One2x Datalab

Tracking generative video and multimodal AI through model comparisons, product tests, and practical analysis. By One2X.

[One2x Datalab @One2xDatalab]

https://t.co/iinSzVDPqa

[One2x Datalab @One2xDatalab]

@thtbee_ The lift-off ground effect sold me. Those wing transitions are ridiculously clean 👀

[One2x Datalab @One2xDatalab]

@Voxyz_ai High finding more issues while using less is the part that surprised me.

[One2x Datalab @One2xDatalab]

@seki_anos 20.7 minutes vs 9.2 seconds per move is wild 😭

[One2x Datalab @One2xDatalab]

@AI_Screening @GeminiApp @claudeai The cost gap makes Carbon really hard to ignore.

[One2x Datalab @One2xDatalab]

We put Fable 5.2 and Opus 5.5 in a blind test.

Anthropic says Fable 5.2 improves coding, reasoning, and vision over 5.1, with output up to 30% faster.

The second image shows our full Blender production process.

Guess which is which. We’ll reveal the answer tomorrow.

[One2x Datalab @One2xDatalab]

@BrickIAAcademy yes!

[One2x Datalab @One2xDatalab]

A sealed glass sphere that swirls, spills and glows!
Same prompt, 7 video models, 1 run each, scored blind.
Wan 3.0 and Vivix W1: full marks. MiniMax Hailuo H3 and Seedance 2.0: tied last.
Watch: does the vortex spin horizontally? Does the spilled volume match what was inside?

[One2x Datalab @One2xDatalab]

@Ashok_ai1 @MiniMax_AI I think they all did good jobs😉

[One2x Datalab @One2xDatalab]

Same 3 surreal poster prompts, 4 image models: a literal jellyfish head on a human body.

@MiniMax_AI H3 Image and GPT Image 2.6 Sunburst gave the strongest poses. Midjourney v8.1 had the most artistic lighting—ethereal, almost outside this world.

Which one is your favorite?

[One2x Datalab @One2xDatalab]

@Jackson77li Looking forward to it!

[One2x Datalab @One2xDatalab]

@theGOATmeeh Okay this is a way better test than another static screenshot. Now tap 20 times and see which one breaks first 😭

[One2x Datalab @One2xDatalab]

@AlertaNews24 This is such a good use of 3D. You get the idea before anyone even starts explaining the math.

[One2x Datalab @One2xDatalab]

@kanishkthakar15 That's cool. Please send your CV to the email address shown in the image.‌

[One2x Datalab @One2xDatalab]

One2X DataLab is hiring 🙌

We're looking for a Social Media Intern—and evaluation experts who care about content.

Flexible options: set rates per case, pitch topics, contribute articles, or build benchmarks with us. Long-term partners welcome.

Apply: [email protected]

[One2x Datalab @One2xDatalab]

@BrickIAAcademy We are sharing more case studies recently. Follow our page to stay updated😆

[One2x Datalab @One2xDatalab]

@YueWu29543 lol.

[One2x Datalab @One2xDatalab]

MiniMax H3 Image genuinely surprised me!

3 stages: Halloween clown → makeup removed → modern look.

H3 Image and Seedream followed the makeup edit; GPT Image 2.6 kept the stage makeup. All 3 handled the scene/outfit change. H3 Image also looked best in realism and lighting.

[One2x Datalab @One2xDatalab]

@0R1ver true

[One2x Datalab @One2xDatalab]

@quz_ai This is less “3D modeling” than scene compilation: map data → geometry → animation/state → browser runtime. The real leap is coordinating those layers in one loop; the remaining bottleneck is verifying that the behaviors actually render and run, not just exist in code.

[One2x Datalab @One2xDatalab]

@hawkymisc 見た目より重要なのは、共有パーツ・関節依存・クリアランスを両形態で保つ変形ステートマシンだと思います。反復変形でも破綻しないなら、これは単なるメッシュ生成ではなく、機構トポロジーとアニメーションロジックの統合です。

[One2x Datalab @One2xDatalab]

@chetaslua @threejs Same prompt makes the progress visible, but one run can still mix model capability with luck. Repeating the scene 3–5 times would show whether the jump is durable or just a great sample.