Vermilion Venus style for Anima preview 3 checkpoint!
Activation tag:
@VermilionVenus
Other tags:
thick lips, huge breasts
optional:
narrow waist
Known characters:
Ciri, white hair, grey hair, scar across eye, white shirt, brown corset, leather pants, knee boots
Anna henrietta, yellow corset, yellow dress, earrings, bead necklace, long hair, forehead, orange hair, green eyes, freckles, freckles, crown
Sylvia Anna, black shirt, shoulder armor, cleavage cutout, black pants, futanari, large penis, short hair, black hair
SETTINGS FOR ANIMA PREV 3:
Short guide:
Same aesthetic settings (masterpiece, best quality, good quality, highres, absurdres, worst quality, low quality, bad quality, lowres, extra digits in negative.) as for V-pred but you don't need "very awa" in positive and "worst aesthetic" in negative because these tags were exclusively for NoobAI V-pred.
Also don't use "score_". Anima supports these tags but they make you pics AI-slop like Pony because score-based aesthetic tags can cause overcontrast.
Generations settings:
Euler A / ER SDE for txt2img, Euler / ER SDE for highres fix (euler a can break backgrounds in highres fix stage so I prefer euler highres fix. If you don't have option to choose in forge / forge neo go to settings and choose "[Hires. fix]: Show checkpoint, sampler, scheduler, and cfg options"
Also I prefer "FlowMatchEulerDiscrete" scheduler instead of "SGM Uniform".
CFG 3.5. You can try CFG 4-5 but personally I like visual of CFG 3 - 3.5 more.
30-40 steps because Anima is more gluttonous than SDXL and 10-20 highres steps, 0.25 - 0.4 denoising for highres (depends on scene, sometimes txt2img pic is already masterpiece, lol).
Adetailer - use very high pixel padding - about 176-232 pixels instead of 32 pixels because Anima understands context much better. Denoising of adetailer - 0.25-0.3.
My preview can help!
Description
Anima prev 3 version.
FAQ
Comments (6)
the lora is good , but i think characters are too overbaked, the crown from henrietta slips too much
It can be so because dataset is very small. Usually I use 150-200+ for styles but for Venus I barely collected 40 pics :/ When dataset is small details of each character in dataset have more weights in lora and can be created randomly. Same with poses. But putting "crown, tiara" in negative can help :3
yeah i get that, i just dont fun of the negative , the more u know , good job anyway!
Nice LORA!
๐๐๐ป
The first 3 & 4 post - The Fiona from "Shrek".














