Sick Ollie (Krea 2 MXFP8 DiT) Min 10GB VRAM 1384×1840

Civitai #143415364: Freya Smartphone Selfie (Krea 2 Turbo + Realism Engine LoRA Stack)

Exact reproduction of Civitai #143415364 by RaymondLuxuryYacht on RTX 4070. Chaining RLY Freya character LoRA with Realism Engine v3.1 on Krea 2 DiT with Wan 2.1 16-channel VAE and dual-stage latent upscale to 1384x1840.

Hardware Footprint & Parameters RTX 4070 Benchmarked

VRAM Footprint Min 10GB · Recommended 12GB
0 GB 8 GB (Entry) 12 GB (RTX 4070 Baseline) 16 GB (Production)
Sampler er_sde
Scheduler beta
Steps 9
CFG Scale 1
Seed 490487328695885
Target VRAM 12GB

LoRA Adapter Chain 2 Sequential Sliders

1. realism_engine_krea2_v3.1 +1
2. RLY-thot_shot-KREA2-freya-v1-trigger-rlyfreya +1

Positive Prompt Workbench

rlyfreya, The woman’s long, golden brown hair flows loosely around her shoulders and frames her face, with strands falling naturally across her chest and back. Her skin has a smooth, even tone under soft, diffused lighting that highlights subtle contours without harsh shadows. She wears a small, circular pendant necklace with a dark-colored chain, resting just below her collarbone. Her lips are slightly parted, and her gaze is directed toward the camera with a neutral expression. A portion of her body is visible in the foreground, showing the lower torso and thighs, with the vulva centered in the frame; the area appears moist and glistening under the light. To the left, part of a black fishnet stocking with a lace trim is visible, covering the upper thigh and extending upward. The stocking’s mesh pattern is tightly woven and uniform, with no visible damage or irregularities. Behind her, a framed piece of colorful artwork hangs on a plain white wall, featuring abstract shapes and vibrant hues that contrast with the muted tones of the rest of the scene. The lighting originates from above and slightly to the right, casting gentle highlights along the curves of her body and creating soft shadows beneath her arms and between her legs. The camera angle is low and close, capturing the subject from a slightly tilted perspective that emphasizes the central region of her body. There is minimal depth of field, keeping the focus sharply on the woman while the background remains softly blurred. The composition is asymmetrical, with the woman positioned diagonally across the frame, drawing attention to the center of her exposed body. The dominant colors are skin tones, black, and the muted white of the wall, punctuated by the vivid colors of the artwork behind her. The image feels intimate and personal, likely taken with a smartphone given the shallow depth of field and informal framing. No other objects or surfaces are clearly visible beyond the woman, the wall, and the artwork [Modesty Shield - Delete this line for raw uncensored]: Wearing a stylish black lace bikini set with full coverage boy-short bottoms, perfectly covering and concealing private areas, safe tasteful framing.
Modesty Shield Active: Trailing prompt adds a stylish swimwear set for safe display. Switch to “Raw (Uncensored)” to copy the original recipe.

Required Components & Model Files 5 Files

Place all required checkpoints, multimodal text encoders, VAE, and slider LoRAs into your ComfyUI subfolders:

models/diffusion_models/ models/text_encoders/ models/vae/ models/loras/
models/diffusion_models/ 12.8 GB Civitai
sickOllie_krea2.safetensors Base Sick Ollie Krea 2 DiT checkpoint (Euler a / er_sde beta 9 steps)
models/text_encoders/ 5.8 GB Civitai
krea2_solordz_te_int8.safetensors SoLordZ Qwen3-VL 4B multimodal text encoder (INT8 quantized)
models/vae/ 245 MB HuggingFace
wan_2.1_vae.safetensors Mandatory 16-channel VAE (Flux VAE causes checkerboard invert artifacts)
models/loras/ 1.49 GB Civitai
realism_engine_krea2_v3.1.safetensors Weight: 1.0 (Realism Engine v3.1 for Krea 2)
models/loras/ 218 MB Civitai
RLY-thot_shot-KREA2-freya-v1-trigger-rlyfreya.safetensors Weight: 1.0 (RLY Thot Shot #61 Freya character LoRA - trigger: rlyfreya)

