Updated to v2.0:
This version enhances likeness and flexibility by leveraging a higher-quality dataset and introducing a new captioning strategy. Based on my tests, it also appears to perform much better than v1.0 when used with other LoRAs. Your feedback is always welcome and greatly appreciated. ❤️
Trigger word: m0l3r4t mole-rat
Strength: 1, Lower it as needed, for example, when working with other LoRAs or if the image appears overly realistic.
Important: If the characteristics of a naked mole-rat are not appearing, it can be helpful to add descriptions of its typical features into your prompt. This approach draws more from the model's training data rather than increasing the model strength, which could lead to over-specialization, as the model becomes too focused on the training data and loses its ability to generalize to new inputs.
I found that adding this description to the prompt works exceptionally well.
The m0l3r4t mole-rat has prominent incisors, and exposed, pink skin.________________________________________________________________________________________________
v1.0 - deprecated
Usage:
Trigger word: mole rat
Strength: 1
Add this description to enhance the effect if necessary: The mole rat has wrinkled face, beady black eyes, sharp incisors, and exposed, pink skin.
If you enjoy this LoRA, please give it a LIKE! Better yet, share an image created with this LoRA by clicking the "+ Add Post" button on this page. It will help me gain buzz to train LoRAs in the future.
I created this LoRA as a learning project, and I’d greatly appreciate any feedback you may have. ❤️
Description
This version enhances likeness and flexibility by leveraging a higher-quality dataset and introducing a new captioning strategy. Based on my tests, it also appears to perform better when used with other LoRAs. Your feedback is always welcome and greatly appreciated.
FAQ
Comments (2)
I really like it as this LoRA is so... different. ;)
What did you learn about captioning? What didn't work so well and what is working great now?
I’m glad you like it! For this version, I used the captioning strategy shared in the tutorial. In short, I avoided all terminology that Flux has a strong association with in the captions. For the Naked Mole-rat LoRA, this includes terms like 'mole,' 'rat,' 'naked,' and 'animal.' Instead, I used a unique keyword to refer to the target concept and focused on describing the physical, observable details of the target object.
Previously, I tried different trigger words and captioning methods. However, they all tended to introduce characteristics of moles or rats, or resulted in the subject being overly furry if I used the class name 'animal.' This was especially noticeable when working with other LoRAs.
I hope this information is helpful for you. Happy training! ❤️
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.