ComfyPilotComfyPilot
模型库 / 整模型 / mossMODE
mossMODE

mossMODE

3059655

整模型 17 次下载 2 喜欢 更新于 2026-11-23

底模:MOSS Image

下载模型 在源站打开

文件清单 线路:原始(huggingface.co) · 切换

下载后放进对应的 ComfyUI 目录。models/ 指 ComfyUI 的模型根目录。

权重 mossMODE_v1_12_step_turbo.safetensors 542 MB 下载
权重 mossMODE_v1_int8_convrot.safetensors 5 GB 下载
权重 mossMODE_v1_bf16.safetensors 9.5 GB 下载

说明

mossMODE

作者:anyMODE

mossMODE v1 is a tuned MOSS-Image checkpoint built for natural photography, expressive illustration/painterly styles, and strong prompt adherence without the over-processed look. Available in full BF16 and a fast, low-VRAM INT8 version. Pros • Natural photographic look: Rich lighting, authentic skin texture, and balanced film-like contrast out of the box. • Versatile across styles: Handles candid photography, studio editorial, macro/product, landscapes, oil/watercolor illustration, and typography cleanly. • High-resolution & Turbo friendly: Scales cleanly to 2.0 MP and pairs seamlessly with 12-step Turbo LoRAs (CFG 1.0). • Two precision options: Full BF16 for maximum fidelity, or INT8 (~5 GB) for fast generation on 8–16 GB GPUs with minimal quality loss. Cons • Anatomy & hands: Hand, finger, and complex pose artifacts can still occur occasionally, though at a noticeably lower rate than base MOSS. • Complex object interactions: Fine interactions (pouring liquids, crowded multi-person scenes, small background text) may occasionally need a seed reroll. • Turbo trade-off: Using the 12-step Turbo LoRA trades a small amount of micro-detail and pose stability for ~2.5× speed. Recommended Settings • Standard: 28 steps · CFG 3.0 · euler / simple • Turbo (with 12-step Turbo LoRA @ 0.9): 12 steps · CFG 1.0 · euler / simple • Resolution: 1.0 – 2.0 MP (multiples of 16) • Text Encoder / VAE: Qwen3.5-9B (moss_image) + FLUX.2 VAE

示例图