You can support my work by posting your images here, by tipping Buzz, and by leaving a like if you enjoyed it! Get exclusive NSFW models at DeviantArt or Booth.pm. Follow for more!
Tags: sitting on person
Recommended quality tags: score_9, score_8_up, score_7_up, score_6_up. Optional: source_anime
Recommended negative tags: score_4, score_5. Optional: 3d, lips, realistic
Note that the end result can be affected by the base model used.
Description
FAQ
Comments (12)
Waifu riding me into battle with this one
randomly it does very great outputs especially when leash/collar is involved , there was similiar lora but this one is better somehow with realism
thanks for the feedback
Idk what is the differences, but you LoRAs seems always work really well with Regional Prompter. Other people's LoRAs are kind of hit or miss when I use it with Regional Prompter.
thanks. I'm not sure either. haven't used regional prompter in a while. but I also don't use the most common training parameters
I use this NAI3 style base model:https://civitai.com/models/317902/t-ponynai3
I found that when the Lora Strength value is higher than 0.8, the image style generated by the model will be biased towards 3d and realistic.
I cannot add 3D and realistic tags to negative tags, as this would significantly degrade the quality of the generated images.
That's an interesting remark. But it most likely has to do with that base model: the dataset for this lora does not include any 3d image, there is no "realistic" or "3d" in the tags, and it was trained on pony v6. So I see no reason for that effect. Have you tried to see if the "sitting" or "person" tags when used with high weights on that model (without loras) gives that kind of output?
@psoft My previous statement was not very accurate; to be precise, the generated art style leans more towards Western anime style rather than Japanese anime style.
@psoft The author of the NAI3 style model said, "The word 'anime' is used to train the art style of nai3. If European and American art styles appear during use, this word can be used to stabilize the nai3 art style. This is a method I tried to reduce overtrain."However, after using your Lora model, the generated images will not be in the NAI3 style, regardless of whether "anime" is added to the positive tags or not.
@FrostKinetics I see. I only used anime images in the training, so maybe the output could be leaning towards the base Pony aesthetics instead of NAI3, since the latter is still based on Pony. From my understanding, T-ponynai3 was created by training style loras and merging into Pony, instead of actual fine tuning. It may be that if I trained this lora with the tag "anime" it could help when using that model, but it could deteriorate the result for the users of other checkpoints. I may do some testing on that.
Maybe you can refer to this author's Lora model: https://civitai.com/models/445021?modelVersionId=495569
The Lora model he trained does not seem to use "anime" as a training label for the dataset, but his Lora model can generate images in various styles quite well, the only drawback is that the details of the characters' eyes and hands are relatively poor.
damn son
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.