Roman AI

AI & tech insights | Content Creator | Sharing AI development & free resources | Dm for Collaborations 💌

[Roman AI @Aiwithroman]

Made with Seedance 2.5 in 1080p. 🎬🔥

Cinematic AI is reaching a whole new level.

PROMPT ⬇

[Roman AI @Aiwithroman]

This is the kind of benchmark that exposes the gap between “the agent can code” and “the agent can actually own a software migration.”

Those are two very different claims.

[Roman AI @Aiwithroman]

The native audio is what caught me here.

30 seconds.
Visuals + sound.
Up to 20 reference assets.

All in one generation.

That changes what you can actually build in one pass.

Worth testing → https://t.co/dDce6mZM1t

[Nawi @Aria_Nawi]

Every "revolutionary" AI model launch comes with the same playbook: hype thread, limited-time discount, countdown timer creating fake urgency.

WAN 3.0 launched today on the Pika API Club and skipped all of that. No expiring deal. It's just less expensive there, permanently.
30s native generation, up to 20 reference assets, and editing precision that actually holds up....worth testing yourself before the price talk becomes the whole conversation.

Get started here https://t.co/jQ7nO33T4t

[Roman AI @Aiwithroman]

@Aria_Nawi The hard part isn’t the first 90% of the migration. It’s keeping the agent coherent when the task spans hundreds of decisions.

[Nawi @Aria_Nawi]

Every "revolutionary" AI model launch comes with the same playbook: hype thread, limited-time discount, countdown timer creating fake urgency.

WAN 3.0 launched today on the Pika API Club and skipped all of that. No expiring deal. It's just less expensive there, permanently.
30s native generation, up to 20 reference assets, and editing precision that actually holds up....worth testing yourself before the price talk becomes the whole conversation.

Get started here https://t.co/jQ7nO33T4t

[Roman AI @Aiwithroman]

This is the kind of benchmark that exposes the gap between “the agent can code” and “the agent can actually own a software migration.”

Those are two very different claims.

[Einsia @EinsiaAI]

The ultimate test for coding agents isn't local editing—
it's whole-repo evolution, and right now, the survival rate is 5.4%.

Today we’re releasing SWE Refactor Bench, a benchmark for long-horizon, whole-repository software stack migration.

Coding agents are getting very good at fixing bugs.
But can they refactor an entire system, C → Rust, Maven → Gradle, POSIX → WebAssembly?

We built 20 real migrations across projects, including SQLite, zlib, libsodium, and GraphHopper.

520 runs. Only 28 survived all 3 stages. 13/20 tasks were solved by nobody.

System-scale migration is still wide open.

Full breakdown 👇
GitHub: [https://t.co/xXyLQ3qq0C]
Paper Link: [https://t.co/lO1Enh63q7]
Einsia Website: [https://t.co/AGn0hF5gwL]

[Roman AI @Aiwithroman]

@EinsiaAI This is where the gap between a good coding demo and production-grade autonomy becomes very obvious.

[Einsia @EinsiaAI]

The ultimate test for coding agents isn't local editing—
it's whole-repo evolution, and right now, the survival rate is 5.4%.

Today we’re releasing SWE Refactor Bench, a benchmark for long-horizon, whole-repository software stack migration.

Coding agents are getting very good at fixing bugs.
But can they refactor an entire system, C → Rust, Maven → Gradle, POSIX → WebAssembly?

We built 20 real migrations across projects, including SQLite, zlib, libsodium, and GraphHopper.

520 runs. Only 28 survived all 3 stages. 13/20 tasks were solved by nobody.

System-scale migration is still wide open.

Full breakdown 👇
GitHub: [https://t.co/xXyLQ3qq0C]
Paper Link: [https://t.co/lO1Enh63q7]
Einsia Website: [https://t.co/AGn0hF5gwL]

[Roman AI @Aiwithroman]

@AiwithNomi Good

[Roman AI @Aiwithroman]

@AiwithShoaib Next level

[Shoaib AI @AiwithShoaib]

The agent has ~31 days of tokens left, no revenue model, and somehow ends up cold-emailing Toby Ord by itself.

That’s a very strange sentence to write about an AI system.

Is it:

a) an agent in trouble
b) an agent doing growth
c) an agent treating survival and growth as the same problem

I lean c.

[Shoaib AI @AiwithShoaib]

AI agents don’t always need more intelligence.

Sometimes they just need the right skills.

Aident Loadout comes with 400+ expert-built skills for research, outreach, content & ops — so your agent can start working with real capabilities from day one. 🔥

[Roman AI @Aiwithroman]

@AiwithLucas_ Amazing

[Lucas AI @AiwithLucas_]

Really interesting to see how this is evolving. AI is making it easier for creators and businesses to experiment, build faster, and discover new opportunities without needing complicated workflows or technical skills.

[Roman AI @Aiwithroman]

@AiwithFatima Great setup

[Fatima AI @AiwithFatima]

The real innovation in Ornith-1.5 isn’t just its benchmark scores.

It’s the idea of a model continuously creating its own learning experiences.

Generate tasks → build scaffolds → solve → learn → repeat.

If this approach scales, open-source models could get much better at improving themselves. 🔥

[Roman AI @Aiwithroman]

@AiwithShoaib This is very interesting 🤔

[Shoaib AI @AiwithShoaib]

Open-source models are entering a different era.

Ornith-1.5 isn’t just scaling parameters — it’s scaling how models learn.

A self-improving loop where the model creates tasks, builds scaffolds, generates solutions, and learns from them through RL.

That’s a very interesting direction for the future of AI. 🔥

[Roman AI @Aiwithroman]

@AiwithLucas_ @Lart_AI Great 👍

[Lucas AI @AiwithLucas_]

No animators. No studio. Just a prompt.

This entire cinematic scene was generated with AI — here's the tool I used 👇

Made with @Lart_AI On Mini Max H3
https://t.co/TMvIIBTtDN

Prompt 👇

New Character & Story Anime cinematic aesthetic, soft hand-painted backgrounds, expressive character animation, and a nostalgic melancholic atmosphere. A lanky old golden retriever with a graying muzzle and a frayed red bandana walks beside a boy in a faded blue school uniform, backpack slung over one shoulder, along a quiet coastal cliffside path at dawn. Begin with a medium tracking shot from behind as the boy and dog walk together, mist curling low over the grass and the dog's paws leaving faint prints in the dew. Cut to intimate ground-level close-ups of the dog pausing at the cliff edge, nose lifted to the sea breeze, ears twitching as gulls cry in the distance. The camera slowly rises past the dog before tilting to reveal the boy staring quietly out at the horizon, his uniform jacket rippling in the wind. Tall sea grass bends and sways around them, catching the first pale light of sunrise. Transition into a smooth cinematic pull-back as the boy kneels down and the dog rests its head against his shoulder. Soft dawn light breaks through low clouds, creating hazy god-rays, gentle lens flares, and a faint golden shimmer across the ocean mist. End with a breathtaking wide-angle shot from behind: the boy and dog appear small atop the cliff, framed by an endless pale-blue sea, distant fishing boats, and a soft peach-and-lavender sunrise sky. The wet grass glistens with dew like scattered light, creating a wistful, bittersweet anime-film ending — a quiet goodbye rendered in warmth rather than sorrow. Highly detailed backgrounds, natural movement, cinematic depth of field, subtle film grain, atmospheric perspective, soft rim lighting, delicate mist and cloud shading, smooth camera motion, emotionally expressive storytelling, dawn color palette, Makoto Shinkai-inspired sense of quiet wonder.

[Roman AI @Aiwithroman]

@AiwithLucas_ @Lart_AI Outstanding