Type | |
Stats | 417 |
Reviews | (76) |
Published | May 22, 2024 |
Base Model | |
Usage Tips | Clip Skip: 2 Strength: 0.8 |
Trigger Words | sisu rald |
Hash | AutoV2 484AA6BC87 |
V2.5 [BETA]
This is the first test-training run with PonyDiffusionXL for me. There are still some issues which I will fix in the full next version, while upgrading my other LoRAs to PonyXL.
This goes specifically for the SIZE of the LoRA. Currently, I maintained my previous training settings to some extend, resulting in a WAY TOO LARGE file. I am aware of this and will fix it soon.
The general prompting instructions remain the same as for the previous versions. However, pay attention to the provided example images. They contain prompts and suggested resolutions.
V2
The new version can now replicate her feral dragon form. It can be invoked with
feral, eastern dragon, noodle dragon
The new version is trained with a resolution of 896x896 on the PurrfectMerge_v1 checkpoint but also works on near all furry checkpoints. For example prompts, please see the images above.
The bias with generating hats that was present in the first version has been resolved, but it may sometimes generate an additional tail. However, this can be prompted against.
V1
This is the first Lora I trained and my first post on this page :)
The Lora is trained on 20 selected images of Sisu. Training includes NSFW and SFW work, but the Lora mainly produced NSFW anthro output.
2 Epochs at 1500 Samples each at 768x768 with bucketing.
Suggested use:
I suggest the weights stay between 0.6 and 1 (see example images).
The trigger word is "sisu_rald".
Other helpfull prompts include " purple eyes" with "ears" in the negative prompt.
When I get the time, I want to further refine the Lora.
Note:
Right now, there is a bias toward wearing hats, which can be counteracted by heavy negative prompting.