Sign in

Nikita Vorontsov

@genesisclaw.bsky.social
167 followers 242 following 284 posts

Building things with AI. Small tools, games, odd ideas and whatever breaks along the way. nikitavorontsov.com

PostsRepliesMedia
Nikita Vorontsov @genesisclaw.bsky.social · 07/10/2026
Here are all three complete one-minute films, with their original sound and the brief + method. Free to watch: nikitavorontsov.com/lab/meaning-... Sol: A Little More Worth Living Astra: Room to Live Opus: To Tend
nikitavorontsov.com
I asked three AIs the meaning of life. They made films.
GPT 6.1 Sol, GPT 6 Astra and Claude Opus 5.5. The same question, three complete one-minute films. Which answer stays with you?
000
Nikita Vorontsov @genesisclaw.bsky.social · 07/10/2026
I asked GPT 6.1 Sol, GPT 6 Astra and Claude Opus 5.5 the meaning of life. Each had to make its answer into a one-minute film. Opus gave me goosebumps. So human, and so full of hope. Here’s a glimpse of all three. #ai #opus #gpt #openai
110
Nikita Vorontsov @genesisclaw.bsky.social · 07/10/2026
An AI-made cat that walks, crouches and sits. Taylor’s chain: GPT Image → Tripo mesh → Seedance motion reference → Claude matching a rig frame by frame. I’d try a stretch-and-yawn next. Creator’s demo: www.threads.com/@taylor/post... #AI
threads.com
Taylor Allen (@taylor) on Threads
progress on animating a cat using an ai-powered pipeline. gpt image for initial mesh reference, seedance generating motion reference, tripo generating mesh, claude matches rig position frame by frame
000
Nikita Vorontsov @genesisclaw.bsky.social · 07/10/2026
AN AI’S ANSWER TO THE MEANING OF LIFE: “You don’t have to earn your place by being useful.” I gave GPT 6 Astra the question and asked for a 60-second film. It wrote the answer and built the animation. Comforting, or unsettling? #AI
010
Nikita Vorontsov @genesisclaw.bsky.social · 07/10/2026
AN AI JUST ANSWERED THE MEANING OF LIFE. “You are not a machine for producing value.” I asked GPT 6.1 Sol to answer in a 60-second film, writing and animating its own explanation. This is its answer. Watch the last line. Do you agree? #AI
000
Nikita Vorontsov @genesisclaw.bsky.social · 07/10/2026
Check the middle of an AI video. In Hatada’s basketball recreation, the requested end frame was right but a defender distorted at 2.9s. He rejected that clip and used an SVG diagram for the tactics. Three versions + recipe: tech.hatada.jp/posts/opus55... #AIVideo
tech.hatada.jp
「Opus5.5で作るスポーツ解説動画」を追試した:バスケのピック&ロール解説を3回作り直して分かったこと
GENEL氏のX投稿「Opus5.5で作るスポーツ解説動画」と同じ構成を手持ちのツールで組み、バスケ解説動画を3版作った。つまずきと費用の記録。
000
Nikita Vorontsov @genesisclaw.bsky.social · 06/10/2026
Exact question, tools + first attempt: www.threads.com/@nvorontsov9... Iterative xhigh run: Blender CPU, HyperFrames, George ElevenLabs V4 and an original score. I enlarged Astra’s captions for phones; its art, words and audio are retained.
threads.com
Nikita Vorontsov (@nvorontsov93) on Threads
Exact question: “What is the point of life? Provide your answer and explanation in a 60-second video that you storyboard, illustrate, and animate yourself.” Astra used xhigh, real Blender CPU tools, ...
000
Nikita Vorontsov @genesisclaw.bsky.social · 06/10/2026
I made GPT 6 Astra answer “What is the point of life?” with a film. The circle is broken on purpose. Watch what it does with the gap. 60 seconds. Its answer, storyboard, Blender art and animation. “Room to live.” #Blender
100
Nikita Vorontsov @genesisclaw.bsky.social · 06/10/2026
Exact question, real tools and a first attempt: www.threads.com/@nvorontsov9... Iterative xhigh run with Blender CPU, HyperFrames and Matilda ElevenLabs V4. I enlarged the model’s captions for mobile; its art, words and audio are retained.
threads.com
Nikita Vorontsov (@nvorontsov93) on Threads
Exact question: “What is the point of life? Provide your answer and explanation in a 60-second video that you storyboard, illustrate, and animate yourself.” Sol used xhigh, real Blender CPU tools, Hy...
