Sign In

Mechanica: Clockwork Concept

91

505

128

31

Verified:

SafeTensor

Type

LoRA

Stats

505

128

1.2k

Reviews

Published

Nov 2, 2024

Base Model

Flux.1 D

Training

Steps: 456

Usage Tips

Strength: 1

Trigger Words

clckwrk

Hash

AutoV2
797D6A0295

The FLUX.1 [dev] Model is licensed by Black Forest Labs. Inc. under the FLUX.1 [dev] Non-Commercial License. Copyright Black Forest Labs. Inc.

IN NO EVENT SHALL BLACK FOREST LABS, INC. BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH USE OF THIS MODEL.

The Art of Clockwork Creation

Enter a world of intricate clockwork art and mechanical creatures with Mechanica, a LoRA designed to transform your subjects into creations of gears, cogs, and metallic beauty. Inspired by a mix of steampunk, clockwork sculptures, and mechanized industrial fantasy, Mechanica reimagines animals, people, and objects as finely crafted clockwork automata. It'll create anything from vigilant clockwork animals with jeweled eyes to mechanical armor-adorned warriors.

Flux thinks anything clockwork just needs to look steampunk, but this LoRA will add gears, cogs, and wheels that are needed to really make it into a clockwork automaton. It also makes cool mechanical machines and fine clockwork parts for actual clocks.

Usage

To use the most recent version of the LoRA, use the following settings:

  • Trigger word: clckwrk

  • Other tokens that work well: describing subjects as "clockwork" or "mechanical" help things go in the right direction. In addition, explain how you want the "gears" "cogs" and "wheels" to look.

  • Lora Strength: A strength between 0.8 and 1.2 is recommended for Flux prompts written in natural language, but pull back to 0.6 to 1.0 for Danbooru tag Flux prompts that usually name more anime-style images.

For differences in previous versions, see the version notes below.

Version Notes

So, starting with v4.2 may seem odd, but this concept was surprisingly difficult to train.

Version 4.2

After trying out many variations, I discovered that the concept trained better with just four images over the 20 or so I had tried to use in v1 and v2 (failed versions). Training this LoRA at 512 resolution worked better than 1024, with a batch size of 4 working better than 1. Captioning was also tricky, so I had to describe things as if they were not clockwork to get the training to stick. It mainly will work for animals and some people, but I've managed to get some excellent clockwork items as well.