Sick Ollie (Krea 2 MXFP8 DiT)Min 10GB VRAM592×888

Civitai #143439252 Original Stack: Sick Ollie Krea 2 + 9 LoRA Sliders

Exact reproduction of Civitai #143439252 on RTX 4070. Chaining 9 sequential LoRA slider adapters on Sick Ollie (Krea 2 MXFP8) with Wan 2.1 VAE and Euler a beta sampling in 8 steps.

Generation Parameters

Samplereuler_ancestral

Schedulerbeta

Steps8

CFG Scale1

Seed385

Recommended VRAM12GB

Positive Prompt

.looking away, biting lip, but body leaning in .cute makeup, pale skin, .black choker, messe long straight brown hair, no bangs. paparazzi shot, harsh direct flash, strong specular highlights, sharp shadow on wall, blown out highlights , Detailed Emotions and Expressions, Large breasts

Engineering Field NotesRTX 4070 Benchmarked

Pixel-exact reproduction of Civitai #143439252: This recipe reproduces the exact stack of the original Civitai creator. It chains 9 sequential LoRA slider adapters on Sick Ollie (Krea 2 MXFP8 base): Detailer (1.0), Areola Slider (-10.0), Refusal Reduction (1.0), Breasts Slider (+8.5), Real/Fake Breast Slider (-2.6), Detailed Emotions (1.0), Ass and Thighs (+4.0), Nose Slider (-2.0), and SNOFS v1.4 (1.0). The generation completes in just 8 steps at CFG 1.0 using Euler Ancestral with beta schedule. The latent is decoded with Wan 2.1 16-channel VAE. Running this pipeline on an RTX 4070 consumes ~10.4 GB VRAM with 0 OOM errors. Compare this with our Qwen-Image-2.1 clean-room distillation recipe which produces the same paparazzi flash aesthetic in a single clean pass without needing 9 separate adapter layers.

Hardware Footprint & Timings

RTX 4070 (12GB VRAM): ~10.4 GB peak VRAM. Generation time: 38 seconds (8 steps at 592x888 across 9 sequential LoRA hooks). Zero CUDA out-of-memory errors on consumer desktop hardware.