by OpenAI

GPT-6 Astra Prompts

OpenAI's GPT-6 Astra — the frontier reasoning model inside ChatGPT that plans first, then directs image and video generation from a single conversational prompt.

imagevideo

Rebuild my Codex setup around this tree: • Configure persistent dot agent on GPT-6 Astra with cloud container and headless browser access • Set up workspace bridges to route routine dev logs to files and critical architectural gates to Slack • Delegate multi-file coding tickets to local Codex CLI environments with automatic task locking • Add one rule to AGENTS.md: pause autonomous execution and request manual take-over before modifying external infrastructure or security configs Show every change as a diff first. No edits until I say go

build a huge, highly detailed recreation of New York in this Minecraft world. area: 1,024 by 1,024 blocks, centered on the spawn point. plan first, build second. write the plan to a file before you place a single block. the plan must list the districts: financial district, midtown, Central Park and the waterfront. it must also list the road grid with avenues and cross streets, at least 100 skyscrapers of different heights and materials, and the main landmarks. use only vanilla blocks and Minecraft commands: /fill, /setblock, /forceload. build in modules, one district at a time. after each module, check that the blocks are where the plan says, fix what's off, then move on. write one progress line after every module. don't ask me questions, keep working until the whole map is done.

Configure my Codex workspace for hierarchical model selection and subagents: 1. Audit ~/.codex/agents and .codex/agents for subagent profiles fitting lead, workers, and adversary: > Give explorer, worker, and researcher model = 'gpt-6-luna', model_reasoning_effort = 'medium' > Assign worker an isolated git worktree branch to eliminate concurrent file conflicts > Create .codex/agents/adversary.toml with model = 'gpt-6-astra' (upgradeable to 'gpt-6.1-astra'), model_reasoning_effort = 'high', sandbox_mode = 'read-only' 2. Configure model selection in ~/.codex/config.toml: [session] model = 'gpt-6.1-sol' model_reasoning_effort = 'high' [agents] default_subagent_model = 'gpt-6-luna' default_subagent_reasoning_effort = 'medium' max_concurrent_threads_per_session = 8 3. Verify environment variables and routing: check for flags overriding reasoning effort or disabling multi-agent tools. Report them, change nothing 4. Add automatic delegation rules to AGENTS.md: > Summon @ adversary before finalizing multi-file architecture plans > Automatically trigger @ adversary when a test or compiler error repeats twice > Enforce @ adversary contract audit before declaring tasks complete or staging git commits Show every configuration diff first. Do not apply edits until confirmed

generate a pale girl with two black braids and blunt bangs like Wendesday from Netflix in ChatGPT pick a plain white room and the dance everyone already knows swap her into the dance with:

An arctic fox breaks out of an ice cave, sprints across the glacier, leaps off the edge, turns into a snowy owl, dives into the sea and ends as an orca. One chain of transformations, zero cuts

a character swinging on the edge of a snowy cliff, heavy snowfall, a lantern glowing nearby

