This is a simple workflow for chaining multiple LoRAs in text-to-image to produce stable multi-character images with clean faces and controlled style.
It uses LoRA Loader (Block Weights) from the COMFYUI Inspire Pack.
You can chain multiple LoRAs like a sequence:
Checkpoint -> LoRA 1 (Character 1) -> LoRA 2 (Character 2) -> ... -> LoRA N (Style, if needed) -> KSampler
Key Tips:
Model weight flow is important: use a descending strength approach to avoid overbaked shading and face distortion.
Block weights control how strongly a LoRA affects each layer of the network:
Example:
LoRA 1 (Character 1):
1,1,1,1,0,0,0,0,0,0,0,0LoRA 2 (Character 2):
0,0,0,0,1,1,1,0,0,0,0,0
Style LoRAs usually get higher values in later blocks to preserve the artistic style without affecting the face.
Following this workflow ensures:
Multi-character images without bleeding
Clean, stable faces
Controlled style and shading
Description
first version
Comments (3)
hi! do you have any tips on how to work with 3+ character LORAs? it works incredible with just 2 LORAs, but I can't figure out how to not mix faces&clothes with more of them
In the description i just gave a rough idea about it
# LORA BLOCK WEIGHT PRESETS (12 Blocks)
# LoRA 1: Focuses on the first 3 blocks
1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0
# LoRA 2: Focuses on the middle-front 3 blocks
0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0
# LoRA 3: Focuses on the middle-back 3 blocks
0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0
# LoRA 4 (Style/Detail): Focuses on the last 3 blocks
0, 0, 0, 0, 0, 0, 0, 0
, 0, 1, 1, 1
Fine-Tuning: You don't have to use just 1 or 0. If a LoRA is too strong and ruining the face, try lowering its specific blocks to 0.7 or 0.8 instead of a full 1.
Also if you do not need a style lora you can just distribute them among the first three(or more if you have more lora)
# LORA BLOCK WEIGHT PRESETS (12 Blocks)
# LoRA 1: Focuses on the first 3 blocks
1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0
# LoRA 2: Focuses on the middle-front 3 blocks
0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0
# LoRA 3: Focuses on the middle-back 3 blocks
0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0
# LoRA 4 (Style/Detail): Focuses on the last 3 blocks
0, 0, 0, 0, 0, 0, 0, 0
, 0, 1, 1, 1
Fine-Tuning: You don't have to use just 1 or 0. If a LoRA is too strong and ruining the face, try lowering its specific blocks to 0.7 or 0.8 instead of a full 1.
Also if you do not need a style lora you can just distribute them among the first three(or more if you have more lora)