[Feature]: xyz lora strength with multiple loras

[SAC020]

Issue Description

I try to generate an xyz with a number of loras (same lora, different epochs) at various strenghts.

Expected behavior: iterate each lora for 0.2, 0.5, 1.5 strengths

In reality, it does:

Lora1 1.0 / 0.2 / 0.5
Lora2 1.5 / 0.2 / 0.5
Lora3 1.5 / 0.2 / 0.5

Log excerpts below, I've tried to keep the clutter at a minimum

Full log attached as a file

130855-807475 INFO XYZ grid sta.txt

11:33:39-659849 INFO XYZ grid start: images=30 grid=1 shape=10x3 cells=1 steps=1200
11:33:39-661843 DEBUG XYZ grid apply LoRA: "sacbf-dalle-flux-v3-colab-concept-concise_captions-adam_000000500"
11:33:39-662840 DEBUG XYZ grid apply setting: extra_networks_default_multiplier=0.2
11:33:40-580892 INFO Load network: type=LoRA
apply=['sacbf-dalle-flux-v3-colab-concept-concise_captions-adam_000000500'] te=[1.0]
unet=[[1.0, 1.0, 1.0]] dims=[None] load=0.92

12:29:19-694640 DEBUG XYZ grid apply LoRA: "sacbf-dalle-flux-v3-colab-concept-concise_captions-adam_000000500"
12:29:19-695638 DEBUG XYZ grid apply setting: extra_networks_default_multiplier=0.5
12:29:28-870061 INFO Load network: type=LoRA
apply=['sacbf-dalle-flux-v3-colab-concept-concise_captions-adam_000000500'] te=[0.2]
unet=[[0.2, 0.2, 0.2]] dims=[None] load=9.17

12:30:19-562619 DEBUG XYZ grid apply LoRA: "sacbf-dalle-flux-v3-colab-concept-concise_captions-adam_000000500"
12:30:19-563617 DEBUG XYZ grid apply setting: extra_networks_default_multiplier=1.5
12:30:20-238280 INFO Load network: type=LoRA
apply=['sacbf-dalle-flux-v3-colab-concept-concise_captions-adam_000000500'] te=[0.5]
unet=[[0.5, 0.5, 0.5]] dims=[None] load=0.67

12:31:03-149304 DEBUG XYZ grid apply LoRA: "sacbf-dalle-flux-v3-colab-concept-concise_captions-adam_000001000"
12:31:03-150792 DEBUG XYZ grid apply setting: extra_networks_default_multiplier=0.2
12:31:04-668023 INFO Load network: type=LoRA
apply=['sacbf-dalle-flux-v3-colab-concept-concise_captions-adam_000001000'] te=[1.5]
unet=[[1.5, 1.5, 1.5]] dims=[None] load=1.52

12:31:47-490868 DEBUG XYZ grid apply LoRA: "sacbf-dalle-flux-v3-colab-concept-concise_captions-adam_000001000"
12:31:47-491867 DEBUG XYZ grid apply setting: extra_networks_default_multiplier=0.5
12:31:48-185527 INFO Load network: type=LoRA
apply=['sacbf-dalle-flux-v3-colab-concept-concise_captions-adam_000001000'] te=[0.2]
unet=[[0.2, 0.2, 0.2]] dims=[None] load=0.69

12:32:31-703321 DEBUG XYZ grid apply LoRA: "sacbf-dalle-flux-v3-colab-concept-concise_captions-adam_000001000"
12:32:31-704319 DEBUG XYZ grid apply setting: extra_networks_default_multiplier=1.5
12:32:32-421913 INFO Load network: type=LoRA
apply=['sacbf-dalle-flux-v3-colab-concept-concise_captions-adam_000001000'] te=[0.5]
unet=[[0.5, 0.5, 0.5]] dims=[None] load=0.72

