Fascium KREA ART ConvRot — 99-Voice Painterly Merge (int8_convrot & int4_convrot)

One checkpoint, ninety-nine painterly voices, two quantized editions. Fascium KREA ART ConvRot is a KREA2-based merge of 99 full-rank artist LoRAs — oil masters, engravers, illustrators, poster artists and anime-production styles — converted with rotation-based quantization (ConvRot) so the merged brushwork survives at a fraction of the VRAM.

What is inside

• 99 merged artistic voices, grouped into families: Old Master oil glaze (Rembrandt, Velázquez, Vermeer grades), Baroque and Spanish black-on-black, Early Renaissance tempera (Botticelli grade), Impressionist broken brushwork (Renoir grade), Pre-Raphaelite painterly detail (Waterhouse grade), Symbolist gold ornament (Klimt grade), Art Nouveau sinuous line (Mucha grade), Russian realist and folk illustration (Repin, Bilibin grades), mid-century American conceptual illustration (Anderson-grade gouache), copperplate engraving (Piranesi, Dürer grades), poster lithography (Toulouse-Lautrec grade), atmospheric Romantic gradient (Turner grade), and studio-grade anime illustration.

• Painterly only — by design. No photographic voices are encoded here; the photographic sibling is Fascium KREA ConvRot. If you need photographed light, use that file; this one supplies painted light.

• SFW by design. Fully clothed adult subjects, no explicit content, no suggestive posing.

• Craft grammar baked in: single-key lighting translated into paint, prop-as-witness storytelling, muted school-accurate palettes, and reliable Cyrillic display typography for captioned poster work on textured grounds.

• Text encoder protected in BF16 during conversion, so long-prompt adherence and glyph rendering survive quantization on both editions.

Files & quantization honesty table

File

Precision

Role

Known behavior

fascium_krea_art_convrot_int8.safetensors

int8_convrot

Recommended for final output

Best brushstroke coherence, cross-hatch density, fine linework, small typography, smooth sky gradients

fascium_krea_art_convrot_int4.safetensors

int4_convrot

Fast drafts & previews

~4× VRAM savings; excellent for composition blocking and contact sheets. Known limits: fine engraving linework may soften, small Cyrillic typography may degrade, smooth gradients can band

BF16 master

bf16

Private archival source

Not distributed

Quantization performed with the Forge Neo Converter extension (port of the Starnodes Model Converter); ConvRot rotation protects outlier weights during quantization.

Which file to choose

• Final gallery posts, prints, captioned posters: use int8

• Quick composition tests, style exploration, batch previews: use int4 for speed

• Engraving-heavy prompts (Piranesi, Dürer): int8 strongly recommended — fine burin lines are the first to soften at 4 bits

• Anime prompts: both editions perform well; int4 is a viable production choice here

Recommended settings

• Sampler: Euler a · Steps: 8–12 · CFG: 1–3

• Resolution: 1024×1536 (vertical plate) or 1536×1024 (panoramic veduta)

• For Cyrillic caption prompts, keep steps at 10–12 for glyph stability on textured grounds

Example style triggers (voice families)

in the manner of Rembrandt · in the style of Diego Velázquez · in the manner of Vermeer · in the manner of Sandro Botticelli · in the manner of Pierre-Auguste Renoir · in the style of John William Waterhouse · in the style of Gustav Klimt · in the manner of Alphonse Mucha · in the style of Ilya Repin · in the manner of Ivan Bilibin · Harry Anderson style · Piranesi style · in the manner of Albrecht Dürer · in the style of Henri de Toulouse-Lautrec · in the style of J.M.W. Turner · in the style of John Singer Sargent · in the style of Edward Hopper · studio-grade anime illustration

Ethics & license notes

• Public-domain masters are referenced respectfully; living artists' voices are included as unofficial style studies and homages. Respect upstream licenses of any derivative work.

• All subjects are synthetic adults; no real-person likenesses are encoded or intended.

• SFW output only. Please keep gallery posts and derivatives within the same discipline.

Dataset

20 original painterly stress prompts ship with this model (see gallery). Each line names the school, the stress element, and — where captioned — the reinforced Cyrillic typography anchor.