Civitai #144401284: Dahlia Editorial Studio Portrait - Krea 2 Turbo

Full Resolution

Generation Specs RTX 4070

Sampler euler

Scheduler simple

Steps 8

CFG Scale 1

Seed 503512476372652

Target VRAM 16GB

VRAM 12GB - 16GB

GPU Compatibility

Optimal Performance

Runs entirely in VRAM with no memory swapping.

Krea 2 Turbo (FP8 Scaled DiT) Civitai #144401284 Min 12GB VRAM 832×1248 RTX 4070 Verified

Dahlia Editorial Studio Portrait - Krea 2 Turbo

High-resolution studio portrait of Dahlia deconstructed from Civitai #144401284 by TeeKay using Krea 2 Turbo FP8, Wan 2.1 16-channel VAE, and a 3-LoRA character stack.

Blueprint Summary RTX 4070 Verified

Reproducible ComfyUI workflow for Civitai #144401284: Dahlia Editorial Studio Portrait - Krea 2 Turbo using Krea 2 Turbo (FP8 Scaled DiT) at 832×1248 resolution. Requires minimum 12GB VRAM with sampler euler and scheduler simple (8 steps). Includes 1-click terminal model sync and canvas JSON graph.

Pipeline Flow 12 Nodes

Open in Resolver

01 Load Models

UNETLoader

6 loaders (DiT, CLIP, VAE)

02 Adapters

LoraLoader

3 adapters

03 Conditioning

CLIPTextEncode

2 prompt encodings

04 KSampler

KSampler

DiT latent denoising

05 Decode & Save

VAEDecode

Latent to pixel space

Execution DAG Topology Interactive Visualizer

Drag to pan · Scroll to zoom · Hover wires

Model & Asset Setup Setup Script

Run in your ComfyUI root:

curl -fsSL https://decomfy.com/api/scripts/civitai-dahlia-blonde-studio-krea2.sh | bash

Positive Prompt

The woman’s long, glossy blonde hair cascades loosely around her shoulders, strands catching soft ambient light that highlights their natural sheen and texture. Her skin carries a smooth, even tone with subtle natural contours, framed by meticulously applied makeup: sharp, winged eyeliner in a dark hue defines her eyes, which gaze directly into the camera with an intense, composed expression. Her lips are subtly tinted, maintaining a neutral-to-pink hue, contributing to a poised aesthetic. She wears a snug tailored white graphic t-shirt. The garment fits elegantly, emphasizing clean modern silhouette. A single dangling earring, metallic and delicate, rests near her earlobe, its surface catching a faint glint from the directional lighting. The background is a plain, off-white studio wall, offering no distractions and allowing the subject to dominate the frame. The foreground is minimal, with only the woman’s upper body and shoulder area visible, creating a tight, intimate portrait composition. The lighting is soft diffused studio key light, casting gentle shadows that contour her features without harshness. The camera angle is slightly elevated, capturing her from the chest up, with a shallow depth of field that blurs the edges of the frame, drawing focus entirely to her face and eyes. The dominant color palette is clean monochrome with warm natural skin tones. High-resolution editorial portrait photography, crisp lines, exquisite facial details, photorealistic 8k masterpiece.

Negative Prompt

unclothed, disrobed, revealing clothes, lowres, blurry, bad anatomy, deformed hands, extra fingers, low quality, artifacts, distorted proportions, text, watermark, signature

Prompt Customization: This workflow is pre-wired and calibrated. Swap this prompt with your own character, subject, or style without breaking the pipeline.

LoRA Adapter Stack 3 Adapters

RTX 4070 Calibrated Weights

01 Character_Mix_DahliaMixer_Kr2.safetensors +1.1

02 Kr2_snofs_krea_v1_4.safetensors +0.5

03 Kr2_Emotions V1.safetensors +0.5

Required Models 6 Models

Required Storage: 17.9 GB (6 models · DiT/Base: 12.6 GB · Text Encoder: 4.5 GB · LoRAs: 684 MB · VAE: 253 MB)

models/diffusion_models/ 12.56 GB Civitai

krea2_turbo_fp8_scaled.safetensors Official Krea 2 Turbo FP8 Scaled DiT (Model #2732656, Version #3072332)

