Valiant Stallion v3, reposted with permission from the original Author: Scratchproof
Settings are similar to the Virile Stallion model by the same author:
Sampler: DPM++ SDE, Euler A, UniPC
Steps: 25
CFG: 5
Clip Skip: 1/0
HiRes Steps: 10-12
Denoising: 0.2-0.4 (0.2-0.25 at 2x upscale, higher for closeups and lower upscale values)
Upscale: 1.5-2x
Adetailer: useful, but not strictly necessary unless you can afford to spend for max detail
the prompt is designed to maximize face and eye detail, so upscaling will fix most issues
Adetailer/facedetailer will improve details, unsure if adetailer works in creation mode, but facedetailer works fine in comfy workflow
Note: with the correct prompt, it is extremely good at realism, but it pays for it by being less capable with a number of more common Pony strengths. That said, it is extremely good at img2img tasks from the output of other pony models.
[space reserved for link to Super Valiant - super realistic pony to Valiant Stallion workflow, low cost and creative -- coming soon]
Core Prompt:
realistic, real life, CHARACTER/SUBJECT, looking at viewer, depth of field, bokeh, CLOTHES, EXPRESSION, LOCATION, CAMERA LOCATION (from below/above/side, etc), dynamic angle,
Core Negative:
low quality, lowres, bad anatomy, normal quality, worst quality, watermark, text, logo
Advanced Prompting ideas: (use what works best situationally)
score_9_up, score_8_up, score_7_up, (Best quality, 8k, 32k, raw photo, photorealistic, UHD:1.2),lifelike rendering, cinematic photo, dramatic lighting, CAMERA LOCATION (from below/above/side, etc), dynamic angle, dynamic POSE/MOTION/ACTION, cinematic film still, BREAK
CHARACTER/SUBJECT/CORE PROMPT.
His/Her highly-detailed, realistic eyes sparkle with mesmerizing iris colors, drawing the viewer's gaze to his/her beautiful eyes. The natural/artificial/neon/etc LIGHTING_TYPE lighting of the LOCATION adds to the STYLE aesthetic, highly-detailed eyes, expressive eyes.
CHARACTER/SUBJECT, CLOTHES, ACTIONS/DOING, ADDITIONAL DETAILS, realistic photograph captured with a Canon EOS R6 Mark II and a 35mm f/2.8 lens, BREAK
realistic eyes, highly-detailed eyes (place here if not using below core prompt), expressive eyes, dreamy reflective eyes, soft tones, intricate details, ornate patterns, impeccable backlighting, depth of field, bokeh. aesthetic TYPE_OF_PHOTO/THEME photography, vivacious, depth and emotion, captivating and enigmatic/stark and sinister/PROJECTED_TONE, MOOD/EMOTION/AMBIENT atmosphere, artistic and elegant/simple and rustic/etc, very detailed, high resolution digital photography, soft illumination, backlighting, professional color grading, muted skin colors, perfectly clear, sharp focus on subject, detailed skin, imperfect skin texture, skin pores
Advanced Negatives:
low quality, lowres, bad anatomy, normal quality, worst quality, 3D, render, illustration, drawing, comic, watermark, text, logo, website, bad anatomy, bad hands, unnatural face, unnatural body, error, (extra fingers), missing fingers, disfigured, deformed, misplaced hair, boring, old, child, man, chibi, huge head, skinny_neck, huge eyes, nude, huge eyes, undetailed faces, undetailed eyes
or
low quality, lowres, bad anatomy, normal quality, worst quality, 3D, render, illustration, drawing, comic, watermark, text, logo, website, bad anatomy, bad hands, unnatural face, unnatural body, glitch, error, (extra fingers:1.2), (missing fingers:1.2), disfigured, deformed, mutated, ugly, boring, bland, deformed pupils, deformed iris, rhombus pupils, misplaced hair, child, (old, man, old man:1.2), chibi, huge head, skinny_neck, nude, huge eyes, undetailed faces, undetailed eyes
NSFW:
add to the negative:
inverted nipples, censored,
remove nude for full nudity. leave it in for more creative clothed nsfw
add to the positive, right after the score_ section
rating_explicit, uncensored,
place core NSFW actions at the top of the prompt, either directly above or below the character,
reinforce NSFW actions below main prompt, if applicable
add nude, naked for more nudity
for the opposite effect, place the above in the negative
add rating_safe in the positive
for manly women and more masculine men, remove 'man' from the negative
depth of field, bokeh: remove these or even place them in the negative if you want the background to be in focus. Note that hand and eye detail will probably suffer.
add perfect teeth for scenes with an open mouth
Hyper:
This model works just fine with a Hyper LoRA at strength 1.0, giving nearly identical results at half the step count: (round up for 13 steps)
trigger word: 无需
if you're using a lot of LoRAs and need space, drop it and set the steps back up to 25
If your results are too saturated, you can load a pony desaturation LoRA:
workflow users can load a saturation node right before the final output (save image) and set the factor to around 0.85. This is lighter weight and leads to a more stable prompt
precise settings for both will vary, as some images will be perfectly fine without it, but others will be very overblown
lighting conditions can affect this also, so keep those in mind
Other useful LoRAs:
Consistency Corrector - SDXL -- corrective/prompt adherence
Midjourney Image Enhancer XL Extreme - SDXL -- corrective enhancements
Pika's Super Realistic Detailed Eyes [SRDE] - SDXL -- enhance eyes and realism -- weight: 0.3
HUBG_Detail+丨More Reasonable Details -- more details and corrective enhancements
there are two HUBG Detail+ models. Unsure which is the proper author. Please leave a message if you find out, and I will update the info/link
Midjourney V6 Style (experimental) -- reasonable MJ-ish style added
let me know if you see any errors or mistakes in this info, or have/discover better methods
