⚠️ UPDATE COMFYUI-PIXAROMA TO v1.4.175 OR NEWER BEFORE LOADING ⚠️

Older versions break the Set/Get nodes on load or refresh.

A fire-and-forget wildcard workflow for MOSS-Image. Hit queue and it rolls a random location, time/weather and lighting for every image, so you get huge variety without writing prompts. 🥳

---

Models needed (~12.1 GB total download):

• moss_image_sft_int8_convrot.safetensors (5.0 GB), goes in models/diffusion_models

• qwen3.5_9b_w4a8_gptq.safetensors (6.8 GB), goes in models/text_encoders

• flux2-vae_small.safetensors (0.2 GB), goes in models/vae

All three are on the Comfy-Org MOSS-Image page.

VRAM: tested on a 16 GB RTX 4080. Around 15s per image at 768x1344 - using the below settings.

---

Recommended settings: sampler er_sde, scheduler sgm_uniform, 30 steps, CFG 4, 768x1344.

Pairs with: Animora Style for MOSS (style LoRA at 1.0 STR). (link)

Required custom nodes: ComfyUI-Pixaroma (v1.4.175+) and Impact Pack (wildcard processor).

---

Tips:

• Add your own character or LoRA in the Character & Pose box.

• If a back view looks scrunched, add long torso, natural proportions to the positive prompt and lower the style LoRA to about 0.7.

• Text in images (signs, name tags) will be gibberish, which is normal.