Uploaded by request. A merge of a few custom checkpoints I made in my personal collection. A specialty with the female form in a hotly saturated cartoon style. SFW/NSFW
Illustrious V4
Trained at 1024x1024 - Even stronger western and semi-realistic 2.5D tones
Sampler/Scheduler: DPM ++2M/Karras
Steps: 20-25
CFG best at 5
Adetailer is recommend for face sizes lower than 1/8th of the generated image size.
Optionally emphasize the realism by using the prompt:
masterpiece, ultra-detailed, cinematic lighting, highres, soft shadows, global illumination, volumetric lighting, subsurface scattering, bounce lighting, physically accurate lighting,
Illustrious V3
Trained at 1024x1024 - Vector Adjacent 2D
Sampler/Scheduler: DPM ++2M/Karras
Steps: 20-25
CFG best at 5
Adetailer is recommend for face sizes lower than 1/8th of the generated image size.
Illustrious V2
Trained at 1024x1024 - Stronger western and semi-realistic 2.5D tones
Sampler/Scheduler: DPM ++2M/Karras
Steps: 20-30
CFG best at 3.5 - 5
Adetailer is recommend for face sizes lower than 1/8th of the generated image size.
Emphasize the realism by using the prompt:
masterpiece, ultra-detailed, cinematic lighting, highres, soft shadows, global illumination, volumetric lighting, subsurface scattering, bounce lighting, physically accurate lighting,
Illustrious V1
Trained at 1024x1024
CFG best at 3.5 - 5
CD-Tuner - Contrast(con1) +6 for darker images in model example images
--
Pony V1
Trained with SD15 dataset. Leaning towards solid vector coloring.
Pony V2
More semi-realistic tones and lighting. Added details and removed common errors.
Pony V3
Sketchy throwback style. "Cuter" and more expressive.
Pony V4
Strong semi-realistic tones.
CD-Tuner - Contrast(con1) +6 for darker images in model example images
--
SD 1.5
Works best at 768x768 but I've had pleasing results at 512x1024 and 1024x1024.
I've found that an abundance of negative prompts hurt the style, but by all means can be used.
Character LoRAs have little issue also.
ADetailer recommended for portraits sized 512x768, 512x1024 and 768x1024.
Extensively tested with LCM FP8 precision.
Description
PonyXL V2 data trained at 1024x1024, 720x1280 and 896x1152. Merged along with many other checkpoints in small amounts. Corrected graphical errors, upped the detail and added softer lighting.