RedCraft 4.0 - Intimate Editorial Lookbook (K2T DUAL)

1. CIVITAI REMIX & RESOURCE DECOMPOSITION: Ground-truth verification confirms that this image relies strictly on a SINGLE fine-tuned model resource: RedCraft | Hybrid H3 & Krea2T DUAL MASO W6A8 (version 赤佬 4.0 受虐版 K2T DUAL, Civitai Model #958009, Version #3311198, File #3196558). There are zero LoRAs attached in the generation graph. The warm, tactile 35mm film texture, nuanced skin subsurface scattering, and balanced macro depth-of-field are native characteristics trained directly into this hybrid checkpoint. 2. WORKFLOW REUSE & REPRODUCTION STANDARD: When remixing or generating locally with RedCraft 4.0, no external LoRA adapters are required. The pipeline connects UNETLoader (redcraftHybridH3Krea2t_40K2TDUAL_3196558.safetensors, fp8_e4m3fn) directly to Qwen3-VL 4B text encoder and Qwen Image VAE. Latent sampling runs via euler with simple scheduler across 8 steps at CFG 1.0, decoded seamlessly through 16-channel VAE, and cleanly scaled by 2.0x bicubic to 1888x3328. 3. HARDWARE BENCHMARKS & CONSUMER GPU FOOTPRINT: Benchmarked on NVIDIA GeForce RTX 4070 (12GB VRAM, CUDA 12.4). Because zero runtime LoRA patches are applied to the quantized weights, the single checkpoint runs natively at peak speed (~5.8s per step) consuming 11.85 GB VRAM without lowvram layer offload thrashing.