<h1 id="fascium-krea-convrot-31-voice-photographic-merge-(int8_convrot-and-int4_convrot)">Fascium KREA ConvRot — 31-Voice Photographic Merge (int8_convrot &amp; int4_convrot)</h1>One checkpoint, thirty-one photographic voices, two quantized editions. Fascium KREA ConvRot is a KREA2-based merge of 31 full-rank photographer LoRAs — portrait, fine-art, editorial and documentary schools — converted with rotation-based quantization (ConvRot) for dramatically lower VRAM while keeping every merged lighting signature intact.<h2 id="what-is-inside">What is inside</h2><ul><li>31 merged photographic voices, grouped into families: soft-window medium-format portraiture (Kazantsev-school muted canvases and 85mm f/1.2 cream), dark-realism chiaroscuro (Zachar-school black voids with a single sculpting key), noir editorial (Kozlov-school dramatic key and fabric texture), soft glamour window light (Archer-school tonal roll-off), intimate window-light realism (Drozhzhin-school), plus cinematic street, analog film-grain documentary and fine-art editorial voices.</li><li>Photographic only — by design. No painterly, engraving or anime styles are encoded in this merge. If you need oil, gouache or copperplate looks, this is not the file; its superpower is the full range of photographed light.</li><li>Genre grammar lives in the prompt, not in the weights. Gothic realism, magical realism, literary character cards and captioned poster work run on top of this photographic base through prompt triggers and composition tokens — the merge supplies honest light, skin and fabric, your prompt supplies the miracle.</li><li>SFW by design. Fully clothed adult subjects, no explicit content, no suggestive posing. NSFW use is outside this model's design and discipline.</li><li>Craft grammar baked in: single-key lighting with deep velvet falloff, prop-as-witness storytelling, muted editorial grades, and reliable Cyrillic display typography for captioned poster work.</li><li>Text encoder protected in BF16 during conversion, so long-prompt adherence and glyph rendering survive quantization on the int8 edition.</li></ul>Files &amp; quantization honesty tableFilePrecisionRoleKnown behaviorfascium_krea_convrot_int8.safetensorsint8_convrotRecommended production filePassed an 18-frame internal stress bench (deep shadows, skin micro-texture at 100% zoom, small props, Cyrillic capitals, multi-layer fabrics) with no visible quantization artifacts versus the BF16 masterfascium_krea_convrot_int4.safetensorsint4_convrotFast draft / preview file~4× VRAM savings; excellent for composition drafts and contact sheets. Known limits: small typography may degrade, smooth gradients can band, micro-texture softens. Not recommended for final captioned postersThe BF16 master is retained privately as the archival source and is not distributed. Quantization performed with the Forge Neo Converter extension (port of the Starnodes Model Converter); ConvRot rotation protects outlier weights during quantization.<h2 id="recommended-settings">Recommended settings</h2><ul><li>Sampler: Euler a · Steps: 8–12 · CFG: 1–3</li><li>Resolution: 1024×1536 (vertical) or 1536×1024 (panoramic)</li><li>For Cyrillic caption prompts, use the int8 file</li></ul><h2 id="example-style-triggers-(photographic-voice-families)">Example style triggers (photographic voice families)</h2><code>Aleksey Kazantsev style</code> · <code>Zachar style</code> · <code>Evgeniy Kozlov style</code> · <code>Sean Archer style</code> · <code>Denis Drozhzhin style</code> · <code>gothic realism photography</code> · <code>magical realism photography</code> · <code>cinematic street portrait</code> · <code>analog film-grain documentary portrait</code> · <code>fine art editorial portrait</code><h2 id="ethics-and-license-notes">Ethics &amp; license notes</h2><ul><li>All subjects are synthetic adults; no real-person likenesses are encoded or intended.</li><li>Living photographers' voices are included as unofficial style studies and homages; respect upstream licenses of any derivative work.</li><li>SFW output only. Please keep gallery posts and derivatives within the same discipline.</li></ul><h2 id="dataset">Dataset</h2>20 original photographic sample prompts ship with this model (see gallery). They cover the voice families, the lighting grammar, and the Cyrillic caption signature — every line is fully photographic, no painterly triggers.

Fascium KREA2 ConvRot — 31-Voice Photographic Merge (int8_convrot & int4_convrot)

fasciumKREA2Convrot31Voice_v30_nvfp4.safetensors

Files with SHA256: 831fbc4dfe06ca62c32ff40eb96808cfc50bc99efbb1ef559c516ac8f2e9681f