Загрузка видео...

Не удалось загрузить видео

На главную

I gave 4 frontier models the same prompt to create an Ecco the Dolphin style video game 100% procedural 🐬 Same task, official chat UIs, max thinking. Watch 👇 Claude Fable 5 Z.ai GLM 5.2 @ChatGPTapp GPT 5.5 Pro Google Gemini Flash 3.5 Note: Fable 5 version was created...

44,413 просмотров • 4 месяцев назад •via X (Twitter)

Комментарии: 55

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

Prompt if you want to test: --- Build "TIDESINGER", a complete Ecco the Dolphin-style game as ONE self-contained HTML file. No external assets, no libraries, no network. Canvas 2D only. HARD CONSTRAINTS - Single file, runs from file:// on desktop and mobile Safari/Chrome - 100% procedural graphics (code-drawn) and 100% procedural WebAudio - Allocation-free hot loop: typed-array pools w/ swap-remove, prerendered offscreen sprite canvases for all glows/bubbles/motes, cached Path2D terrain, viewport culling on every entity, DPR capped at 2 - 60fps target on mid-range phones; fixed max dt clamp GAME FEEL (non-negotiable) - Momentum swimming: thrust + exponential drag underwater, real gravity above the waterline so breaching produces ballistic arcs with splash in/out - Charge burst on cooldown (Ecco's dash), damages predators at speed - Sonar pulse: expanding ring; stuns jellyfish, spooks sharks, interrogates glyphs, fires a direction ping toward current objective - Dynamic camera: velocity lookahead + speed-based zoom out + cinematic zoom/letterbox/slow-mo on story beats; hit-stop on heavy impacts WORLD: 5 biomes across one continuous 15k-unit map, distinct palette grading 1 SUNLIT SHALLOWS animated caustics, coral, tutorial 2 KELP FOREST dense weeds, lateral current, crystal 1 3 OPEN TRENCH floor dips deep, downdraft current, passing WHALE silhouette w/ moan (ambient signature moment), crystal 2 4 ABYSS CAVERN rock ceiling blocks surfacing; darkness with a light radius around the dolphin (sonar widens it); AIR POCKETS under ceiling domes are the only oxygen; crystal 3 5 SONG GATE + HOME REEF barrier wall opens only when all 3 songs are sung at the glyph; goal ring ends the run SYSTEMS - Air meter (drown damage at zero), vitality, fish schools that flee and heal when eaten, jellyfish drift hazards, patrolling/chasing sharks (2 hp, flee when beaten), urchin spike hazards on rocks - Checkpoint shells per biome (respawn keeps progress), score with fish-streak multiplier, breach style bonuses, floating score popups, minimap progress bar with objective markers - States: animated title, pause, death, win (time + score + fish stats) RENDERING QUALITY BAR - Spine-based dolphin (18 segs): undulation injected along spine, counter-shaded body perpendicular to travel, dorsal/pectoral/fluke with independent motion, blowhole spray on surfacing - Sunset sky, sun glow, parallax islands; 3-layer rock parallax underwater; god rays; drifting plankton motes; depth fog + per-biome color grade (multiply pass); caustic light webs in shallows; foam flecks on wave crests; vignette; additive glow sprites for all emissive objects AUDIO (WebAudio, all synthesized) - Filtered-noise ocean bed + detuned drone, depth-tracked lowpass - Pentatonic sonar chirps, crystal arpeggio, splash noise bursts, gate rumble, whale moan, low-air heartbeat CONTROLS - Touch: left-half floating stick, SONAR + CHARGE buttons, pause button - Keyboard: WASD/arrows, SPACE sonar, SHIFT/X charge, P pause Ship it tested: syntax-checked and headless-simulated through a full run.

Фото профиля ᴍᴀɴɴ | Dungeonpunk Online
ᴍᴀɴɴ | Dungeonpunk Online4 месяцев назад

Very cool. Fable really is in a class of it's own, the scene composition seems really polished, but each had some interesting surprises.

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

Fable version really surprised me

Фото профиля ᴍᴀɴɴ | Dungeonpunk Online
ᴍᴀɴɴ | Dungeonpunk Online4 месяцев назад

The amount of juice in the movement and sound is really what got me. It was using little pops of movement to drive attention and discovery. Stuff that I consider next level for most devs.

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@Zai_org @ChatGPTapp @GeminiApp Play the Claude Fable 5 version

Фото профиля ESchwaa
ESchwaa4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp And some say they don’t notice any difference with Fable

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp Pretty clear here right?

Фото профиля Haider.
Haider.4 месяцев назад

glm-5.2 did so good

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

Yeah, I was surprised

Фото профиля Unperceivable Truth
Unperceivable Truth4 месяцев назад

@haider1 Were these all one shot? Or did you have to nurse any along? It looks like Fable 5, glm 5.2, flash 3.5 and then gpt5.5 ranked from best to worse.

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@haider1 All one shot

Фото профиля Jules ⭐
Jules ⭐4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp Including Flash 3.5 in with the others was just evil... until I saw the results! It's not the best here, but it's easily 10x better than my expectations. Maybe there is hope for 3.5 Pro after all.

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp I'm trying to motivate @GoogleAI 😆

Фото профиля Chong-U
Chong-U4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp 1. Ecco the Dolphin 2. Ecco the Dolphin x Another World Remix Edition 3. Slecco the Swirlfin 4. Ecco the Salamander

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp Great minds think alike 😂

Фото профиля AI Mastery Guide
AI Mastery Guide4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp Same prompt, four models, completely different feel for each game, love this format

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp Cheers 🍻

Фото профиля AI Mastery Guide
AI Mastery Guide4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp nice, cheersss

Фото профиля Adel Bucetta
Adel Bucetta4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp that's the most interesting part each model's procedural generation is so different even when given the same task, it highlights the variance in their internal representations

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp It's a good benchmark to see actual differences

Фото профиля Philippe Tremblay
Philippe Tremblay4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp Could you show Opus?