Build a Chrome extension called Signal Crawler using Manifest V3. The extension should help me find business problems in public web pages. It must work on the page I have open, not silently crawl other sites. CORE EXPERIENCE When I click the extension icon, a small animated crawler appears on the page. It moves between readable text blocks, visually scans them and highlights useful passages. A floating panel shows what it has found in real time. Clicking the icon again removes the crawler, panel, highlights and all changes to the page. WHAT TO ANALYZE Support three source types: 1. Job posts: look for manual, repetitive work, slow workflows and tasks that could be automated. 2. Customer reviews: look for repeated complaints, workarounds, missing features and problems that cost the customer time or money. 3. Product and company pages: look for what the business sells, who it serves and how a proposed fix could fit its existing workflow. Ignore navigation, ads, cookie banners, forms, code blocks and hidden text. Handle pages that load more content while scrolling. THE CRAWLER Draw the crawler on a transparent canvas above the page without blocking clicks or text selection. Give it a compact body, eight jointed legs and visible stepping motion. It should move automatically from one relevant text block to another. When it finds a useful passage, pause briefly, place a foot near it, highlight the passage and send a small animated signal from that passage to the floating panel. Use a restrained visual style: warm yellow for findings, white for the body, dark outlines and a subtle glow. Movement should feel quick and alive, but the text must remain readable. SIGNAL BOARD For each finding, show: - the exact quote - the page title and URL - the type of signal - the problem it suggests - how many similar signals have been found Group related findings into clusters. Do not treat one isolated sentence as a proven market. Make repeated evidence visually more prominent than single mentions. ASTRA ANALYSIS Add an optional GPT Astra analysis step. The user enters their own API key and model name in the extension settings. Store the key locally and make API requests from the background service worker, never from the page. For each cluster, ask Astra to produce: - a plain-English description of the problem - who experiences it - evidence for why it matters - the smallest plausible fix - a short draft offer - what still needs human verification Astra must not invent quotes, companies, prices or customer counts. Every conclusion in the panel should link back to its source passages. PANEL CONTROLS Include Pause, Resume, Analyze, Copy Findings, Export JSON, Clear and Close. The panel should show scanning progress, the number of passages checked, findings collected and clusters formed. Let me open any finding and jump back to its original position on the page. QUALITY REQUIREMENTS The extension must not break scrolling, links, page layout or text selection. It should not send messages, contact companies or publish anything automatically. On Close, remove every canvas element, highlight, listener, observer and timer. Support reduced-motion preferences. Handle API failures by keeping the collected evidence visible. Use plain JavaScript, HTML and CSS. Provide every file needed for a working Manifest V3 extension, then give me clear installation and testing steps.

Set up model routing for this repo: 1. List this month's task types and how long each answer was. Table only, change nothing 2. Anything that ended past 128K goes to Gemini 4 Argon as one call, no chunking > Terminal and shell work stays on Opus 5.5 > Fixes that failed twice go to GPT-6 Astra > Cache the repo context on Argon so every run reuses it 3. Price each task type at Argon's intro rate next to what it costs today 4. Print the routing table with the price column, then wait for my yes before switching anything.

Create a complete, polished motion graphics video called “SIGNAL — Turn ideas into motion” Generate a 20-second motion graphics video at 1920×1080 and 120 FPS. Create the entire animation procedurally in Python. Include the complete editable Python source file. Build the animation, render every frame, inspect the output, and export the finished MP4. Do not stop at a storyboard, mockup, explanation, or individual frames. The final result must be a fully rendered motion graphics video. No external visual assets are necessary. Generate all typography, shapes, backgrounds, grids, lines, cards, transitions, and visual effects directly through code. Create a premium futuristic motion identity for a fictional creative technology company called SIGNAL. The visual language should feel like a combination of modern product design, editorial typography, Swiss graphic design, and high-end technology branding. Palette: Near-black: #111214 Warm white: #F4F1E8 Electric lime: #C7FF45 Soft gray: #A7AAA5 Use a clean geometric sans-serif font such as Geist, Inter, Helvetica, or a comparable system font. Typography should be large, confident, and tightly composed. Use generous negative space. Maintain consistent margins and alignment throughout the animation. Avoid unnecessary decoration. Every scene should have one dominant visual idea. The central visual motif is a SIGNAL LINE. The line begins as a tiny pulse. It expands into a waveform. The waveform becomes a grid. The grid becomes cards. The cards collapse into a single line. That line eventually constructs the SIGNAL wordmark. Every transformation should visibly originate from the previous object. Nothing should feel like an unrelated scene replacement. Motion should feel precise, fluid, mechanical, and slightly weighted. Fast transformations should always resolve into readable compositions. 0–2 seconds: Begin with a near-black background. Place a tiny electric-lime horizontal line in the exact center. The line should initially appear almost motionless. A subtle pulse travels through it. As the pulse reaches the center, the line expands horizontally. Reveal large warm-white typography: “IDEAS” Then underneath: “NEED SIGNAL.” Keep the typography left aligned. Small supporting text: “Turn noise into something worth noticing.” The lime signal line remains visible on the right side of the composition. Allow approximately one second of readable hold. 2–4 seconds: The signal line begins oscillating. Transform it smoothly into a clean waveform. The waveform should react rhythmically rather than randomly. Introduce several thin gray guide lines behind it. The waveform stretches across most of the screen. Large typography changes to: “FIND THE PATTERN” Small supporting copy: “Every idea starts as noise.” Animate individual waveform peaks with slight timing offsets. Keep movement controlled. Do not create chaotic audio-visualizer movement. 4–6 seconds: Freeze the waveform momentarily. Extend vertical lines downward from selected waveform peaks. Extend horizontal lines outward. These lines gradually construct a modular grid. The waveform itself should become one of the grid's horizontal lines. Do not remove it and introduce a new grid separately. Once the grid is established, reveal small labels inside several cells: “IDEA” “BUILD” “TEST” “BREAK” “LEARN” “SHIP” Highlight one cell in electric lime. Headline: “STRUCTURE THE NOISE” Hold long enough for the composition to remain readable. 6–9 seconds: Selected grid cells detach from the grid. They become floating rectangular cards. Each card should preserve its original dimensions and position during the beginning of the transformation. Then allow the cards to reorganize around the center. Use approximately six cards. Cards should have restrained shadows and minimal depth. Most cards should be warm white. One card should remain electric lime. Use short labels: “01 / IDEA” “02 / BUILD” “03 / TEST” “04 / REFINE” “05 / SHIP” “06 / REPEAT” The cards move into a clean horizontal sequence. Headline above: “MOVE WITH INTENT” Motion should have slight inertia while remaining controlled. No bouncing. No excessive overshoot. 9–12 seconds: Compress the cards toward the center. As they approach each other, reduce the gaps between them. The cards merge into one long horizontal strip. The strip becomes electric lime. Preserve the same horizontal axis. The strip begins bending smoothly. Transform it into a flowing continuous line that travels across the screen. Switch the background from warm white to near-black through a geometric wipe generated by the moving strip itself. Large warm-white typography appears: “MOMENTUM” Small supporting line: “ONE MOVE CREATES THE NEXT.” The signal line should travel around the typography without obscuring it. Give the line subtle depth using restrained shading. Do not use glow. Do not use neon bloom. 12–14 seconds: The moving line begins drawing geometric letterforms. First construct: S Then: I Then: G Then: N Then: A Then: L Each letter should emerge from the same continuous line. Avoid simply fading the word into existence. The viewer should clearly understand that the moving signal is physically constructing the wordmark. Once complete, display: SIGNAL Centered. Small text underneath: “TURN IDEAS INTO MOTION” Allow the composition to settle. 14–17 seconds: Collapse the SIGNAL wordmark back into the horizontal line. The transformation should reverse naturally. The line travels toward the center. Create three small pulses alon

