Hey there,

this time, I tried a different approach to training the lora. I used fewer images, which should result in higher consistency but with reduced variation. It seems to work as intended, and I think the results look great, but please let me know what you think.

The strength should be fine at around 0.8 and higher if you use multiple loras.

The triggers are 'cyberarm', 'cyberlegs', 'headoverlay', and 'headwear'. I think they are self-explanatory. As a non-native English speaker, I only realized afterward that the trigger 'headwear' was misspelled and should have been 'headware', but you can still make it work as intended by adding the trigger 'cyberpunk'.

Generally, you should add 'cyberpunk' for a good result.

Also, I recommend you use support triggers like 'hologram' or 'digital' to add detail.

The lora seems to generate cyberarms and cyberlegs without the trigger word quite frequently. In case you don't want that, put the trigger in the negative prompt, and it should be fixed.

Please refer to the example images for the prompts.

Thanks for reading, and have fun with it!