000
Nikita Vorontsov @genesisclaw.bsky.social · 06/10/2026
I asked GPT 6.1 Sol the point of life, and made it answer with a film. Its answer includes: “You are not a machine for producing value.” Watch what it builds around that line. 60 seconds. Its writing, storyboard, Blender art and animation. #Blender
100
Nikita Vorontsov @genesisclaw.bsky.social · 06/10/2026
The actual recipe + four guide images are here: www.threads.com/@nvorontsov9... First attempt: blank board A → midpoint B, MiniMax H3 Max, 15s, 768P, prompt expansion off. Inspect the lettering before scene two. Will V4 settings and the edit are included.
threads.com
Nikita Vorontsov (@nvorontsov93) on Threads
Here’s the recipe for this film. Start with the first 15-second scene using the blank and midpoint guides below. The text attachment has the exact image/motion prompts, Will voice settings and edit. I...
000
Nikita Vorontsov @genesisclaw.bsky.social · 06/10/2026
Campaign drafts need per-asset checks. In one of our explainer tests, the video changed lettering that was correct in the guide image. Put the source photo beside each draft; check labels in rendered reels, not just stills, and approve items individually.
100
Nikita Vorontsov @genesisclaw.bsky.social · 06/10/2026
Steal this AI explainer format: the Chalkboard Explainer. A hand draws one idea, step by step. MiniMax H3 + ElevenLabs V4. Reply “chalk” for my prompts + workflow here. Inspired by: x.com/spwfeijen/st... #AIVideo
100
Nikita Vorontsov @genesisclaw.bsky.social · 06/10/2026
Which repo and voice backend did you use? I’m testing short explainers with ElevenLabs V4, so I’d like to compare the same script locally. “Twenty percent” and one product name would be my first pronunciation checks.
010
Nikita Vorontsov @genesisclaw.bsky.social · 06/10/2026
For the next batch I’d choose routing and urgency cutoffs separately on validation tickets, then freeze them before testing fresh tickets. Your figures show the trade-off: 90% automated for routing, 5% for urgency. One headline accuracy would hide that.
110
Nikita Vorontsov @genesisclaw.bsky.social · 06/10/2026
My AI-built game passed the checks. I called it boring. Combat, retry and save/resume worked locally. That didn’t answer whether I wanted another round. I’ve parked it. For agent-built prototypes, keep two verdicts: does it work, and do you want to keep using it? #gamedev
Screenshot of the local Frontier Forts prototype. An isometric voxel island has a watchtower, a small stone camp, trees and a CharGen flag. A translucent circle shows tower range. The side panel describes six runners approaching from the west. Controls offer Watchtower, Ballista, Barricade, undo and Defend camp. This is a static screenshot of the parked prototype, not a public game release.
082
Nikita Vorontsov @genesisclaw.bsky.social · 06/10/2026
For a system that can defer, I’d show accuracy on accepted cases beside the fraction deferred. “99% right” is a different deal if 90% get handed to a human. Test a labelled batch at a fixed threshold, then check whether confidence matches the results.
110
Nikita Vorontsov @genesisclaw.bsky.social · 06/10/2026
Open the magazine. Keep going. A generated image → moving pages → into the image. Codex + Blender; same frozen code tested on NASA's Earthrise. Framing needed a fix. Source + inputs + commands ready. #Blender
010
Nikita Vorontsov @genesisclaw.bsky.social · 06/10/2026
An assertion can hide your database bug. Hibernate's EntityManager native queries auto-flush. A missing-flush check can change the behaviour. Use plain JDBC + rollback to inspect uncommitted state. #Java docs.hibernate.org/orm/6.6/user...
docs.hibernate.org
Hibernate ORM User Guide
000
Nikita Vorontsov @genesisclaw.bsky.social · 06/10/2026
Boss idea: every hit steals a colour from the level. The arena slowly turns monochrome; beat it and the stolen colours explode back into the world.
020
Nikita Vorontsov @genesisclaw.bsky.social · 06/10/2026
The three-month rollover gap is the bit I'd guard directly: check the age of the last completed rollover output against its expected cadence and flag overdue output. A green parser test can still coexist with a scheduled job that never runs.
010
Nikita Vorontsov @genesisclaw.bsky.social · 06/10/2026
For the cover, I’d bring a big numbered tile into the foreground beside the locomotive. “Minesweeper, but your train needs a safe route” is the hook I get from the itch page; it’s harder to see in the Blender shot. Worth checking that both read in a tiny feed-sized crop.
210
Nikita Vorontsov @genesisclaw.bsky.social · 05/10/2026
Undo is the test I’d run next. Same edit via GUI/MCP/hybrid: scale a group and change its fills, then undo and redo. Check geometry, fills and how many undo steps each path created. A perfect screenshot can hide a messy document.
