How to Create Stunning Clay‑Style Images with Stable Diffusion

Social platforms are being swept by a new creative wave called “clay AI,” a quirky yet adorable style that has become a hot topic across hashtags like #clay and #myclayworld.

Generating Clay‑Style Images with Stable Diffusion

Three large models—ReminiClay, Ugly‑Cute Internet Clay Filter, and Internet Clay Filter 1.5—can directly produce clay‑style pictures.

Example Using ReminiClay

Prompt (positive): A penguin walks on the grass, high quality, masterpiece, rich detail, Clay

Negative prompt: ng_deepnegative_v1_75t, (badhandv4:1.2), (worst quality:2), (low quality:2), (normal quality:2)

Sampler: DPM++ SDE Karras

Steps: 30

Resolution: 1024×1024

You can also start from realistic or 3D base models such as 麦橘写实, 墨幽人造人, AWPoly3D, or 3D角色IP.

Text‑to‑Image Method

Choose a base model, fill in the parameters below, and let Stable Diffusion generate the image.

Prompt (positive): A girl walks on the street, Clay, upper body, close‑up, real, masterpiece, high quality, detail

Negative prompt: ng_deepnegative_v1_75t, (badhandv4:1.2), (worst quality:2), (low quality:2), (normal quality:2)

Sampler: Euler a

Steps: 30

Resolution: 512×768

Image‑to‑Image (图生图) Method

Using the Lora model “Clay World SD1.5,” upload a reference image and apply the following settings.

Model: ReVAnimated_v122

Prompt (positive): A black ceramic robot that looks like a bear, masterpiece, best quality

Negative prompt: low quality, lowres, watermark, worstquality, beard

Sampler: Euler a

Steps: 30

Resolution: 1024×1024

After generating a basic result, you can refine it by adjusting prompts. For example, to remove unwanted beards, add beard to the negative prompt. To enhance eyebrows, add eyebrow to the positive prompt, and to fix eye issues, add eye.

Iterative Refinement Examples

By repeatedly tweaking the positive and negative prompts (adding or removing terms like beard, eyebrow, eye) and keeping the same sampler and steps, the generated images gradually approach the desired clay‑style appearance.

Finally, a satisfactory image is obtained, demonstrating that the KuPai AI platform can provide fast, free image generation and deep‑customized solutions for non‑design staff across departments.

Below are several example results illustrating the process.

In summary, the tutorial shows how to achieve high‑quality clay‑style images using Stable Diffusion by selecting appropriate models, crafting detailed prompts, and iteratively refining negative and positive terms.