Lesson 84 of 2116
Flux Dev — open-source fine-tuning
Flux Dev is the LoRA-friendly middle tier of the Flux family. Here is how to train a style on your own art without renting a farm.
Lesson map
What this lesson covers
Learning path
The main moves in order
- 1Why Dev exists
- 2Training a style LoRA
- 3Suggested workflow
- 4When not to fine-tune
Concept cluster
Terms to connect while reading
Section 1
Why Dev exists
Flux Schnell is too distilled for stable fine-tuning. Flux Pro is closed. Flux Dev is the sweet spot: open weights good enough to train a LoRA on in a few hours on a single 4090 or A100.
Section 2
Training a style LoRA
- 1Gather 20-50 consistent-style reference images
- 2Write captions that describe what varies (subject) and bake in what stays constant (style) as a trigger word
- 3Train with ai-toolkit or kohya_ss for ~1500 steps
- 4Evaluate with held-out prompts before shipping
Compare the options
| Training target | Flux Dev LoRA | SDXL LoRA | Midjourney style ref |
|---|---|---|---|
| Training cost | ~$5-15 GPU | ~$2-10 GPU | Free but limited |
| Quality ceiling | High | Medium | Medium |
| Commercial use | Non-commercial (Dev) | OK | Per MJ TOS |
| Control | High | High | Limited |
The license trap
Section 3
Suggested workflow
- Train on Flux Dev locally or via Replicate's trainer
- Iterate prompts until the LoRA holds style consistently
- For production renders, apply the same prompts to Flux Pro (note: Pro does not accept your LoRA — use it as a style reference via BFL's fine-tune endpoints instead)
- Keep your trigger word consistent so collaborators can reproduce results
A minimal LoRA training config. Budget 2-4 hours on a 4090.
# ai-toolkit flux config
model: black-forest-labs/flux-dev
steps: 1500
lr: 1e-4
rank: 32
trigger: mystyle_v1Section 4
When not to fine-tune
If you only need 10 images in a consistent style, prompt engineering plus reference images is faster and cheaper than training. Fine-tune when you need hundreds of renders with identical aesthetic DNA.
Key terms in this lesson
End-of-lesson quiz
Check what stuck
15 questions · Score saves to your progress.
Tutor
Curious about “Flux Dev — open-source fine-tuning”?
Ask anything about this lesson. I’ll answer using just what you’re reading — short, friendly, grounded.
Progress saved locally in this browser. Sign in to sync across devices.
Related lessons
Keep going
Creators · 32 min
Kimi Research Mode — autonomous deep research
Kimi's Research Mode plans, browses, and synthesizes across dozens of sources. Here is how to get the most out of it.
Creators · 36 min
Flux Schnell vs. Flux Pro
Black Forest Labs offers three Flux tiers. Schnell is free-speed, Pro is the paid flagship. Here is when each wins.
Creators · 38 min
Midjourney niji — anime mode
Niji is Midjourney's anime-specialist model. Here is how to prompt it and when it beats general Midjourney for stylized art.