Loop Forge

The template's own example, on five seeds

From MOSS-Image add-ons, tested locally

Model
MOSS-Image SFT (moss_image_sft_int8_convrot.safetensors)
Output
1024×1024
Sampling
25 steps, CFG 2.5, er_sde / simple
Seeds
101, 7, 42, 300, 777
Seed 101
43.0 s, peak VRAM 10.7 GiB
Ran on
RTX 3060, 12GB (local)
Limit

The big title and the knight land, but the small paragraphs wander out of their boxes, and that happened on every seed I tried.

The prompts

Download workflow
Dark fantasy editorial poster in a lush overgrown ruin: a knight in moss-covered plate armour in a heroic kneeling pose, torso upright and head lifted, one hand gripping a massive greatsword planted point-down beside him, the other forearm on his raised knee, a dense bouquet of daisies and pink roses growing out of the open helmet instead of a face, ivy and crumbling columns behind, petals drifting in the air, soft volumetric light from above, deep green and moss palette with muted cream type.

Notes

The boxes are drawn in the template's Create Bounding Boxes node. Each one is a text box (the exact words to spell and how they should look) or an object box (a description), and the MOSS-Image Layout Prompt node turns the caption and the boxes into the layout prompt the model was trained on. The Caption and Boxes tabs are exactly what those two nodes were given.

Boxes 04 and 05, the two small quotes, are drawn on each result. The other seeds use the same workflow with only the seed changed.

Comfy-Org workflow templates (the MOSS-Image templates ship with ComfyUI)

Run it yourself

This test loads no input files. Put the models below in place and drag the workflow onto ComfyUI.

FilePut it inSize
moss_image_sft_int8_convrot.safetensorsmodels/diffusion_models5.4 GB
qwen3.5_9b_w4a8_gptq.safetensorsmodels/text_encoders7.3 GB
flux2-vae_small.safetensorsmodels/vae249.5 MB

No custom nodes. Every node is part of ComfyUI 0.43.1 (frontend 1.57.6, templates 0.11.100).