A quick guide in the sequences/orders of using LoRA

Hello everyone, in this guide I will give a brief explanation of how to use LoRA, especially in terms of the sequence/order of placement.

Question: I want to generate a text2image image. In addition to using the Base Model, I will also use several LoRAs. The question is, will the order of the LoRAs affect the image? For example: the first LoRA is Extra Detailer, the second LoRA is Enhanced Lightning, and the third LoRA is Realistic Skin.

The short answer: yes, the order of LoRA can have an effect, but it's not always dramatic. Think of LoRA as a transparent layer layered on top of a base painting. The order determines who gets the last word in the overlapping areas. 🎨

Let's analyze it carefully.

1️⃣ How LoRA works (brief but precise)

Most LoRA models work by adding biases or weight modifications to the base model on the same layer.

If two LoRA models touch the same area (e.g., facial details, lighting, or skin texture), then:

LoRAs loaded later tend to be more dominant
Especially if the weight is similar or quite high

If LoRA touches different areas, the sequence is barely noticeable.

2️⃣ Analyze the LoRA example you are asking about

Order:

Extra Detailer
Enhanced Lightning
Realistic Skin

What happened?

Extra Detailer
Adding micro-details, textures, and sharpness. This is global in nature.
Enhanced Lightning
Change contrast, highlights, shadows, sometimes color tones.
Realistic Skin (last)
Usually focuses on the face and skin: pores, softness, color blending.

📌 Since Realistic Skin was loaded last:

He can override the rough detail effects of Extra Detailer.
The skin becomes more natural, not too "crispy"
Lighting is still felt, but adjusted to the skin

➡️ This is a pretty reasonable order for realistic results.

3️⃣ Example if the order is reversed

For example:

Realistic Skin
Extra Detailer
Enhanced Lightning

The results can be:

The skin is initially realistic
Then “sharpened again” by Extra Detailer → pores can be too strong
The final lighting can make the skin look plastic or over-contrast.

It's not wrong, but the characters are different.

4️⃣ Rules of thumb (that practitioners actually use)

Use this principle:

Global LoRA first → specific LoRA later

Ideal order:

🔹 Style / Detail / Texture LoRA
🔹 Lighting / Mood / Color LoRA
🔹 Anatomy / Face / Skin / Character LoRA

In other words:

The ones you most want to “maintain” the effect of, put last.

5️⃣ Important notes that are often forgotten

Weight is more influential than order
- LoRA weight 1.2 at the beginning can still override LoRA weight 0.6 at the end
Clip vs UNet
- If LoRA targets are different (e.g. one is dominant in CLIP, one in UNet), the order becomes less significant.
Sampler & CFG
- High CFG can make LoRA conflicts more visible

Conclusion 🎯

✔️ The LoRA order does matter, especially if:

The effects overlap each other
Similar weight
Touching the face / lighting / details together

Question: If I use the following LoRA settings globally: Face, Detailer, Lighting, Realistic, Correction (hand/finger), Background Detail. And all the weights are the same, say they all use 0.8 (the default weight). What would be a good LoRA order?
Answer: Imagine you're putting together a movie scene on a mini-stage. There's a set crew, a lighting designer, a makeup artist, and a final editor. They all have the same volume (weight 0.8), so whoever comes in last will have the most say in the final touches. 🎬

Assuming all weights are the same (0.8) and that LoRAs potentially touch overlapping areas, this is the most stable and “sane” sequence globally:

🔑 Main principles (in summary)

General → specific → corrective → finishing touches

Those who shape the world are done first, those who correct mistakes come later.

✅ Recommended LoRA order

1️⃣ Background details

Reason:

Most global and insensitive
Determine the texture of the environment, architecture, foliage, etc.
Hardly any conflict with the face or anatomy

It's like building a stage set first.

2️⃣ Lighting

Reason:

Lighting affects the perception of all objects
It is better to be “accepted” by the face and skin, not to be inflicted on it.

