<p>Kreros 1.0 is a photorealistic merge built on Krea 2 Turbo (INT8 ConvRot), blended with amateur photography and realism LoRAs. It specializes in raw, candid, amateur-style photos — the kind of look you get from a smartphone shot in a bedroom, bathroom, or party: imperfect framing, natural window or flash lighting, real skin texture with blemishes and tan lines, and an overall unpolished, authentic feel. NSFW capable. Recommended settings: 8 steps, CFG 1.0, Euler / simple, 1K-2K resolution. Works great with vertical (9:16) prompts. Pair it with the standard Krea 2 text encoder (Qwen3VL 4B) and Qwen Image VAE. Tips: add words like 'candid', 'amateur photo', 'smartphone photo', 'shot on iPhone', 'flash photo at night' or 'analog film' to the prompt for stronger authenticity; describe lighting and camera angle explicitly for best results. Trained words: not required. This is an experimental personal merge shared for the community — no commercial restrictions on outputs. Base model license: Krea AI Community License. Merges and derivatives are allowed, credit is appreciated but not required.</p>

Kreros 1.0

<p>Kreros 1.0 w4a8 is a pre-quantized INT4 (weights) / INT8 (activations) build of Kreros 1.0 — same weights baked into a single 8.17 GB safetensors. Load it with OTUNetLoaderW8A8 (quantization_mode int4_mixed, int4_mixed_ratio 0.25) for dynamic INT4, or with a standard UNETLoader using the baked weights as-is. WHY THIS VERSION EXISTS: 8.17 GB vs 12.57 GB INT8 (-35%) — fits 12 GB cards or leaves headroom on 16 GB for CLIP (Qwen3VL 4B FP8) + VAE. Same prompts and aesthetic as INT8, no re-learning. Baked weights = faster load, no on-the-fly quant pass. TRADE-OFFS VS INT8: INT4 loses some fine detail — slightly softer micro-textures (skin pores, hair strands, fabric weave) at 100% zoom. Extreme resolutions / tiny batches degrade more; int4_mixed keeps sensitive layers at higher precision (ratio 0.25) — raise the ratio if you push it. Not bit-identical: same seed gives close but not pixel-identical results. INT8 stays the reference when VRAM is not a constraint. WHEN TO USE: 12-16 GB GPUs (RTX 4070, 5060 Ti 16GB, 3060 12GB); long batches where headroom matters more than last-percent fidelity; quick drafts in the Kreros look. RECOMMENDED SETTINGS (same as INT8): er_sde | 8 steps | CFG 1.0 | 1776x2368 native | CLIP qwen3vl_4b_fp8_scaled | VAE qwen_image_vae. All example images were generated with this exact file; the 3 shots via standard UNETLoader confirm the baked weights run outside the OTU loader too.</p>

Files with SHA256: 08d34518789431602edbaa3fd0f4f3df64a784aa1186b02ee6550ae998db6f47