FLUX.2 Is Now Open Source, But I Also See the Struggles of Small Companies

FLUX.2 Is Now Open Source, But I Also See the Struggles of Small Companies

FLUX.2 Goes Open-Source — A Deep Dive & Reality Check

Yesterday marked an interesting event in the AI image generation world — FLUX.2 officially released part of its technology as open source.

image

From King of AI Art to Relative Silence

FLUX was once considered the reigning champion of AI image generation, almost replacing the original Stable Diffusion ecosystem, and becoming a mainstream base model.

But times have changed — Nano Banana Pro has taken the spotlight, leaving FLUX with much less buzz. It’s a bit tragic, really.

Still, given it’s one of the few sparks left in AI image creation, it’s worth discussing.

---

What's in the Release?

They’ve released:

  • 4 Base Models
  • Pro (closed source) — most powerful
  • Flex (closed source) — most powerful
  • Dev (open source) — already available
  • Klein (distilled, open source soon)
  • 1 VAE model
image

Open-Source 'Dev' Model

image

🔗 Open-source link:

https://huggingface.co/black-forest-labs/FLUX.2-dev

Soon-to-be Open Source 'Klein'

This distilled model will be open-sourced soon.

Once integrated with Liblib, these models should work there. You can also try them on FLUX’s official site (VPN required):

https://playground.bfl.ai/

image

---

Prompt Testing — FLUX.2 vs Nano Banana Pro

Using identical prompts to compare results:

---

Test 1: Archaeologist Prompt

Prompt: A human archaeologist at a pyramid excavation site discovers a rotating metal sphere, shot handheld in documentary realism style.

image

---

Test 2: Casual iPhone Street Photo

Prompt: Snapshot of an Asian woman in a relaxed outfit, sweater + wide-leg jeans, standing casually on the street or café entrance, soft natural light.

image

---

Test 3: Anime Dragon Scene

Prompt: Shenron from Dragon Ball Z gazing at Dragon Balls, colorful pencil style, dynamic composition.

image

---

Test 4: Chinese Artistic Poster

Prompt: Tianshu Qitan poster in gongbi Chinese landscape style, clear readable text.

image

---

Prompt-Based Editing

Example: Cosplay Transformation

Original image:

image

Prompt: Make left character cosplay the one on the right.

Nano Banana Pro result:

image

FLUX.2 result:

image

Nano’s output matches perfectly; FLUX’s misses the mark entirely.

---

Example: Convert Anime to Real Human

Original:

image

Results:

image

---

World Knowledge — Where the Gap Widens

Behind Nano Banana Pro: Gemini 3 Pro — a top-tier multimodal model.

Behind Flux.2: Mistral-3 24B.

image

---

Example: One Piece Battle Power Infographic

Nano Banana Pro:

image

Shows rich understanding of One Piece.

Flux.2:

image

Appears to lack knowledge entirely.

---

Karl’s Famous Prompt — Group of Anime & Cartoon Characters

Nano Banana Pro result:

image

Nearly perfect in identifying all characters.

Book Page Generation Prompt:

Nano Banana Pro:

image
image

Flux.2:

image

Only produced garbled text.

---

What This Reveals

Flux.2’s struggles highlight a major trend:

Once big tech commits full resources, smaller AI labs fall behind quickly.

Image generation is no longer just visual skill — it’s about world models:

  • Knowing anime protagonists and cultural icons
  • Understanding cross-media references
  • Integrating text, video, code, and dialogue sources via super multimodal models

---

Resources Drive the Gap

  • Data
  • Compute Power
  • Talent Density
  • Money — lots of it

Flux’s images may feel clumsy compared to Nano’s, but instead of ridicule, they inspire regret — and respect for the open-source gesture.

For SMEs and indie developers, Flux.2’s release offers a foundation to build:

  • Train custom models
  • Experiment with workflows
  • Explore creative possibilities

In that sense, they are heroes — with a touch of Don Quixote tragedy.

---

The Hard Truth for Entrepreneurs

The AI era is:

  • Best of times — democratized access to tech
  • Worst of times — deep waters ruled by resource competition

The ideals of innovation often hit the hard wall of resource inequality. Yet game-changing breakthroughs can grow quietly beneath it.

This is harder than the Internet or mobile Internet eras — and as an entrepreneur, I deeply relate.

---

A Note for Creators — Platforms Like AiToEarn

Open-source ecosystems like AiToEarn官网 can help bridge gaps:

  • AI content generation
  • Seamless multi-platform publishing
  • Monetization tools
  • Ranking insights via AI模型排名
  • Platforms supported: Douyin, Kwai, WeChat, Bilibili, Xiaohongshu, Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, X (Twitter)

Such integrated infrastructure could be the missing link for indie creators leveraging models like FLUX.2 or Nano Banana Pro.

---

Final Thoughts

I wish for all entrepreneurs — myself included — to be brave pioneers of our time,

to endure storm and tide,

and to reach the farthest shore.

May we encourage each other.

---

If you’ve read this far and enjoyed it, please like, share, or save. Want updates? Add a ⭐ — thanks for reading, and see you next time.

Read more

Translate the following blog post title into English, concise and natural. Return plain text only without quotes. 哈佛大学 R 编程课程介绍

Harvard CS50: Introduction to Programming with R Harvard University offers exceptional beginner-friendly computer science courses. We’re excited to announce the release of Harvard CS50’s Introduction to Programming in R, a powerful language widely used for statistical computing, data science, and graphics. This course was developed by Carter Zenke.