Get

models/text_encoders/ 4.45 GB Civitai

krea2_solordz_te_int8.safetensors Solordz INT8 Text Encoder for Krea 2 (Offloaded to CPU for 12GB VRAM headroom)

Get

models/vae/ 253 MB HuggingFace

wan_2.1_vae.safetensors Wan 2.1 16-Channel Fast VAE Decoder (3s decode for Krea 2 latents)

Get

models/loras/ 228 MB Civitai

Character_Mix_DahliaMixer_Kr2.safetensors Teekay's Custom Characters: Dahlia v2.0 Krea 2 Character Mixer (Weight: 1.1)

Get

models/loras/ 228 MB Civitai

Kr2_snofs_krea_v1_4.safetensors SNOFS Realism Refinement v1.4 for Krea 2 (Weight: 0.5)

Get

models/loras/ 228 MB Civitai

Kr2_Emotions V1.safetensors Emotions V1 Expression Micro-tuning for Krea 2 (Weight: 0.5)

Get

Field Notes RTX 4070 Benchmark

Benchmark: Rendered on NVIDIA GeForce RTX 4070 (12GB VRAM, CUDA 12.4): 8 steps euler simple @ CFG 1.0, 832x1248, 11.6 GB peak VRAM, Wan 2.1 16-channel VAE fast decode (3s).

1. CIVITAI PROVENANCE & CHARACTER TOPOLOGY: Forensic extraction of Civitai image #144401284 by creator TeeKay revealed a native Krea 2 Turbo architecture (ModelVersion #3072332) utilizing an 8-step Euler Simple schedule at CFG 1.0. The facial likeness and stylistic fidelity rely on a triple-LoRA stack: TeeKay's custom Dahlia character mixer (Character_Mix_DahliaMixer_Kr2 at strength 1.1, ModelVersion #3375434), SNOFS v1.4 realism refinement (Kr2_snofs_krea_v1_4 at strength 0.5, ModelVersion #3290120), and Emotions V1 expression micro-tuning (Kr2_Emotions V1 at strength 0.5, ModelVersion #3075498). 2. WHITE-WEAPON SFW RE-ENGINEERING & COLOR PROFILE: Under the White-Weapon Protocol ('Tẩy Trắng'), positive conditioning anchors Dahlia in a snug tailored white graphic t-shirt and off-white minimalist studio backdrop. The composition preserves the intimate elevated portrait angle, delicate metallic earring glint, and glossy blonde hair cascade while ensuring 100% Google SafeSearch compliance. ConditioningZeroOut provides stable latent guidance without semantic dilution, generating photorealistic skin micro-textures and winged eyeliner definition. 3. HARDWARE BENCHMARKS & 12GB VRAM MEMORY ARCHITECTURE: Rendered on external GPU host wingpu (NVIDIA GeForce RTX 4070 12GB VRAM, CUDA 12.4). Because Krea 2 Turbo is a 13.5 GB diffusion transformer, loading both the 4.9 GB Qwen3-VL text encoder and the DiT simultaneously on GPU would force 3.5 GB over PCIe per step. By offloading the Solordz INT8 text encoder to host system RAM (34 GB available) via device: 'cpu', 100% of usable GPU VRAM remains dedicated to the DiT in float8_e4m3fn format. Replacing the multi-pass Qwen image VAE with the compatible Wan 2.1 16-channel VAE reduces decoding time from several minutes to just 3 seconds.

Frequently Asked Questions FAQ

What GPU and VRAM are required to run Civitai #144401284: Dahlia Editorial Studio Portrait - Krea 2 Turbo?

This workflow requires a minimum of 12GB VRAM (recommended 16GB VRAM). Tested and verified on NVIDIA GeForce RTX 4070 (12GB VRAM) at 832x1248 resolution.

How do I resolve missing custom nodes for this workflow?

You can drop the workflow JSON into our client-side Missing Node Auto-Resolver at https://decomfy.com/resolve/ to detect missing nodes and generate install commands, or run the 1-click terminal setup script provided below.

Which VAE and conditioning text encoders are required for Krea 2 DiT?

Krea 2 DiT single-stream models strictly require the 16-channel Wan2.1 VAE and Qwen3-VL text encoder. Using standard Flux or SDXL VAE causes inverted checkerboard artifacts. Also pair with RescaleCFG for high-contrast portrait clarity.