Build a single self-contained HTML file: a 3D marble computer, in the spirit of Wintergatan's marble machine and the Digi-Comp II, that adds two 4-bit numbers using nothing but marbles falling through mechanical toggles. Hard rules: 1. One HTML file. Three.js from https://t.co/KCQbEZLusb via an import map, plus its OrbitControls addon. No other libraries, no physics engine, no external assets. 2. The mechanism is the computer. Each bit of the result is stored in a two-state toggle (flip-flop) that a marble physically flips as it passes. Carries happen because a marble leaving one toggle rolls on to the next. Do not compute the answer in JavaScript arithmetic anywhere. The displayed result must be read from the toggle states after the last marble stops. 3. Marbles move along tracks with believable gravity and timing. Nothing teleports. You may script motion along paths, but every toggle flip must be caused by a marble arriving at it. 4. Input: two rows of four switches (A and B, 0 to 15), set by clicking. A lever labelled RUN releases the marbles. A counter shows how many marbles have fallen. 5. Output: a row of five result toggles, most significant bit on the left, and a readout that shows the binary and decimal value read from those toggles. 6. A DEMO button runs 5+6, then 9+7, then 15+15, resetting between runs, so it can be screen-recorded without touching anything. 7. Looks: a wooden board, brass toggles, glass marbles, warm studio lighting, soft shadows, a slow camera that follows the busiest part of the machine. It should look good in a square 1080x1080 recording. 8. Short click sounds via the Web Audio API when a toggle flips, started only after the first user click. 9. It must run at 60 fps on a normal laptop and have no console errors. Output only the complete HTML file.

