5.5 has the best taste for design. This recommendation applies to October 2026 and may change with the next model.

02How I prompt Opus 5.5

Three rules from my own work: Opus should respect everything that already exists and improve it intelligently. In parallel, I let it build a second version for which it reviews every feature, rethinks the architecture and builds everything better from scratch. I keep the plan short and often skip it entirely.

Prompt · Improve what exists

Review the whole project before you change anything: every feature, piece of content, data source and earlier decision. Respect everything that exists and improve it intelligently. Keep what works and justify every larger change in one sentence. Then test the most important workflows from the user’s perspective and show me what changed.

Prompt · Rethink a second version

Review every feature of this project and rethink the architecture. Consider how it would look and work if you built it better from scratch today. Build this second version alongside the existing one and leave the first version unchanged. Finally, compare both versions and list what the new one does better and what it still lacks.

03Audio drama, voices and music

Create an API key in Google AI Studio and pass it to your agent through an environment variable rather than in the chat. That gives it Gemini 3.8 Flash TTS for voices with breathing, emotion and accents, and Lyria for music. Opus 5.5 writes the script, casts the voices and directs music and sound. For songs with vocals, Suno remains my first choice; commercial rights require a paid plan. If music generated in code sounds like MIDI, use a finished track from Suno or Lyria that you hold the rights to.

04Images and video

For image editing over many rounds, I showed FLUX 3. In the comparison, the image stayed stable after many changes, while quality with Images 2.5 declined with each round. Seedance 2.5 suits video with multiple shots and audio, and ElevenLabs’ new Eleven v4 suits voices.

05Dots in Europe

In the EU, Dots is currently only enabled on the Business plan, probably for data protection reasons. In the session I showed how I activated Dots once through a VPN; the free version of Proton VPN worked for this. After connecting it to Slack once, the dot remained reachable there without the VPN. Dots runs on Astra without a token limit and has its own cloud computer with Blender, GIMP, Godot, CAD tools and a terminal. Check OpenAI’s terms of use first, and whether this use is permissible for you and your data.

06Computer use in Claude Code

Claude Code can use the same computer use as Codex; the guide is in the linked post by argofowl. In my test, it worked in the background with its own mouse pointer while I kept working in parallel. Give the agent access only to the applications and accounts it needs for the task.

07Projects, mods and Dots

A project in Claude Code connects to GitHub, knows your repositories and keeps running in the cloud. While your computer is on, it fetches context from it when needed. You can close the laptop and continue from your phone. A project belongs to one topic and creates its own threads for new subtasks. Dots, by contrast, is a single agent for everything that reaches out proactively. Mods load a bundle of skills, context and plugins for a particular kind of work, such as videos, websites or presentations. My impression is that mods have not yet fully rolled out in Europe.

08Jev for fast decisions

Jev from TypeSafe AI returns values and probabilities. That makes many small decisions cheap: which part of a page is relevant to you, which model should answer a question or how well two profiles match. According to the provider, output is free and input is very cheap. Jev is currently in early access.

09Hermes, Paperclip and local models

Hermes is open source under the MIT licence and lets you choose the models freely. At first I wanted a single agent that did everything. Clearly scoped tasks and several agents work better. For that, I put Paperclip on top; it manages communication, budgets and skills. Underneath, I run Hermes, Claude Code and Codex. For local models, Can I Use LLM and CanIRun.ai show what runs on your hardware and which quantisation you need. On arena.ai you can compare open-source models by task area.

Original sources and follow-up reading

The X posts come from the collection I showed during the evening. They document creator and vendor demonstrations, not independent verification. Official pages and documentation provide follow-up context. Figures from posts reflect what the creators report.

01

Models and tools from the session

Official sources for the models behind the examples shown.

  • Claude Opus 5.5

    Anthropic’s introduction to Opus 5.5, including use cases and pricing.

    www.anthropic.com

  • Claude Code

    Documentation for Claude Code, the workspace where I use Opus 5.5.

    code.claude.com

  • Day 1: faster default speed · Tibo

    According to an OpenAI employee, Astra and Sol 6.1 run about 50 percent faster at default speed. Part of the daily launch series.

    Creator / vendor post

  • Jev · TypeSafe AI

    Jev answers typed questions with values and probabilities instead of text. Currently in early access.

    typesafe.ai

  • System One models · documentation

    The concept behind Jev, with examples of queries and answer formats.

    docs.typesafe.ai

  • FLUX 3 Image

    Image model from Black Forest Labs. In the comparison shown, images stayed stable across many editing rounds.

    bfl.ai

  • Seedance 2.5

    ByteDance video model with audio, multiple shots and reference images.

    seed.bytedance.com

  • Eleven v4

    ElevenLabs’ new speech model, shown together with Seedance 2.5 during the session.

    elevenlabs.io

  • Raphael AI

    Image generator with several models and a free tier, according to the provider.

    raphael.app

02

Audio drama, voices and music

What you need for an audio drama from one prompt, and where usage rights are set out.

03

Agents and computer use

Dots, computer use in Claude Code and the new mods.