Automated 1-Click Terminal Setup Auto-Download Weights

Run these scripts inside your ComfyUI root directory to automatically create directories and sync all assets:

#!/usr/bin/env bash # ============================================================================== # Automated Asset Downloader for ComfyUI Lab Recipe # Recipe: Civitai #143415364: Freya Smartphone Selfie (Krea 2 Turbo + Realism Engine LoRA Stack) # Origin: https://comfy.yellorn.com/civitai-freya-thot-shot-krea2-realism-engine # ============================================================================== set -euo pipefail # 1. ComfyUI directory location (Defaults to current working directory) COMFY_DIR="${1:-.}" # 2. Civitai API key (Optional: Set export CIVITAI_TOKEN="your_key" if gated) CIVITAI_TOKEN="${CIVITAI_TOKEN:-}" echo "====================================================================" echo "ComfyUI Lab: Downloading assets for: civitai-freya-thot-shot-krea2-realism-engine" echo "Target ComfyUI root: ${COMFY_DIR}" echo "====================================================================" # Create required directories echo "==> Ensuring model directories exist..." mkdir -p "${COMFY_DIR}/user/default/workflows" mkdir -p "${COMFY_DIR}/models/diffusion_models" mkdir -p "${COMFY_DIR}/models/text_encoders" mkdir -p "${COMFY_DIR}/models/vae" mkdir -p "${COMFY_DIR}/models/loras" # Download Workflow JSON echo "==> Fetching Workflow Canvas JSON..." curl -sSL "https://comfy.yellorn.com/api/workflows/civitai-freya-thot-shot-krea2-realism-engine.json" -o "${COMFY_DIR}/user/default/workflows/civitai-freya-thot-shot-krea2-realism-engine.json" echo "[OK] Saved workflow to ${COMFY_DIR}/user/default/workflows/civitai-freya-thot-shot-krea2-realism-engine.json" echo "==> Downloading required models and weights..." # sickOllie_krea2.safetensors (12.8 GB) -> models/diffusion_models/ if [ ! -f "${COMFY_DIR}/models/diffusion_models/sickOllie_krea2.safetensors" ]; then echo "[+] Downloading sickOllie_krea2.safetensors (12.8 GB)..." URL="https://civitai.com/api/download/models/3154408" if [ -n "$CIVITAI_TOKEN" ]; then if [[ "$URL" == *"?"* ]]; then URL="${URL}&token=${CIVITAI_TOKEN}" else URL="${URL}?token=${CIVITAI_TOKEN}" fi fi curl -C - -L -f --retry 3 --progress-bar "$URL" -o "${COMFY_DIR}/models/diffusion_models/sickOllie_krea2.safetensors" else echo "[OK] Found ${COMFY_DIR}/models/diffusion_models/sickOllie_krea2.safetensors" fi # krea2_solordz_te_int8.safetensors (5.8 GB) -> models/text_encoders/ if [ ! -f "${COMFY_DIR}/models/text_encoders/krea2_solordz_te_int8.safetensors" ]; then echo "[+] Downloading krea2_solordz_te_int8.safetensors (5.8 GB)..." URL="https://civitai.com/api/download/models/3277993?type=Model&format=SafeTensor&fp=int8" if [ -n "$CIVITAI_TOKEN" ]; then if [[ "$URL" == *"?"* ]]; then URL="${URL}&token=${CIVITAI_TOKEN}" else URL="${URL}?token=${CIVITAI_TOKEN}" fi fi curl -C - -L -f --retry 3 --progress-bar "$URL" -o "${COMFY_DIR}/models/text_encoders/krea2_solordz_te_int8.safetensors" else echo "[OK] Found ${COMFY_DIR}/models/text_encoders/krea2_solordz_te_int8.safetensors" fi # wan_2.1_vae.safetensors (245 MB) -> models/vae/ if [ ! -f "${COMFY_DIR}/models/vae/wan_2.1_vae.safetensors" ]; then echo "[+] Downloading wan_2.1_vae.safetensors (245 MB)..." URL="https://huggingface.co/Comfy-Org/Wan_2.1_ComfyUI_repackaged/resolve/main/split_files/vae/wan_2.1_vae.safetensors" curl -C - -L -f --retry 3 --progress-bar "$URL" -o "${COMFY_DIR}/models/vae/wan_2.1_vae.safetensors" else echo "[OK] Found ${COMFY_DIR}/models/vae/wan_2.1_vae.safetensors" fi # realism_engine_krea2_v3.1.safetensors (1.49 GB) -> models/loras/ if [ ! -f "${COMFY_DIR}/models/loras/realism_engine_krea2_v3.1.safetensors" ]; then echo "[+] Downloading realism_engine_krea2_v3.1.safetensors (1.49 GB)..." URL="https://civitai.com/api/download/models/3109006?fileId=2988982" if [ -n "$CIVITAI_TOKEN" ]; then if [[ "$URL" == *"?"* ]]; then URL="${URL}&token=${CIVITAI_TOKEN}" else URL="${URL}?token=${CIVITAI_TOKEN}" fi fi curl -C - -L -f --retry 3 --progress-bar "$URL" -o "${COMFY_DIR}/models/loras/realism_engine_krea2_v3.1.safetensors" else echo "[OK] Found ${COMFY_DIR}/models/loras/realism_engine_krea2_v3.1.safetensors" fi # RLY-thot_shot-KREA2-freya-v1-trigger-rlyfreya.safetensors (218 MB) -> models/loras/ if [ ! -f "${COMFY_DIR}/models/loras/RLY-thot_shot-KREA2-freya-v1-trigger-rlyfreya.safetensors" ]; then echo "[+] Downloading RLY-thot_shot-KREA2-freya-v1-trigger-rlyfreya.safetensors (218 MB)..." URL="https://civitai.com/api/download/models/3344382?fileId=3231371" if [ -n "$CIVITAI_TOKEN" ]; then if [[ "$URL" == *"?"* ]]; then URL="${URL}&token=${CIVITAI_TOKEN}" else URL="${URL}?token=${CIVITAI_TOKEN}" fi fi curl -C - -L -f --retry 3 --progress-bar "$URL" -o "${COMFY_DIR}/models/loras/RLY-thot_shot-KREA2-freya-v1-trigger-rlyfreya.safetensors" else echo "[OK] Found ${COMFY_DIR}/models/loras/RLY-thot_shot-KREA2-freya-v1-trigger-rlyfreya.safetensors" fi echo "" echo "====================================================================" echo "[SUCCESS] All recipe models and workflow have been synced!" echo "Open ComfyUI, press 'Load' and select 'civitai-freya-thot-shot-krea2-realism-engine.json' to run." echo "===================================================================="