Create a complete, polished motion graphics video called “FOLD — Make room for focus” Generate an 18-second video at 1920×1080, 60 fps, with original or commercially licensed music and subtle sound effects Include the complete editable source Build the animation, render it, inspect it, and deliver the finished MP4 Do not stop at a storyboard or explanation No external assets are necessary: create the graphics procedurally A premium minimalist motion identity for a fictional creative workspace Palette: Warm paper # F2EFE6 Near-black ink # 181C1D Coral # F45D41 Off-white cards Use Geist or a comparable clean sans serif Large, confident typography with tight tracking Generous empty space, consistent margins, restrained shadows Keep one dominant visual idea in each composition The central motif is folded paper: one coral tile becomes notes, the notes become a ribbon, and the ribbon becomes a custom F monogram Motion should feel precise, fluid, and slightly weighty Fast transitions should resolve into readable compositions Use subtle material shading on the ribbon, without glow or shiny plastic 0–2 seconds: Warm paper background Large black headline on the left: “Make room” “for focus” Small supporting line: “A little less noise. A lot more possibility.” A coral rounded square sits on the right, containing a bold geometric F monogram made from three folded strokes Give the opening a subtle entrance and a readable hold 2–4 seconds: The coral tile travels toward the center and unfolds into seven note cards Stagger their movement slightly so they feel connected Arrange them across two or three loosely balanced rows Use short labels such as: “A fresh idea” “Something to make” “A useful note” “The next move” “Room to explore” “A small detail” “What matters” Most cards are warm white; one is coral Headline above: “A little less noise” 4–6 seconds: The scattered cards smoothly align into one centered vertical stack Headline changes to “One clear direction” Remove the card labels before the cards become too small to read Compress the entire stack into a narrow horizontal coral form Preserve its central position so the next transformation is easy to follow 6–10 seconds: An ink-colored aperture expands from behind the compressed form, filling the background The coral form unfurls into a broad flowing ribbon across the center Construct the ribbon from a continuous ruled surface or tightly joined strips It should gently twist, with coherent front/back shading and no visible cracks Large off-white headline: “Find your flow” Small supporting line below: “SPACE TO THINK / MOMENTUM TO MAKE” Let the ribbon carry the movement while the typography stays stable 10–12 seconds: The ribbon curls into a circular band Keep the same material and color through the transformation Reveal “01” inside the ring Below it, small text: “THING AT A TIME” The ring settles with controlled deceleration 12–14 seconds: Warm paper expands from the center, replacing the dark background The coral ring remains the anchor Show the supporting line: “That’s where good work begins” Allow a short readable pause before the final transformation 14–16 seconds: The ring folds and packs into the geometric F monogram The monogram must be constructed from the same moving geometry rather than appearing as an unrelated replacement Reveal “FOLD” beside it in large black type Below: “Make room for focus” 16–18 seconds: The monogram moves toward its original position on the right A coral tile forms around the moving monogram and follows it Restore the original left-side headline and supporting line Finish in the opening composition, ready to loop Use one coherent coordinate system and persistent objects Preserve a recognizable position, edge, material, or silhouette through every transition Do not crossfade between unrelated full-screen scenes Use critically damped springs or smooth minimum-jerk curves No repeated bouncing Position, scale, and rotation must have continuous motion Text has separate entrance and exit timing from its container Use masked text reveals and restrained fades Never allow outgoing and incoming labels to overlap The final visual state must match the first, including object positions and velocities 120 BPM, nine four-beat bars, exactly 18 seconds Use a restrained electronic rhythm with warm tones, soft percussion, and light transition accents Major transformations should anticipate or land on musical beats Keep effects below the music Use original synthesized audio or assets with a verified commercial-use license Do not reuse audio from reference videos Prefer a single HTML/Canvas/WebGL project with an async window seek(t) function Every visual property must be computed from absolute time No CSS animation clocks, setInterval, requestAnimationFrame-driven state, or accumulated physics during export Seeking frames out of order must produce identical results Render at true 60 fps Use spatial antialiasing and 3–5 temporal samples per output frame for subtle motion blur Keep text sharp during holds Do not use frame interpolation to disguise poor motion Inspect frames throughout the film and dense frame sequences around every transformation Fix overlap, ribbon seams, abrupt replacements, clipped typography, and loop discontinuities before delivery Generic slideshow transitions Particles, shockwaves, neon glows, lens flares, camera shake Random decorative movement Large elastic overshoots Unrelated stock footage Long empty pauses A logo reveal that disconnects from the preceding geometry

Make a dynamic 15-second motion graphics video that shows what an incredible motion designer you are, like it's your showreel for a resume. Go all out.

make a dynamic 15-second motion graphics video that shows what an incredible motion designer you are, like it's your showreel for a resume. go crazy

make a dynamic 10-second motion graphics video that shows what an incredible motion designer you are, like it's your showreel for a résumé. go all out

Build an original SFX palette for my short visual edit in Codex, without a DAW or third-party samples. First map the visual events by function: micro-cut, object lock, texture movement, transition, resolve. Start with only three distinct dry mechanical-digital sounds, not a full pack. Avoid tonal beeps, pitch-sweep lasers, bells, conventional drums and long reverb. Export individual 48 kHz / 24-bit WAVs, a level-matched audition reel and a cue sheet. Check peaks, clean endings and mono compatibility. Don’t claim you can hear the result: ask me which prototype works, which fails and why; iterate from that feedback before making variations.

Act as a security engineer reviewing a repository Analyze the supplied scanner findings and affected source files Prioritize exploitable issues over informational warnings For each confirmed issue, provide the root cause, affected location, minimal patch, and regression test Do not invent vulnerabilities or claim a fix passed without test evidence Never expose secret values in the report

ROLE AND OBJECTIVE You are an elite motion-design studio composed of a creative director, senior 3D artist, motion graphics designer, cinematographer, lighting artist, technical animator, compositing artist, and sound designer. Reconstruct the supplied 16.55-second Sanctum / INF launch advertisement with extremely high visual fidelity. Your goal is a professional, broadcast-quality, 16:9 motion-design film that matches the reference video in its visual identity, animation language, camera choreography, pacing, compositions, transitions, typography, dimensional materials, and overall sense of polish. Do not create a generic cryptocurrency commercial. Reproduce the distinct artistic decisions of the reference. Study the reference video before producing anything. It takes priority over any approximate parameters in this specification. The final animation must feel intentionally designed and directed, not like a sequence of unrelated AI-generated clips. TECHNICAL DELIVERY SPECIFICATION Target duration: 16.55 seconds. Aspect ratio: 16:9. Master render resolution: 3840 × 2160. Working preview resolution: 1920 × 1080. Frame rate: 24 FPS. Progressive scan. High-quality antialiasing. Physically based 3D rendering. Consistent color management across shots. Carefully controlled bloom, reflections, depth of field, and motion blur. No unintended frame interpolation or discontinuity. Continuous audio with synchronized sound design. Final output: high-bitrate H.264/H.265 MP4. Preserve editable source files, assets, and animation timelines. AESTHETIC IDENTITY Create an exceptionally polished visual language combining premium 3D brand animation, futuristic financial-product storytelling, playful geometric composition, and cinematic lighting. The design should feel: Vibrant but controlled. Youthful but premium. Technically sophisticated yet friendly. Visually playful without being childish. High-energy but never chaotic. Glossy and dimensional rather than flat. Full of subtle optical effects, translucency, and colorful reflections. Combine two distinct worlds. WORLD A — BRIGHT IRIDESCENT BLUE: An immersive bright electric-blue gradient environment with cyan, sapphire, and subtle greenish or pastel illumination. The background contains large soft gradients, gently fluctuating light, and faint atmospheric texture. Floating glossy black INF tokens exist at different depths, some small and distant, others enormous and partially cropped by the boundaries. Small decorative geometric particles float among the tokens: four-pointed sparkles, soft concave diamond shapes, neon-green stars, yellow and orange squares, tiny magenta pieces, and white reflective accents. There is no literal room, landscape, ground plane, horizon, or architectural structure. The camera occupies an abstract, infinite, dimensional space. WORLD B — DARK CINEMATIC STUDIO: A nearly black or midnight-navy void illuminated with luminous cobalt, violet, and cyan spotlights. Subtle floor reflections suggest a glossy studio floor beneath the objects. Soft pools of blue light create grounding without introducing recognizable walls. Objects cast soft shadows and contribute reflected light to nearby surfaces. The scene has dimensional volume, controlled darkness, and pronounced separation between background and glowing 3D elements. Never use a simple black solid as the only environmental treatment. Preserve nuanced lighting, ambient diffusion, and slight colored atmospheric depth. CORE OBJECT DESIGN — INF TOKEN Model the primary INF token as a thick, glossy, circular 3D coin. GEOMETRY: Circular silhouette with significant visible thickness. Rounded edge bevel. Smooth polished cylindrical outer wall. Dark black or charcoal material. Slight glossy edge highlights. A circular face with a centered abstract INF graphic. FACE ARTWORK: A soft, multi-lobed, organically curved shape. Intersecting or overlapping rounded color fields. Iridescent gradient regions flowing from pastel lavender and blue into cyan, vivid lime, yellow, orange, and subtle magenta. The mark feels like abstract, liquid color geometry inside the black coin. Smooth boundaries and softly blended color transitions. No noisy texture, excessive surface detail, or illegible ornamentation. MATERIAL: Polished black polymer or ceramic outer body. Subtle metallic and glasslike response. Strong controlled specular reflection. Physical bevel highlights. Rich dark surface that never looks pure flat black. The center artwork is colorful and luminous but retains clear shape definition. ROTATION: Coins rotate along more than one axis. Use different angular velocities and phases for each instance. Their movements should reveal thickness and surface reflections. Avoid repetitive synchronized spinning. SECONDARY TOKEN SYSTEM Create a collection of distinct 3D liquid-staking token designs. Include recognizable variations of: Black SOL-marked token with colorful curved striped symbol. Purple and magenta glowing coins. Solana-inspired striped tokens. Yellow and orange branded coins. Cyan-and-white glowing symbol coins. Black-and-gold ring tokens. Smaller INF-style tokens. Each token must have: Real thickness. Individual material treatment. Physically coherent reflections. Distinct orientation. Independent rotation. Rim highlights. Controlled emissive or neon accents where appropriate. Appropriate front and side textures. Do not make every token identical except for the face image. PARTICLE SYSTEM Use a limited field of designed geometric accents. Types: Concave four-point sparkle forms. Bright lime-green diamond stars. Small orange and yellow quadrilaterals. Tiny violet and magenta specks. White glimmering star highlights. Distant miniature coins. Particles must vary in size, focus, distance, and movement. Foreground element