  • Dots · OpenAI

    Official introduction to the always-on agent. Check EU availability and the terms of use.

    openai.com

  • Computer use in Claude Code · argofowl

    Creator guide to using Codex computer use from Claude Code. Review permissions before setting it up.

    Creator / vendor post

  • Claude Code mods · video

    German-language video on mods: bundles of skills, context and plugins for a particular kind of work.

    www.youtube.com

04

Open source, multiple agents and local models

From the Q&A: Hermes, Paperclip and how to choose a local model.

  • Hermes Agent

    Open-source agent under the MIT licence, with a free choice of models.

    github.com

  • Paperclip

    Organises multiple agents with roles, budgets and rules for who talks to whom.

    paperclip.ing

  • Paperclip · GitHub

    Source code under the MIT licence.

    github.com

  • Can I Use LLM?

    Detects your hardware and shows which open models run locally and at which quantisation. Shared in the session chat.

    www.caniusellm.com

  • CanIRun.ai

    A second option for the same check: which open models run on your hardware and at which quantisation.

    canirun.ai

  • Arena · Leaderboard

    Rankings by task area, including open-source models.

    arena.ai

05

Opus 5.5: motion graphics and design

Creator examples shown in the session. Runtimes, costs and prompts are as reported by the creators.

  • 3Blue1Brown · YouTube

    The channel behind the explainer style Opus 5.5 can now recreate. Shared by Eva-Maria in the chat.

    www.youtube.com

  • Animation from one prompt · donald

    According to the creator, five minutes of speaking, then Opus 5.5 worked for twelve hours. The prompt is in the thread.

    Creator / vendor post

  • Visual design · NotinReality

    The creator’s assessment of Opus 5.5’s visual design, based on their own model comparison.

    Creator / vendor post

  • Showreel comparison · kirill sh

    The same prompt for a 15-second clip with Opus 5.5 and GPT-6 Astra. A single comparison with the creator’s verdict.

    Creator / vendor post

  • Boat ride in three.js · Meng To

    Playable 3D scene with weather, day and night lighting and characters, generated with Opus 5.5 according to the creator. Steered live during the session.

    Creator / vendor post

  • Valley · live demo

    Linked demo for the boat ride. Availability and rendering may change.

    valley.mengto.here.now

  • A $1,874 island · Dan Greenheck

    Detailed 3D world with animals and wind. The creator puts the token cost at US$1,874.40.

    Creator / vendor post

  • A paper as an explainer · Deedy

    Eight minutes in the style of 3Blue1Brown from a research paper, made with Opus 5.5 according to the creator. Check the content with an expert.

    Creator / vendor post

  • One prompt and music · Min Choi

    One prompt plus a music file; according to the creator, twelve hours to the finished video.

    Creator / vendor post

  • Motion design studio · Movez

    A detailed course on getting high-quality motion design from Opus 5.5. Recommended in the session as a learning resource.

    Creator / vendor post

  • Space game · Rubzem

    Game with more than 80 planets, biomes and weather, built from one prompt with Opus 5.5 and three.js according to the creator.

    Creator / vendor post

  • Doodle chess · Kevin Ngo

    Chess that looks like drawings on paper, built with Claude Sonnet 5.5 according to the creator.

    Creator / vendor post

  • Video from code · YouTube

    Short 2D and 3D video, created entirely in code by Opus 5.5 according to its description.

    www.youtube.com

06

Compute becomes affordable

Sol 6.1 and Luna, demand for Astra and what a token budget builds today.

07

Spaces, CAD and robotics

Creator examples in which models work with spaces, mechanics and camera images.

  • Interactive anatomy · ashe

    3D website with 2,234 modelled parts, made with GPT-6 Astra according to the creator. Check medical claims separately.

    Creator / vendor post

  • Studio in 3D · Roberto Nickson

    Nine photos as the starting point for an interactive 3D model of the room, according to the creator.

    Creator / vendor post

  • Agentic CAD · adam

    The creator’s assessment of GPT-6 Astra in mechanical design. An engineering review is still pending.

    Creator / vendor post

  • Brain in 3D · Axultan

    Visualisation with more than 166,000 neurons, according to the creator. Read the original for scientific context.

    Creator / vendor post

  • Hardware by agent · nim

    Astra designed, ordered and documented a mini DJ controller with the creator’s credit card. Purchases by agents need clear limits.

    Creator / vendor post

  • Robot painting · thijs

    Camera feedback and repeated attempts at painting the Golden Gate Bridge.

    Creator / vendor post

  • Robot control · Jay Chooi

    Creator report comparing GPT-6 Astra and Fable 5.1 on a control task. Test conditions are in the thread.

    Creator / vendor post

08

Image, video and voice

Images 2.5, a conversational video model and real-time voice.

  • Images 2.5 · Sam Altman

    OpenAI’s CEO announcing ChatGPT Images 2.5.

    Creator / vendor post

  • Stop-Motion · Charlie Guo

    Creator example of a stop-motion animation made with GPT-Image 2.5.

