🌀 Visual Regional Prompting – Multi-Model Workflows, Native LoRA Hooks & LLLite

Experimental regional LoRA support for the built-in BV Regional Anima Conditioning node.This workflow demonstrates two independently assigned character LoRAs:<ul><li>Nyamena in the left region</li><li>Myaley in the right region</li><li>LoRA strength: <code>0.8</code> each</li><li>Approximately <code>13%</code> interaction overlap</li><li>Fixed seed and reproducible sampler settings</li></ul>The characters face each other and hold hands inside the shared region while their colors, fur and identities remain locally distinct without obvious LoRA bleeding.Connect each external <code>LORA_STACK</code> through BV Named LoRA Stack, assign it to the matching region in BV Regional Prompt, and connect both <code>lora_registry</code> and <code>lora_bindings</code> to BV Regional Anima Conditioning.Regional LoRAs use additional masked model passes and may substantially increase generation time, VRAM usage and system-memory consumption.This feature remains experimental. Results depend on the model, LoRA training, prompt compatibility, strength, seed and region geometry. The included LoRAs come from the same series and creator, making this a favorable validation case rather than a guarantee for arbitrary LoRA combinations.Update BV Node Pack, restart ComfyUI and hard-refresh the browser before loading the workflow.

VisualRegionalPromptingMulti_animaLoraAttnEXP.json

Anima Lora Attn (EXP)

Krea 2 Lora Attn (EXP)

Anima Lora Native (EXP)

Krea 2 (Experimental)

Flux.2 klein 9B

ZImage Turbo

Illustrious/Pony/SDXL

Anima

Files with SHA256: 75d82be251b0bc8906fa271ca216e8f50a27018c419526e58e90fe27179f4812