HOME | DD

Description
Finishing up testing on my first slimepouringdownonhead LoRA for Stable Diffusion XL. I created this as I had an issue with slime on stable diffusion. It will render slime perfectly well. It will render a subject covered in slime maybe adequately is the best way to describe it. Where it failed was trying to get it to pour, it or dump it, much less dump it onto a person.

I am posting some results from a 100-batch run, and what I feel are the top 30 images from the run. While there are others that were arguably "nicer, or prettier subject, or whatever," these are my top 30. My goal is to test the effectiveness of the LoRA so I am looking for consistency of the slime, the interaction with the model being slimed, and the coverage, and in this case the realism. I would say it's a pretty solid LoRA, but always room for improvement. That said these where with a single prompt and single setting in one batch run, and I will include the prompt below, don't expect to get the same results without the LoRA, if you do get them let me know *wink.*

I am just going to do the same for different LoRA epochs at different strengths over the next few coming days and hopefully views, favorites and comments help me determine which one to keep and which one to trash. I have three versions, and plan on keeping only 1, the third I believe is too overcooked, so I don't expect alot of good output. So, thank you in advance for your help by lending me your views, favorites and comments. Being just a test, none of these have been altered in anyway. No cropping, no image correction, no color or saturation adjustment, just pure original output. So are they perfect, absolutely not, but I think they show the potential.

I know the imagery doesn't change much. I have installed dynamic prompt and after the first couple sets, I incorporated it and randomly had each generated image vary the hairstyle, hair color, outfit, outfit color, setting along with the subject kneeling, sitting, or standing. I feel that will give me a better baseline of what to expect from the LoRA going forward. Oddly enough I also get better results using the keywords as part of the prompt, but with the LoRA off the results suck. So, I am doing both. Maybe once I refine a future LoRA incorporating these results back into training even that won't be necessary. I also plan to develop one for You Can't Do That on Television Slime specifically, chunky, and thick. I also plan to test it with other models, and other styles, particularly anime.

Anyway, with all that out of the way, enjoy!

parameters

SPDOH, photo of a young woman, green slime (substance) pouring down on head, wearing black dress, smiling, happy, rim lighting, studio lighting, dslr, ultra quality, sharp focus, tack sharp, dof, film grain, Fujifilm XT3, crystal clear, 8k UHD, award winning,
Negative prompt: disfigured, ugly, bad, immature, cartoon, anime, 3d, painting, b&w, hat
Steps: 150, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3972122629, Size: 768x1280, Model hash: 0f1b80cfe8, Model: dreamshaperXL10_alpha2Xl10, VAE hash: 63aeecb90f, VAE: sdxl_vae.safetensors, Denoising strength: 0.1, Hires upscale: 2, Hires steps: 10, Hires upscaler: 4x_NMKD-Superscale-SP_178000_G, Lora hashes: "SPDOH-000020: 0a733c13270b", Version: v1.6.0

exif comment

SPDOH, photo of a young woman, green slime (substance) pouring down on head, wearing black dress, smiling, happy, rim lighting, studio lighting, dslr, ultra quality, sharp focus, tack sharp, dof, film grain, Fujifilm XT3, crystal clear, 8k UHD, award winning,
Negative prompt: disfigured, ugly, bad, immature, cartoon, anime, 3d, painting, b&w, hat
Steps: 150, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3972122629, Size: 768x1280, Model hash: 0f1b80cfe8, Model: dreamshaperXL10_alpha2Xl10, VAE hash: 63aeecb90f, VAE: sdxl_vae.safetensors, Denoising strength: 0.1, Hires upscale: 2, Hires steps: 10, Hires upscaler: 4x_NMKD-Superscale-SP_178000_G, Lora hashes: "SPDOH-000020: 0a733c13270b", Version: v1.6.0

Description
Finishing up testing on my first slimepouringdownonhead LoRA for Stable Diffusion XL. I created this as I had an issue with slime on stable diffusion. It will render slime perfectly well. It will render a subject covered in slime maybe adequately is the best way to describe it. Where it failed was trying to get it to pour, it or dump it, much less dump it onto a person.

I am posting some results from a 100-batch run, and what I feel are the top 30 images from the run. While there are others that were arguably "nicer, or prettier subject, or whatever," these are my top 30. My goal is to test the effectiveness of the LoRA so I am looking for consistency of the slime, the interaction with the model being slimed, and the coverage, and in this case the realism. I would say it's a pretty solid LoRA, but always room for improvement. That said these where with a single prompt and single setting in one batch run, and I will include the prompt below, don't expect to get the same results without the LoRA, if you do get them let me know *wink.*

I am just going to do the same for different LoRA epochs at different strengths over the next few coming days and hopefully views, favorites and comments help me determine which one to keep and which one to trash. I have three versions, and plan on keeping only 1, the third I believe is too overcooked, so I don't expect alot of good output. So, thank you in advance for your help by lending me your views, favorites and comments. Being just a test, none of these have been altered in anyway. No cropping, no image correction, no color or saturation adjustment, just pure original output. So are they perfect, absolutely not, but I think they show the potential.

I know the imagery doesn't change much. I have installed dynamic prompt and after the first couple sets, I incorporated it and randomly had each generated image vary the hairstyle, hair color, outfit, outfit color, setting along with the subject kneeling, sitting, or standing. I feel that will give me a better baseline of what to expect from the LoRA going forward. Oddly enough I also get better results using the keywords as part of the prompt, but with the LoRA off the results suck. So, I am doing both. Maybe once I refine a future LoRA incorporating these results back into training even that won't be necessary. I also plan to develop one for You Can't Do That on Television Slime specifically, chunky, and thick. I also plan to test it with other models, and other styles, particularly anime.

Anyway, with all that out of the way, enjoy!

parameters

SPDOH, photo of a young woman, green slime (substance) pouring down on head, wearing black dress, smiling, happy, rim lighting, studio lighting, dslr, ultra quality, sharp focus, tack sharp, dof, film grain, Fujifilm XT3, crystal clear, 8k UHD, award winning,
Negative prompt: disfigured, ugly, bad, immature, cartoon, anime, 3d, painting, b&w, hat
Steps: 150, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3972122629, Size: 768x1280, Model hash: 0f1b80cfe8, Model: dreamshaperXL10_alpha2Xl10, VAE hash: 63aeecb90f, VAE: sdxl_vae.safetensors, Denoising strength: 0.1, Hires upscale: 2, Hires steps: 10, Hires upscaler: 4x_NMKD-Superscale-SP_178000_G, Lora hashes: "SPDOH-000020: 0a733c13270b", Version: v1.6.0

exif comment

SPDOH, photo of a young woman, green slime (substance) pouring down on head, wearing black dress, smiling, happy, rim lighting, studio lighting, dslr, ultra quality, sharp focus, tack sharp, dof, film grain, Fujifilm XT3, crystal clear, 8k UHD, award winning,
Negative prompt: disfigured, ugly, bad, immature, cartoon, anime, 3d, painting, b&w, hat
Steps: 150, Sampler: DPM++ 2M Karras, CFG scale: 7, Seed: 3972122629, Size: 768x1280, Model hash: 0f1b80cfe8, Model: dreamshaperXL10_alpha2Xl10, VAE hash: 63aeecb90f, VAE: sdxl_vae.safetensors, Denoising strength: 0.1, Hires upscale: 2, Hires steps: 10, Hires upscaler: 4x_NMKD-Superscale-SP_178000_G, Lora hashes: "SPDOH-000020: 0a733c13270b", Version: v1.6.0