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp

Фото профиля AiRivers.com | For Sale
AiRivers.com | For Sale4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp Ecco the Dolphin is the perfect nostalgic theme to test these new models. In my opinion, Fable did it best but GLM really surprised me.

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@openingai_com @claudeai @Zai_org @ChatGPTapp @GeminiApp Agree on both points

Фото профиля @ChuckBaggett Chuck Baggett Peace Freedom Ad Astra
@ChuckBaggett Chuck Baggett Peace Freedom Ad Astra4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp GLM 5.2 looked the best to me.

Фото профиля Damian Dąbrowski
Damian Dąbrowski4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp Would be awesome to add Opus 4.8 to comparison :)

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp Opus 4.8 did well on logic, but Fable 5 graphics were much better

Фото профиля Philippe Tremblay
Philippe Tremblay4 месяцев назад

@dabrodev @claudeai @Zai_org @ChatGPTapp @GeminiApp Who wins between GLM 5.2 and Opus 4.8 in your opinion?

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@dabrodev @claudeai @Zai_org @ChatGPTapp @GeminiApp Opus 4.8 slightly, but GLM is impressive

Фото профиля anki
anki4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp Fable wins

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp Agree 💯

Фото профиля Chong-U
Chong-U4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp Could i ask how you’re using glm 5.2? Which harness and provider ?

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp Directly on

Фото профиля Philippe Tremblay
Philippe Tremblay4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp it's between Fable and GLM 5.2. With Fable being better on some aspects, and vice-versa.

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp Fable did the cleanest implementation with best graphics imo

Фото профиля Arthur.eth
Arthur.eth4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp 这个对比真的很有说服力

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp 🙏

Фото профиля Rangga Suntya🔻
Rangga Suntya🔻4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp GPT 5.5 why the fish body look like wave

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp Not sure, I tried a few generations and it came out like that.

Фото профиля Kakashi
Kakashi4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp Can I know how much did each cost too ?

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@Meme_God_069 @claudeai @Zai_org @ChatGPTapp @GeminiApp For Claude and ChatGPT I used the subscriptions and official chat interfaces, GLM and Gemini were free.

Фото профиля YEEHAW Games
YEEHAW Games4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp Did you /loop it or goal it haha

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp Ha nope! It's actually just a single prompt, check the replies

Фото профиля YEEHAW Games
YEEHAW Games4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp The one thing I have to say, the UI is always so bad 😭😭 no offense to you, I know it was one shotted, but AI UI is so noticeable now haha

Фото профиля Joshua Riley
Joshua Riley4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp Fable 5 actually looks incredible, still won’t switch back from Codex though, it’s alright for me right now

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp I like Codex for functional code, but I've found Claude superior for game dev

Фото профиля Joshua Stanton
Joshua Stanton4 месяцев назад

@threejs @claudeai @Zai_org @ChatGPTapp @GeminiApp Use 5.5 in codex harness not 5.5 pro

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@ShiftingPathway @threejs @claudeai @Zai_org @ChatGPTapp @GeminiApp Tried it in Codex, got a better dolphin, but the level design still very similar to Pro

Фото профиля Duc Truong
Duc Truong4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp That Gemini is what nightmares are made of

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp 😂 hope 3.5 pro is better! @OfficialLoganK

Фото профиля Myron AI artist
Myron AI artist4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp You should have a warning on there "CANNOT BE UNSEEN"

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp 😂

Фото профиля Zach
Zach4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp Fable looks really good and I also like GLMs

Фото профиля Ali Shirani
Ali Shirani4 месяцев назад

@claudeai @Zai_org @ChatGPTapp @GeminiApp It reminded me of an old game called "Feeding Frenzy." Nice work, Majid.

Фото профиля Majid Manzarpour
Majid Manzarpour4 месяцев назад

@Zai_org @ChatGPTapp @GeminiApp Play the GLM 5.2 version

Похожие видео

GPT-5.6 vs GPT-5.5 on my custom spaceship prompt. I gave both models the exact same custom prompt. This is also the same prompt I previously gave to Fable 5. For context, GPT-5.6 Pro worked for 87 minutes, while GPT-5.5 Extra High worked for 34 minutes and 42 seconds. As I’ve said before, based on great authority GPT-5.6 will be an incremental/soldi improvement over GPT-5.5, not a “Fable killer.” My rough expectation has been that it would trade blows with Fable 5 on some benchmarks, maybe win around half depending on the category, but not clearly surpass it overall. And again fable five will have bigger model smell, but this was expected. After testing this coding output, that view feels pretty accurate. GPT-5.6 is clearly better than GPT-5.5 in several visual areas. The lighting, shading, chairs, object details, and exterior of the spaceship looked noticeably stronger. The scene was also easier to test. I do want to give GPT-5.5 credit though. It built out the rooms much much better and the planets looked better than GPT-5.6’s. It was also interesting that both GPT-5.5 and GPT-5.6 produced better-looking planets than Fable 5 in this specific test. The downside with GPT-5.5 was stability. The game was much glitchier and harder to test compared to GPT-5.6. But when it comes to the core of the demo, which is the spaceship itself, Fable 5 still beat both models pretty comfortably. GPT-5.6 is impressive, but from this test, it looks exactly like what I expected which was a meaningful incremental improvement over GPT-5.5, at least for indie game demos, but not something that replaces Fable 5. In collaboration with Chetaslua

Chris

250,919 просмотров • 5 месяцев назад