The lights are installed before the actor is made up.

3️⃣ Detailer

Reason:

Increasing global sharpness
If placed too late, it can damage the skin or hands.

This is basic sharpening before the touch of realism.

4️⃣ Realistic

Reason:

Adjusting natural texture: skin blending, tone, micro-imperfection
Taming the effects of previous over-detailing

Make-up stage so it doesn't look digital.

5️⃣ Face

Reason:

Very sensitive
Often adjust the proportions, eyes, nose, expression
Must come after lighting & realism for consistency

The main actors are made up after the stage and lights are ready.

6️⃣ Correction (hand/finger)

Reason:

Corrective in nature
Often “override” other LoRAs
The safest place is at the end.

The editor comes last, correcting anything strange.

🧾 Final order (summarized)

Background Detail
→ Lighting
→ Detailer
→ Realistic
→ Face
→ Hand/Finger Correction

⚠️ Important note (small but valuable gold)

With the same weight (0.8), this sequence really feels the effect.
If the hands still don't fit:
- hand correction can be increased to 0.9–1.0 without messing up the face
If the face becomes too “cartoonish”:
- lower the face LoRA to 0.7, don't change the order yet

🎯 Conclusion

✔️ The LoRA order is not about right or wrong, but who has the final say.

Question: I use Flux and SDXL
Answer: For Flux and SDXL, the rules are philosophically similar, but the sense of influence of the LoRA order is different. Think of Flux as fast-flowing, adaptive water, while SDXL is like thick oil paint that leaves brush marks. Therefore, the order needs to be adjusted slightly. 🎨

Below I give a separate optimal sequence, assuming:

All LoRA weight = 0.8
LoRA Types: Face, Detailer, Lighting, Realistic, Hand/finger correction, Background details
Target: realistic and stable

🔷 FLUX (all variants)

Flux is more “context-respecting,” so LoRA coming in last is less brutal, but still impactful.

✅ Best order for FLUX

Background Detail
→ Lighting
→ Detailer
→ Face
→ Realistic
→ Hand/Finger Correction

Why is it different?

Realistic behind the face
- Flux blends more smoothly, so realism can “reset” the face without destroying identity.
Detailing should not be too final
- If it's too far back, Flux tends to make the skin too rough.

📌 Flux Notes:

Hand correction in the last position is almost always the safest.
Flux is relatively tolerant, so this sequence is stable even for full body

🔷 SDXL

SDXL is much more sensitive to the “last word.” The LoRA sequence in SDXL has a more pronounced effect.

✅Best order for SDXL

Background Detail
→ Lighting
→ Detailer
→ Realistic
→ Face
→ Hand/Finger Correction

Why is the face later?

LoRA faces in SDXL often:
- change the structure
- expression
- eye & nose shape
If placed before realism, the result can feel “polished” and lose character.

📌 SDXL Notes:

Realistic functions as a balance of detail
The face should “lock identity” at almost the end

🧠 Quick summary

Model Key sequence
Flux: background → lighting → detailer → face → realistic → correction
SDXL: background → lighting → detailer → realistic → face → correction

⚠️ Hassle-free tuning tips

Broken hand?
👉 increase LoRA correction to 0.9–1.0
Skin too sharp?
👉 lower Detailer to 0.6–0.7
Does face feel “less alive”?
👉 increase the face LoRA 0.85 without changing the order

🎯 Conclusion

✔️ LoRA order still matters in Flux and SDXL, but:

Flux = more flexible
SDXL = more sensitive

✔️ Weight overrides all sequences/orders 👉 Higher weight beats smaller weight for similar or identical LoRA

That's all, folks. I created this guide based on my experience in generating text into images. Since I mostly use the Flux and SDXL/Pony Base Models, I think it won't be much different for other Base Models, so feel free to try it yourself.

Final Note: This is just a guide, so it's not absolute; it all comes down to your creativity and imagination.