Create an original Japanese torii shrine scene using PixVerse and Blender. Complete these steps: Create the concept: Use PixVerse to generate a photoreal reference image. Lock the torii design, materials, lighting, and environment for consistency. Build the 3D scene: Create an editable textured model in Blender. Clean the topology, unwrap the UVs, and add suitable animation controls. Add movement: Include foreground leaves that partially obscure the wide shots, gently moving foliage, and birds flying visibly across the scene. Animate the cameras: Create a 12-second sequence with three shots: an arcing approach, a lateral detail shot, and a rising pullback. Keep movement smooth and framing clear. Render in Blender: Export a native camera video and a separate neutral gray clay reference video using the same animation. Save them beside the Blender project. Generate the photoreal film: Use PixVerse Seedance with the gray video for camera and action reference and the unchanged photoreal image for appearance. Generate a 1080p, 16:9 video with only natural environmental sounds—wind, rustling leaves, and bird wings. No music or voices. Check and deliver: Review every shot for consistent design, photoreal textures, visible birds, smooth motion, and correct audio. Fix issues within the approved budget and document remaining limitations. Save the editable Blender project, videos, reference image, and a brief production assessment in the same folder. Key points: Never use the colored low-poly video as the Seedance reference. Check the entire film, especially the final wide shot.

Is this the highest confidence solution for this session, with no clear gaps, no overengineering, in — AI image prompt — GPT-6 Astra

Is this the highest confidence solution for this session, with no clear gaps, no overengineering, incredibly surgical but still efficient with no drop in quality or performance, and the least likely to cause any negative impact? No drift or staleness in assumptions? Audited a few times over?

read x live for the last 72 hours. find complaints, not hype: people describing a problem a cheap product could fix. return the top 5 problems with post counts, 3 example posts each with links, and how many ads already target it. don’t pick products.