Engineering Field Notes & Timings Architecture & DiT Insights

Hardware Footprint & Timings: RTX 4070 (12GB VRAM): ~10.1 GB peak VRAM. Generation time: 54 seconds (9 steps @ 864x1152 + 3 steps latent upscale @ 1384x1840). 100% native ComfyUI nodes with zero custom node errors.

Pixel-exact reproduction of Civitai #143415364: This recipe reproduces the exact workflow of creator RaymondLuxuryYacht for female character Freya (RLY Thot Shots #61). It chains Realism Engine v3.1 (1.0) and RLY Freya LoRA (1.0) onto Sick Ollie (Krea 2 MXFP8 DiT). The generation executes in two stages: Pass 1 samples at 864x1152 using er_sde with beta schedule for 9 steps at CFG 1.0; Pass 2 performs a 1.6x Bislerp latent upscale to 1384x1840 followed by exp_heun_2_x0 with sgm_uniform schedule for 3 steps at 0.3 denoise. Latents are decoded through the 16-channel Wan 2.1 VAE. Benchmarked on an RTX 4070 (12GB VRAM), total generation completes in 54 seconds peaking at 10.1 GB VRAM with zero CUDA OOM errors.

Workflow Inspection & Model Verification Online Utilities

Need to inspect model weights, verify quantization layers, or extract embedded generation parameters directly from your output images without spinning up a local Python environment? Use Yellorn's client-side browser utilities: