Jiaxiang Liu

Homepage： Of ERNIE-Image

[Lex Fridman @lexfridman]

I'm traveling the world for a bit, starting with China but then hopping around the globe, anywhere. Open to any adventure. No plans, only a backpack. Hoping to meet & get to know humans from all walks of life. The pic is from a long hike on the Great Wall. For me, as a fan of history, this was an epic experience.

In China, first I'm visiting a few big cities & talking to engineers at the heart of China's AI revolution. After that, if feeling crazy enough, I'm hitchhiking (first time) across rural China for a few weeks. Hitchhiking because I think it's the best way to meet rural folks who I would otherwise never get the chance to meet. I hope to do the same in US and other places.

I have a request, if you have a travel recommendation, fill out the form(s) below if you feel like it. Or share with folks who might have advice about such travel.

Form 1 - travel recommendation:
If you can, recommend to me an interesting place I should visit anywhere in the world. For this, fill out form 1. Not touristy stuff, but something off the beaten path, that tourists may not know about, but is legendary. It could be as remote as meeting a herder in the mountains who is a local legend. Asia, Middle East, Europe, India, South/North America, Africa, Australia, anywhere. In China, I'm hoping to visit maybe Heibei, Shanxi, Shaanxi, Gansu, Sichuan, Yunnan, etc, so recommendations for spots to visit are helpful.

Form 2 - coffee:
If you want to grab a coffee with me anywhere in the world, fill out form 2 (please don't use form 1 for that).

Anyway, I hectically tossed stuff in backpack. Realizing I don't have a clear plan of any kind, which is probably the only way to do it. LFG.

Love you all ❤️

[Jiaxiang Liu @lclbrew]

Smaller models catching up to frontier performance is one of the most interesting trends in AI right now.

The direction is clear: the gap between “affordable to train” and “capable enough to use” is narrowing fast.

Hopefully this becomes the norm. Powerful models shouldn’t require frontier-scale budgets

[ERNIE for Developers @ErnieforDevs]

ERNIE 5.1 is here 🚀

ERNIE 5.1 significantly reduces pretraining cost while compressing total parameters to ~1/3 and activated parameters to ~1/2 — using only ~6% of the pretraining cost compared to models at similar scale, while achieving leading performance in its class.

💡Key highlights:

1/ Strong agentic performance approaching leading frontier models. ERNIE 5.1 surpasses DeepSeek-V4-Pro on both τ3-bench and SpreadsheetBench-Verified.

2/ Strong world knowledge and creative writing capabilities, with GPQA and MMLU-Pro performance approaching leading closed-source models, and creative writing ability nearing Gemini 3.1 Pro.

3/ Frontier-level reasoning performance. ERNIE 5.1 scores 99.6 on the challenging AIME26 benchmark with tools, second only to Gemini 3.1 Pro.

4/ Deep search capability. On May 9, ERNIE 5.1 ranked #4 globally and #1 among Chinese models on the Arena Search leaderboard with a score of 1223.

ERNIE 5.1 is now available on ERNIE and the Baidu AI Studio Model Playground:
👉https://t.co/qhd67Lg3B4
👉https://t.co/AaQSqDmVGU
👉https://t.co/uCNiypIu1q

[Jiaxiang Liu @lclbrew]

Congrats to the ERNIE team on a great milestone! 🎉
Efficiency isn’t just a metric，it’s a philosophy. Better results per unit of compute means AI that’s more accessible, more sustainable, and more impactful. That’s a win for everyone. Looking forward to what comes next. 👀

#ERNIE #AI

[ERNIE for Developers @ErnieforDevs]

Introducing ERNIE 5.1 Preview — now live! 🚀
Ranked #13 globally and #1 among Chinese labs on @arena 's Text Arena.
Top-10 worldwide across:
📐 Math — #9
⚖️ Legal & Government — #1
💼 Business, Management & Financial Ops — #4
💻 Software & IT Services — #7

Built on the strong pre-training foundation of ERNIE 5.0, ERNIE 5.1 Preview compresses total parameters to ~1/3 and activated parameters to ~1/2 of its predecessor, while using only ~6% of the pre-training cost of comparable models — delivering leading foundational performance at its scale.

Try it now 👉 https://t.co/jNwEbwrKps
More new models are on the way — stay tuned 👀

[ERNIE for Developers @ErnieforDevs]

Introducing ERNIE 5.1 Preview — now live! 🚀
Ranked #13 globally and #1 among Chinese labs on @arena 's Text Arena.
Top-10 worldwide across:
📐 Math — #9
⚖️ Legal & Government — #1
💼 Business, Management & Financial Ops — #4
💻 Software & IT Services — #7

Built on the strong pre-training foundation of ERNIE 5.0, ERNIE 5.1 Preview compresses total parameters to ~1/3 and activated parameters to ~1/2 of its predecessor, while using only ~6% of the pre-training cost of comparable models — delivering leading foundational performance at its scale.

Try it now 👉 https://t.co/jNwEbwrKps
More new models are on the way — stay tuned 👀

[Baidu Inc. @Baidu_Inc]

It's easy to forget how small we are in the grand scheme of things.

This Earth Day, we're taking a moment to appreciate our planet and the vast ecosystem we share. 🌍

- Images created with ERNIE-Image

[Max For AI @MaxForAI]

真的没人发现百度Ernie-image这个宝藏模型吗？？

都在说GPT-image-2
但我强烈想安利大家看下
目前同参数最强开源生图模型
Erine-image👀

8B的参数连长文本解说和
复杂网页UI都能直接完美输出？！

之前我折腾各种开源模型
目前我实测跑下来
ERNIE-Image绝对是当之无愧
NanoBanana开源最强平替

特别是对于文字的生成能力
无论是中文、英文还是多语言混合
它都能实现高可读性的高保真生成
我在实测里让它画UI界面、多图文排版的海报
文字清晰可见
拼写完全正确

而且指令遵循能力很好
丢给它一段复杂的Markdown或者长文本
它能给你排出规规矩矩的动漫分镜、多格网格图（但Prompt要写很细

而且这次百度真的非常良心
开源了两个版本
大参数版本ERNIE-Image得益于紧凑的 8B 参数规模
24G显存的消费级显卡就能完美运行（我的4090稳吃😋

而且官方还出了个ERNIE-Image Turbo（蒸馏版
只要8步就能出图
速度快到飞起
微调和二次开发的门槛大幅降低（我觉得适合NSFW

目前ERNIE-Image的模型权重和推理代码
已经在Hugging Face和魔搭全面开源（非常厚道的Apache 2.0协议
ComfyUI 工作流也已经就位！
强烈建议所有玩AI绘画的朋友赶紧去官方的星河社区或者Hugging Face/魔搭下个权重跑跑看，绝对会惊艳到你！👇

[DeepSeek @deepseek_ai]

🚀 DeepSeek-V4 Preview is officially live & open-sourced! Welcome to the era of cost-effective 1M context length.

🔹 DeepSeek-V4-Pro: 1.6T total / 49B active params. Performance rivaling the world's top closed-source models.
🔹 DeepSeek-V4-Flash: 284B total / 13B active params. Your fast, efficient, and economical choice.

Try it now at https://t.co/GCdiMzk1Dl via Expert Mode / Instant Mode. API is updated & available today!

📄 Tech Report: https://t.co/drlDrxkYtp
🤗 Open Weights: https://t.co/T13Y8i7SDM

1/n

[Jiaxiang Liu @lclbrew]

@ShunyuYao12 great work

[Zain Shah @zan2434]

Imagine every pixel on your screen, streamed live directly from a model. No HTML, no layout engine, no code. Just exactly what you want to see.

@eddiejiao_obj, @drewocarr and I built a prototype to see how this could actually work, and set out to make it real. We're calling it Flipbook. (1/5)

[Chidanand Tripathi @thetripathi58]

Okay... I spent the morning testing Baidu's new open-source model, ERNIE-Image.

Usually, getting AI to spell text correctly on a mockup or a diagram takes hours of trial and error. You ask for a label, and it gives you random shapes instead of letters.

This one actually got it all right.

I ran two tests: a bilingual product shot and a 3D cutaway diagram.

Check out my results: ↓

[Design Arena @Designarena]

BREAKING: ERNIE Image & ERNIE Image Turbo by @Baidu_Inc are #5 and #8 out of open weight models on Image Arena!

They have Elos of 1154 and 1131, and are in the same performance bands as GLM-Image and Qwen Image, respectively

Congrats to the @ErnieforDevs team on the launch!

[Jiaxiang Liu @lclbrew]

@HaareBlond excellent idea. we were actually going to use it in the next version

[Jiaxiang Liu @lclbrew]

1/ we are excited to release ERNIE-Image, after 3 months of building from scratch.
an 8b text-to-image model from baidu's ernie image team. honestly, we didn't expect an 8b dit to get this far, this fast.
strong instruction following. best-in-class text rendering. runs on a 24gb gpu.
huge thanks to the ERNIE-Image team, this wouldn't exist without an incredibly talented group of people who shipped fast and cared deeply.
thread below. 👇 👇

[Jiaxiang Liu @lclbrew]

@HaareBlond We’d already finished pretraining by the time this technology came out.

[Jiaxiang Liu @lclbrew]

@RongshengWang great

[Jiaxiang Liu @lclbrew]

@EtherCoins Thanks！

[Jiaxiang Liu @lclbrew]

9/ full release: ERNIE-Image, and ERNIE-Image-Turbo (8-step distilled) + prompt enhancer + comfyui workflow + sglang version + unsloth/gguf version .
all of it. today.
we're excited to see what you build.
— ERNIE Image Team

Blog：https://t.co/l5oGYzSMM9
Demo1: https://t.co/3T5FkoAReL
Demo2: https://t.co/gdrV0VRc0A
Hugging Face：
https://t.co/RVPyjg7fPz
https://t.co/Y2vixULdwd
Github：https://t.co/bCNTM0MNM3
Art Gallery : https://t.co/A18iDorLS7

[Jiaxiang Liu @lclbrew]

8/ one more thing, and this matters.
ernie-image works best with long, detailed, well-structured prompts. in practice, users type a short sentence.

so we built a fix: a 3b prompt enhancer, released alongside the model. it expands short inputs into detailed, structured prompts without changing what you asked for.

take a look at the left two panels first. same prompt, same image model, very different outputs depending on pe quality. the difference is telling. that's a practical lever we're explicitly handing to the community, and why we open-sourced the pe instead of keeping it internal.

now look at the right two panels. stronger llms push this even further, prompt enhancement scales with the quality of the enhancer. we're curious to see how far the community can take it.