Additional words that are also essential if you want to capture her full original essence: 1girl, solo, blue hair, short hair, spiked hair, blue eyes, sleeveless dark blue turtleneck, bare shoulders, purple chest strap, silver chest ornament, arm armor, white detached sleeves, black corset, white laces, blue skirt, white ribbons, grey fingerless gloves, black shorts, black thighhighs DO NOT PUT EMPHASIS ON ANY OF HER TRAITS PAST Strength: 1.0!  It will BREAK your image! You can change her outfit by excluding clothing tags. If you find the base clothing still influencing your prompt, try putting her clothing tags that are interfering in the negative. Settings I used for pretty much all of my generationsSampler: Euler ASteps: 32Strength: 1.0 (You can lower it, and it will still do a good job at capturing her.) Upscaler: R-ESRGAN 4x, 15-30 stepsADetailer model: face_<a target="_blank" rel="ugc" href="http://yolov8n.pt/">yolov8n.pt</a> (default settings)This model actually kind of struggles with transferring her to different 3D styles (depending on the style, what checkpoint you're using, and what you're hoping to achieve), but I find that it actually works exceptionally well with 2D styles. If 2D styles aren't looking exactly right, add (3d), (realistic), (render) to the negatives. Not always necessary. While I'm happier with this model than I am with any of the other Aqua models I've seen, this model isn't perfect. Far from it. It's slightly overtrained in some ways, and not trained well enough in others. It was trained on a "sub-optimal" data set and parameters, but as of right now, it's the best I have before I take a break, so I'm releasing it. I'll return and make a model for Aqua that's the closest to perfection I can get. Eventually. I'll run many more training sets and have a far better dataset than I currently do. I've put a lot of work into this, and I'll put in a lot more until it's perfect. If anyone wants to give me any tips or pointers in the right direction for training, I'm open to it.

[Aqua] From Kingdom Hearts

Additional words that are also essential if you want to capture her full original essence: &lt;span style="color: #00aaff;"&gt;1girl, solo, blue hair, short hair, spiked hair, blue eyes, sleeveless dark blue turtleneck, bare shoulders, purple chest strap, silver chest ornament, arm armor, white detached sleeves, black corset, white laces, blue skirt, white ribbons, grey fingerless gloves, black shorts, black thighhighs&lt;/span&gt; DO NOT PUT EMPHASIS ON ANY OF HER TRAITS PAST Strength: 1.0!  It will BREAK your image! You can change her outfit by excluding clothing tags. If you find the base clothing still influencing your prompt, try putting her clothing tags that are interfering in the negative. Settings I used for pretty much all of my generationsSampler: Euler ASteps: 32Strength: 1.0 (You can lower it, and it will still do a good job at capturing her.) Upscaler: R-ESRGAN 4x, 15-30 stepsADetailer model: face_<a target="_blank" rel="ugc" href="http://yolov8n.pt/">yolov8n.pt</a> (default settings)This model actually kind of struggles with transferring her to different 3D styles (depending on the style, what checkpoint you're using, and what you're hoping to achieve), but I find that it actually works exceptionally well with 2D styles. If 2D styles aren't looking exactly right, add (3d), (realistic), (render) to the negatives. Not always necessary. While I'm happier with this model than I am with any of the other Aqua models I've seen, this model isn't perfect. Far from it. It's slightly overtrained in some ways, and not trained well enough in others. It was trained on a "sub-optimal" data set and parameters, but as of right now, it's the best I have before I take a break, so I'm releasing it. I'll return and make a model for Aqua that's the closest to perfection I can get. Eventually. I'll run many more training sets and have a far better dataset than I currently do. I've put a lot of work into this, and I'll put in a lot more until it's perfect. If anyone wants to give me any tips or pointers in the right direction for training, I'm open to it.

AquaPonyV1.0.safetensors

PonyV1.0

V1.0

Additional words that are also essential if you want to capture her full original essence: 1girl, solo, blue hair, short hair, spiked hair, blue eyes, sleeveless dark blue turtleneck, bare shoulders, purple chest strap, silver chest ornament, arm armor, white detached sleeves, black corset, white laces, blue skirt, white ribbons, grey fingerless gloves, black shorts, black thighhighs DO NOT PUT EMPHASIS ON ANY OF HER TRAITS PAST Strength: 1.0! It will BREAK your image! You can change her outfit by excluding clothing tags. If you find the base clothing still influencing your prompt, try putting her clothing tags that are interfering in the negative. Settings I used for pretty much all of my generations Sampler: Euler A Steps: 32 Strength: 1.0 (You can lower it, and it will still do a good job at capturing her.) Upscaler: R-ESRGAN 4x, 15-30 steps ADetailer model: face_yolov8n.pt (default settings) This model actually kind of struggles with transferring her to different 3D styles (depending on the style, what checkpoint you're using, and what you're hoping to achieve), but I find that it actually works exceptionally well with 2D styles. If 2D styles aren't looking exactly right, add (3d), (realistic), (render) to the negatives. Not always necessary. While I'm happier with this model than I am with any of the other Aqua models I've seen, this model isn't perfect. Far from it. It's slightly overtrained in some ways, and not trained well enough in others. It was trained on a "sub-optimal" data set and parameters, but as of right now, it's the best I have before I take a break, so I'm releasing it. I'll return and make a model for Aqua that's the closest to perfection I can get. Eventually. I'll run many more training sets and have a far better dataset than I currently do. I've put a lot of work into this, and I'll put in a lot more until it's perfect. If anyone wants to give me any tips or pointers in the right direction for training, I'm open to it.

Files with SHA256: 9e6ea8cafbd6cd28825e3a4c5c9aef48238007ff7f19e498c9dd99d785c9ab0d