You are the founder, GPT-6 Astra is the operator, and every function of the company becomes a node with an owner, an output and a check. Seven layers do the work: > Functions -> nodes, edges, parallel work. > Agents -> one subagent per node, only the context it needs. > Code nodes -> anything with one correct answer runs as code. > Permissions, gates, return paths, weekly review. The founder approves money, hiring and anything public. Everything else runs in the loop and lands in company.md, checks.md and progress.md

Rebuild my Codex setup around this tree. 1. Agents - scan ~/.codex/agents and .codex/agents for anything that already fits explorer, worker or researcher - draft TOML only for missing roles: explorer and researcher on gpt-6-luna, worker on gpt-6.1-sol, all at model_reasoning_effort medium - explorer and researcher get a read-only sandbox, only worker can write - skip agents pinned to another model and list them 2. Architect - add an architect on gpt-6-astra, model_reasoning_effort high, read-only sandbox - its only jobs: review plans, repeated errors and finished work - feed it the plan, the diff and the failing log, never the full transcript - force one output shape: verdict (proceed / revise / stop), top 3 risks, one cheaper path, under 200 words - cap it at 3 calls per task 3. Main config - in ~/.codex/config.toml set model to gpt-6.1-sol, model_reasoning_effort to high, approvals_reviewer to auto_review 4. Overrides - find anything that would beat this: active profiles, flags in shell aliases, agents.default_subagent_model - report it, change nothing 5. Rules - add to the AGENTS file in the repo root: spawn the architect before a large plan, when the same error signature shows up twice, and before calling a long task done - log every architect call with its trigger and verdict to .codex/architect-log.jsonl Check that every TOML file parses, then show me all changes as one diff. No edits until I say go.

Do a full animation with sound of anthropic mascote with the newset model of Opus 5.5 fighting against OpenAI mascotte with the new astra models Sol etc like the intro of pokemon fire red and leaf green. Same animations

Set up this Dots + Codex workflow for my current project: 1. Have Codex inspect ~/.codex/agents and .codex/agents for explorer, researcher and worker roles. Read the applicable AGENTS.md instructions. Reuse compatible agents; list conflicting instructions and pinned models before proposing replacements. 2. Draft valid TOML files only for missing roles, including name, description and developer_instructions: > explorer: gpt-6-luna, medium effort → trace relevant files, callers and existing tests; no edits > researcher: gpt-6-luna, medium effort → check documentation and API constraints; no edits > worker: gpt-6.1-sol, medium effort → make scoped changes and run targeted tests 3. Draft model = 'gpt-6.1-sol' and model_reasoning_effort = 'high' in ~/.codex/config.toml. Check overrides (project config, active profiles, shell aliases, launch flags, task settings, agents.default_subagent_model and agents.default_subagent_reasoning_effort). Report conflicts without changing them. 4. Draft one workflow rule in AGENTS.md: delegate independent investigations to explorer and researcher, give worker their findings, and return checkpoints before major implementation, after a repeated failure and before completion. Review those checkpoints yourself, then send the same Codex task a revised plan, a targeted investigation or remaining corrections. Show all proposed changes as diffs first. Apply nothing until I approve. After approval, verify the roles load with the intended models. Preserve existing permissions; do not merge or deploy.

make a dynamic 15 second motion graphics video that shows off your best work like it's a resume showreel, go all out

Build a detailed sports car that transforms into a humanoid robot, with an x-ray mode and interactive exploded view

Make a 2-minute sand animation that tells the story of 100 years of Hong Kong history. Keep it lively, engaging, and tasteful. Add appropriate background music and sound design.

make a dynamic 15-second motion graphics video that shows what an incredible motion designer you are, like it’s your launch video. go crazy

make a dynamic 15-second motion graphics video that shows what an incredible motion designer you are, like it's your showreel for a résumé. go all out.

Make a dynamic 20-second motion graphics launch video for my open-source project jev-browser-skill ('Give a site. Write a goal. Jev clicks.'), like it's the hero video on its GitHub README. Go all out.

Related models

GPT Image 2.5

OpenAI's GPT Image 2.5 — up to 50% faster than Images 2, with pinpoint local edits that leave the rest of the frame untouched and reference subjects that stay recognizable across repeated rounds.

GPT Image 2

OpenAI's top-ranked image model — it plans the layout, reasons, and self-checks before it draws, nailing complex multi-part instructions.

Seedance 2.5

Seedance 2.5 is a next-gen AI video model that generates continuous, synchronized 30-second 4K clips from a single prompt.

MiniMax H3

MiniMax H3 is a powerful multimodal video generation model that creates up to 15-second 2K videos with native stereo sound from text, image, video, and audio inputs.

Browse AI Image Prompts Browse AI Video Prompts All models