    Creator / vendor post

  • Images 2 and 2.5 · Chetaslua

    Visual creator comparison. Individual examples do not establish reliable consistency.

    Creator / vendor post

  • Griffin · Tavus

    Vendor claim: 48 percent of people who spoke to the video model live thought it was human.

    Creator / vendor post

  • GPT-Live-1 · OpenAI Developers

    Speech model for natural real-time conversation, now available in the API according to the vendor.

    Creator / vendor post

  • ChatGPT Voice · OpenAI

    Voice mode with plugins for email, calendar and Slack, plus access to Astra, Sol and Luna.

    Creator / vendor post

09

Dots and agents in everyday life

OpenAI’s always-on agent and other agents that work without constant input.

10

Research and the AGI question

Mathematics, cryptography, self-improvement and safety, plus the sources for the AGI countdown.

  • AGI countdown · Alan D. Thompson

    A conservative countdown to AGI with a definition and the reasoning for each step. Currently at 99 percent.

    lifearchitect.ai

  • ASI checklist · Alan D. Thompson

    Checklist for superintelligence, including solved mathematical conjectures.

    lifearchitect.ai

  • SimpleBench

    The test on which, according to lifearchitect.ai, Fable 5.1 overtook the human baseline in September.

    simple-bench.com

  • VibeMathed · stats

    Overview of maths problems solved with AI, including the share of formally verified proofs.

    vibemathed.com

  • Napoleon cipher · Carter Church

    According to the creator, GPT-6 Astra broke a cipher unread for 217 years in about six hours, starting from one image and a goal.

    Creator / vendor post

  • Navier-Stokes · OpenAI

    The vendor’s report on a proposed solution to the Millennium Prize Problem. It does not imply that a prize has been awarded.

    Creator / vendor post

  • Fly brain plays Beat Saber · lyra bubbles

    A short post about a simulation. Check the background and method in the original.

    Creator / vendor post

  • Dream-RSI · Dr Singularity

    Summary of Google DeepMind work on recursive self-improvement. Details are in the paper.

    Creator / vendor post

  • We Must Pace the Frontier · Dario Amodei

    Essay with a three-part plan on the pace of AI development. The author’s position.

    Creator / vendor post

  • Mythos report · ℏεsam

    A creator’s summary of an Anthropic report on Mythos. Read the original report for context.

    Creator / vendor post

  • HomeBody · Stanford and Caltech

    Humanoid robot that explores unfamiliar homes, remembers places and completes household tasks.

    www.youtube.com

11

Built in Salzburg

Projects from my month and from the community, public and ready to click.

  • Colearning · the young people’s website

    Website, soundscape and sales video from one prompt with Opus 5.5, finished in about an hour and a half.

    kimagie.netlify.app

  • Fabian Martin · new website

    Fabian’s website built with Opus 5.5, including an animated string that follows your scrolling.

    fabian-martin.at

  • Freiraum · Fabian Martin

    Fabian’s learning game about personal finance, built with Codex Astra. The first level is playable.

    freiraum-spiel.netlify.app

  • Matching app · Blind Date

    The app from the session: web, iOS and Android from one prompt, using Jev for its assessments. The launch follows later.

    blind-date-relaunch.web.app

  • YMAD · sales film

    Sales film, script and structure produced entirely in code, plus a guided demo.

    zukunftbilden.eu

  • YMAD

    Managed AI marketing department from Salzburg.

    ymad.ai

  • Film “Von Grund auf gedacht”

    A 59.5-second collage film for Freie Schule Salzburg, drawn in JavaScript on canvas.

    freieschulesalzburg.at

  • Heureka · afternoon programme

    The Matura afternoon on the Heureka learning platform, with its landing page, built in three days.

    paideia-lernwelt.web.app

12

Come and visit

Dates and places where we continue the conversation after the Deep Dive.

  • Dragon Dynamics

    Tabletop role-playing at the Penthouse. Free to try on the first Sunday of every month.

    dragondynamics.at

  • Colearning

    Free trial day for young people, every Saturday from 10:00 to 16:30. Get in touch beforehand.

    zukunftbilden.eu

  • Freie Schule Salzburg info evening · 8 October

    Thursday, 8 October 2026, 19:00 to 20:30, at Strubergasse 26 in Salzburg or online. Further info evenings on the second Thursday of each month.

    freieschulesalzburg.at

  • Public quest board

    Submit requests to the school’s children and young people, such as 3D prints or cushion covers.

    paideia-lernwelt.web.app

  • KI-Kickstart

    In three months: a computer, team training, strategy, the first productive agents and a custom platform. 1,000 euros for a referral.

    zukunftbilden.eu

  • Raustruppe · WhatsApp group

    A shared experience once a month, preferably outdoors. Last time, eight of us went hiking.

    chat.whatsapp.com

Downloads

ZukunftBilden

GmbH

Part of ZukunftBilden GmbH

AI Deep Dives are a practical lab within a larger education system for the future.

At zukunftbilden.eu you will find the wider places, courses, communities and systems behind this format.

zukunftbilden.eu