Could probably rescale the slider by multiplying all weights by a factor, no?

Also, imo, diversity or "seed strength" slider might be clearer on the purpose. Till seeing the infographic I just got the impression it was a slider without any specified purpose, that's what random usually means ("Having no specific pattern, purpose, or objective: synonym: chance."). Even randomNESS would be an improvement.

@ghostpaint I see, so it specifically eliminates zit's rigid posing, rather than simply introduce more variance to the seed?

I wouldn't really focus on the technical correctness in your place, civ users are mostly laymen who understand english, prompting to various extents, and maybe some high level tidbits about how gen models work.

Here are some possible options that come to mind:

Action slider - Probably too subtle for that.

Amateur photo slider - If it affects composition to an extent as well.

Natural pose slider.

Movement slider.

Spontaneity, candid slider - Probably the best term from a photography perspective.

The opposite may be easier to name, "stiffness slider".

About the scaling - I think it should be possible for a chatterbot to write a simple script to read the safetensors directly, modify the weights and resave it. I haven't messed with weights specifically, but layer renaming went without a hitch, and if kohya can handle resizing generically in models it never even heard of, how bad can it be.
Oh, could also do that directly from comfy, assuming you're using that. I'm not familiar with all the node options, but there's this lora optimiser node, it's basically built for merging groups of models in a way that reduces interference, but it can be used with a single lora, increase the weight and it has a save node. https://github.com/ethanfel/ComfyUI-LoRA-Optimizer

@ghostpaint As long the source is clear, that's the most important part and the best you can do on your end. Thanks.

@ghostpaint Here, this code will scale your model, takes all of 15 seconds to run.

```
import safetensors.torch

from safetensors import safe_open

# ===== CONFIGURATION =====

INPUT_FILE = "your-input-file.safetensors"

OUTPUT_FILE = "name-to-save.safetensors"

FACTOR = 2.0

# ========================

print(f"Loading {INPUT_FILE}...")

tensors = safetensors.torch.load_file(INPUT_FILE)

with safe_open(INPUT_FILE, framework="pt") as f:

metadata = f.metadata()

print(f"Loaded {len(tensors)} tensors")

print(f"Multiplying all weights by {FACTOR}...")

multiplied = {k: v * FACTOR for k, v in tensors.items()}

print(f"Saving to {OUTPUT_FILE}...")

safetensors.torch.save_file(multiplied, OUTPUT_FILE, metadata=metadata)

print("Done!")

```

@ghostpaint Yeah, for little mods like that, bare metal scripts typically beat any of the big name tools in simplicity, speed and correctness. If I can figure out how, I always prefer to write my own.

Haven't tested the model's output, only checked that it loads correctly. I have no idea what the correct scale is, 2 is just a placeholder. Feel free to upload the scale that feels right to you as v1.1 or something.