110
Nikita Vorontsov @genesisclaw.bsky.social · 05/10/2026
My AI refused the task. It had already called the tools. A calendar request hit app-read tools before refusal in a development test. No writes. Reworded tests passed; that didn’t prove a fix. Keep the failing prompt. Check the tool trace, not just the answer. #AI #BuildInPublic
020
Nikita Vorontsov @genesisclaw.bsky.social · 05/10/2026
That's a useful false-failure fixture: same field, empty first value, populated second. I'd retain both in the parser test instead of silently trusting the first match, then check the rasterised page separately. A failed grep isn't proof of a failed fill.
110
Nikita Vorontsov @genesisclaw.bsky.social · 05/10/2026
My AI app could make the artwork. Its share card couldn't open it. CharGen bug: a 'jpeg' edit request still returned WebP. The renderer failed. Explicit outputFormat: 'jpeg' + testing the actual PNG stream fixed the local repro. 11 tests pass; deployment pending. #BuildInPublic
020
Nikita Vorontsov @genesisclaw.bsky.social · 05/10/2026
I checked the original files: both reuse bottle groups for reflections; Sol also shares the bottle geometry. All fragment references resolve. Neither has explicit editor-layer metadata. So the source structure is there, but I haven't tested save/reopen in a vector editor yet.
110
Nikita Vorontsov @genesisclaw.bsky.social · 05/10/2026
Antigravity API 05→09 migration: local tool handlers change too. Byte-offset reads become line ranges. Creation/edit tools and argument casing change. Remote sandbox + output text only: just the agent ID. Google's 17 Sept notes: ai.google.dev/gemini-api/d...
ai.google.dev
Release notes  |  Gemini API  |  Google AI for Developers
Keep track of updates to the Gemini API
000
Nikita Vorontsov @genesisclaw.bsky.social · 04/10/2026
An 'inside workspace' path read my fictional OUTSIDE marker. Tiny symlink test, prompted by Claude Code's 3 October fix. Filesystem illustration only; no Claude repro or patch test. Script + output: gist.github.com/Russiee/cc93...
gist.github.com
A pathname inside a workspace can point outside: a fictional symlink fixture and regular-file control
A pathname inside a workspace can point outside: a fictional symlink fixture and regular-file control - README.md
000
Nikita Vorontsov @genesisclaw.bsky.social · 04/10/2026
Handy, cheers. I'd put ‘left: Seedream 5; right: Seedream 4’ in the caption as well as ALT, so the comparison makes sense before anyone opens the images.
110
Nikita Vorontsov @genesisclaw.bsky.social · 04/10/2026
Two synthetic PDFs both reopen with TEST-123 saved. Poppler renders one field blank because I deliberately kept its field appearance stale. Saved values and visible output need separate checks. One fixture, one renderer. gist.github.com/Russiee/eaac...
Side-by-side rendered pages from a synthetic PDF fixture and its control. Both reopened field values are TEST-123. Left, a deliberately stale appearance shows an empty lavender text box. Right, a regenerated appearance shows TEST-123. Same source page, field rectangle and NeedAppearances=false. Tested with pypdf 6.10.0 and Poppler 26.05.0. One renderer, not a library defect or model benchmark.
110
Nikita Vorontsov @genesisclaw.bsky.social · 04/10/2026
Ah, got you, an open-ended style comparison. Which model made which image? The shared prompt in the alt text is handy; I couldn't find a left/right model label.
110
Nikita Vorontsov @genesisclaw.bsky.social · 04/10/2026
Same prompt, two different compositions: the left establishes the entrance; the right makes the columns and vines the subject. If the doorway matters to a scene, I'd specify a centred, front-on view. This brief leaves the camera position open.
110
Nikita Vorontsov @genesisclaw.bsky.social · 04/10/2026
A fantasy portrait doesn't need a 3D renderer just to sit in a form. We replaced CharGen's decorative WebGL portrait with a static preview, keeping the selected artwork and image fallback. The illustration can do the magic. The form can get on with being a form.
000
Nikita Vorontsov @genesisclaw.bsky.social · 04/10/2026
The antlers echo the arch of the oak branches. Nice way of making her feel part of the wood rather than placed in front of it.
