Clover Image Tiny 🍀

A compact Stable Diffusion 1.4-class family running on Hugging Face ZeroGPU. Choose Regular for text-to-image and style LoRAs, or Inpainting to upload an image and paint exactly the area Clover should replace. Each request produces one image and the queue processes one generation at a time.

Model card · Inpainting model · Regular generation and Inpainting are safety-free by construction so completed images cannot be discarded after inference finishes.

Mode

Regular Inpainting

Create a new image from text with base Clover or a style LoRA.

Model catalog

Regular Text-to-image · base Clover plus three optional style LoRAs

Inpainting Image + white mask · validated HQ 9-channel inpainting model

Prompt — describe the image or masked replacement

Negative prompt

Editable; clear it for the original empty-negative recipe.

Seed

Guidance scale (CFG)

0 20

Style LoRA

Pick a style, then add its trigger word to your prompt.

Base checkpoint — no trigger word needed.

LoRA weight

1.00 uses the selected style as trained; lower or raise it to tune the effect.

0 1.5

Scheduler

Inference steps (Diffusers scheduler)

4-50 recommended; 51-100 is exploratory and slower.

4 100

Width

256 512

Height

256 512

Generated image

Reproducibility metadata

Style LoRAs

MonetTrigger: Monet Style

PointillismTrigger: pointillism painting

Watercolor AnimeTrigger: watercolor anime