Version Platform Description

PS C:\ai\automatic> .\webui --debug --medvram
Using VENV: C:\ai\automatic\venv
13:45:14-797851 INFO Starting SD.Next
13:45:14-800843 INFO Logger: file="C:\ai\automatic\sdnext.log" level=DEBUG size=65 mode=create
13:45:14-801841 INFO Python: version=3.11.9 platform=Windows bin="C:\ai\automatic\venv\Scripts\Python.exe"
venv="C:\ai\automatic\venv"
13:45:15-079447 INFO Version: app=sd.next updated=2024-10-21 hash=cc4f4819 branch=dev
url=https://github.com/vladmandic/automatic/tree/dev ui=dev
13:45:15-930345 INFO Repository latest available 0d332ca 2024-10-23T18:59:10Z
13:45:15-943385 INFO Platform: arch=AMD64 cpu=Intel64 Family 6 Model 165 Stepping 5, GenuineIntel system=Windows
release=Windows-10-10.0.22631-SP0 python=3.11.9

Relevant log output

No response

Backend

Diffusers

UI

Standard

Branch

Dev

Model

Other

Acknowledgements

• I have read the above and searched for existing issues
• I confirm that this is classified correctly and its not an extension issue

[vladmandic]

lora strength is applied by modifying extra_networks_default_multiplier setting which means its global for all loras.
this is a valid request, but i'll have to treat it as feature request.

[SAC020]

I am not sure I explained very clearly.

I don't have other loras in the prompt, it's just those on X. The problem is some sort of "delayed" action or precedence: first iteration, the lora is called with strength 1 when xyz says otherwise, then every subsequent iteration is the chain is shifted by 1 step.

It's something like:

for i = 1 to max_iterations
{
set xyz_lora_strength = 0.2;
call lora (default_strength);
set default_strength = xyz_lora_strength;
i = i+1;
}

(of course it's my aproximation of the script, but it's exactly the behavior I am seeing)

[vladmandic]

i understand.

[SAC020]

Simplified example: asked for strengths 0 and 0, should produce identical outputs (zero strength)

Output

Log excerpt

08:04:26-760520 INFO XYZ grid start: images=2 grid=1 shape=1x2 cells=1 steps=40
08:04:26-762516 DEBUG XYZ grid process: x=1/1 y=1/2 z=1/1 total=0.50
08:04:26-764510 DEBUG XYZ grid apply LoRA: "amateurphoto-v6-forcu"

08:04:26-765507 DEBUG XYZ grid apply setting: extra_networks_default_multiplier=0.0

08:04:28-536586 INFO Load network: type=LoRA apply=['amateurphoto-v6-forcu'] te=[1.0] unet=[[1.0, 1.0, 1.0]] dims=[None] load=1.77

08:05:26-346604 INFO Processed: images=1 its=0.34 time=59.58 timers={'init': 1.77, 'prepare': 0.03, 'args': 0.03, 'callback': 3.21,
'pipeline': 55.06, 'decode': 2.66, 'post': 0.03} memory={'ram': {'used': 29.18, 'total': 63.92}, 'gpu': {'used':
2.66, 'total': 23.99}, 'retries': 0, 'oom': 0}
08:05:26-349595 DEBUG XYZ grid process: x=1/1 y=2/2 z=1/1 total=1.00
08:05:26-350593 DEBUG XYZ grid apply LoRA: "amateurphoto-v6-forcu"

08:05:26-352587 DEBUG XYZ grid apply setting: extra_networks_default_multiplier=0.0

08:05:28-691847 INFO Load network: type=LoRA apply=['amateurphoto-v6-forcu'] te=[0.0] unet=[[0.0, 0.0, 0.0]] dims=[None] load=2.34

08:05:59-622190 INFO Processed: images=1 its=0.60 time=33.27 timers={'init': 2.34, 'callback': 3.6, 'pipeline': 30.44, 'decode': 0.44,
'post': 0.03} memory={'ram': {'used': 29.3, 'total': 63.92}, 'gpu': {'used': 7.33, 'total': 23.99}, 'retries': 0,
'oom': 0}
08:05:59-668068 INFO XYZ grid complete: images=2 size=(1792, 2322) time=92.86 save=0.04
08:05:59-670062 DEBUG XYZ grid: grid=1 subgrids=0 images=2 total=3
08:05:59-687018 INFO Processed: images=3 its=3760.02 time=0.02 timers={'init': 2.39, 'callback': 3.6, 'pipeline': 30.44, 'decode':
0.44, 'post': 0.05} memory={'ram': {'used': 29.31, 'total': 63.92}, 'gpu': {'used': 7.33, 'total': 23.99},
'retries': 0, 'oom': 0}