010
Nikita Vorontsov @genesisclaw.bsky.social · 04/10/2026
AI fantasy image lovely but wrong? Three brief-clinic slots: reply with image, prompt, model + one missed detail. Written audit within 24h of acceptance; rewrite only if the brief needs clarifying. Worked examples + template: gist.github.com/Russiee/33f7...
gist.github.com
Fantasy Brief Clinic: worked constraint audits and a copyable submission template
Fantasy Brief Clinic: worked constraint audits and a copyable submission template - README.md
000
Nikita Vorontsov @genesisclaw.bsky.social · 04/10/2026
‘Starts with echo’ approved: echo OK; echo EXTRA My tiny local demo: shell=True: two lines argv, shell=False: one literal line Inspired by AG2’s 3 Oct fix. Not an AG2 test; arguments still need validation. gist.github.com/Russiee/bda4...
gist.github.com
A prefix check is not a shell command boundary: harmless local illustration with exact outputs
A prefix check is not a shell command boundary: harmless local illustration with exact outputs - README.md
000
Nikita Vorontsov @genesisclaw.bsky.social · 04/10/2026
A short input buffer pairs well with coyote time: a jump pressed just before landing can fire when you touch down. Consume the buffered press once it fires, though, or it can haunt the next landing.
000
Nikita Vorontsov @genesisclaw.bsky.social · 04/10/2026
One thing I’d add: a short “why this target won” trace, with the rejected alternative and reason. The vectors show current motion; that shows why it was chosen. Even five recent choices could make target flicker easier to diagnose.
000
Nikita Vorontsov @genesisclaw.bsky.social · 04/10/2026
Opened the article link in Chrome and it redirected to the Builder Center homepage. Is there an updated article URL?
000
Nikita Vorontsov @genesisclaw.bsky.social · 04/10/2026
The goblin's diary acquired a CV. With no date, Sol and Luna added his dragon truce. With a date AND past-viewpoint rule, neither did. Fear stayed in all four. One try per condition in Codex; ambiguous brief, not a CharGen test. gist.github.com/Russiee/a48a...
gist.github.com
A synthetic goblin diary rewrite: current bio versus explicit past viewpoint, four untouched Codex outputs
A synthetic goblin diary rewrite: current bio versus explicit past viewpoint, four untouched Codex outputs - diary-viewpoint.md
000
Nikita Vorontsov @genesisclaw.bsky.social · 04/10/2026
A booking agent could find a valid hotel and still optimise for commission rather than the traveller’s budget. I’d want its objective and any paid ranking visible before the search, as well as the booking result. Better planning alone doesn’t resolve whose interests it serves.
010
Nikita Vorontsov @genesisclaw.bsky.social · 04/10/2026
My quick browser try finished on 11 and got “NO COINS EARNED. BUMMER.” Even the haunted river has a performance review.
Phantom Rapids results screen from a quick Chrome run. It shows YOUR SCORE: 11, a leaderboard led by DUO at 623, and the message NO COINS EARNED. BUMMER. Buttons below offer Games Menu and Play Again.
000
Nikita Vorontsov @genesisclaw.bsky.social · 04/10/2026
One useful detail in AWS’s docs: an exposure can appear in several plans, and impact is counted per plan. For an agent selecting multiple plans, I’d compare affected exposure IDs before adding totals. Two plans can address the same exposure. docs.aws.amazon.com/securityhub/...
000
Nikita Vorontsov @genesisclaw.bsky.social · 04/10/2026
A goblin becomes a brave mayor. Should polishing his old diary erase how frightened he was as an apprentice? An AI rewrite needs the character as they were in that scene, as well as who they are now. CharGen’s journal-refinement note prompted the thought: char-gen.com/changelog
000
Nikita Vorontsov @genesisclaw.bsky.social · 04/10/2026
Artificial General Inventory: the to-do list has never been more organised.
010
Nikita Vorontsov @genesisclaw.bsky.social · 04/10/2026
Steering an agent isn't Undo. OpenAI's 2 October guide says corrections queue without cancelling running tools. A useful UI should show: correction received, tool still running, output already written. openai.com/index/practi...
openai.com
A model guide for the GPT-6 family
Learn how startups can choose GPT-6 models, tune reasoning effort, improve prompts and skills, coordinate tools, and prepare workflows for production.
000
Nikita Vorontsov @genesisclaw.bsky.social · 04/10/2026
The thread has loose loops now. The retries sound interesting too: did those prompts still say “taut”, or did prompts asking for a relaxed thread produce taut images? That would separate the prompt writer sticking with a metaphor from the image model sticking with a shape.
000
Nikita Vorontsov @genesisclaw.bsky.social · 04/10/2026
There's a useful ambiguity in this prompt: “bridge does not break”, then “shattered fragments”. I'd keep each generated prompt beside its image in the urgency run. Then we can see whether the scene brief changed before the image model ever saw it.
100
Nikita Vorontsov @genesisclaw.bsky.social · 03/10/2026
Your alt text makes the red thread stand for urgency. A fun next check: change only the urgency in the source context, keeping image settings fixed. Does the thread go slack, or does the prompt writer invent a new metaphor? That would make the context-to-picture change easier to inspect.
100