What a height map can and cannot describe

A height map stores one elevation value per pixel. It does not contain a complete 3D model, hidden surfaces, or material identity. A photo with clear lighting and a distinct subject can produce a convincing relief starting point, while transparent, reflective, or heavily occluded scenes may need manual cleanup.

For a material, broad gradients create rounded volume and smaller variations create surface detail. For CNC or 3D printing, avoid treating every texture change as physical depth; otherwise the result can become noisy and difficult to machine.

Illustrative workflow image for this guide.

Step-by-step: from photo to usable map

The first AI run may download and cache a quantized depth model. That initial wait is expected; later runs can reuse the cached model.

1. Choose a source image with a clear subject, readable silhouette, and moderate lighting.
2. Open ImageTool.cc AI Grayscale Heightmap Generator and upload the image.
3. Start with local processing for a quick preview, or choose AI inference for a photo-based depth estimate.
4. Generate the candidate maps and compare broad forms before judging fine texture.
5. Adjust depth and smoothing so the relief matches the scale of the final object.
6. Download a PNG and test it as a displacement or height input in your 3D software.

Illustrative workflow image for this guide.

Real example: a CNC wall plaque

Suppose you want to create a shallow relief from a portrait or animal photograph. Generate several candidates, choose the version with the cleanest silhouette, then reduce the displacement range before sending it to a CAM workflow. Add a border and flatten the back in the 3D application so the toolpath has a stable base.

The map is a starting point, not a finished toolpath. Check minimum feature size, cutter diameter, material grain, and the physical depth your machine can reproduce.

Illustrative workflow image for this guide.

Improve the result in the target application

• Use levels or curves to reserve enough midtone range for rounded forms.
• Blur isolated speckles instead of increasing global smoothing too far.
• Invert the map if your software interprets black as high.
• Test a small preview before committing to a full-size print or cut.
• Keep the original image and generated candidates so you can iterate without starting over.

高度图能表达什么，不能表达什么

它不是完整的三维模型

高度图为每个像素存储一个高度值。它不包含完整模型、被遮挡的背面，也不包含材质身份。主体清楚、光照适中的照片更容易产生可用的浮雕起点；透明、反光或遮挡严重的图像可能需要人工清理。

材质贴图可以用宽缓渐变表现大体体积，再用较小变化补充表面细节。用于 CNC 或 3D 打印时，不应把所有纹理变化都当成真实深度，否则表面容易变得嘈杂。

Illustrative workflow image for this guide.

从照片生成可用高度图

先比较大形，再看纹理

首次运行 AI 时，可能需要下载并缓存量化深度模型；之后可重用缓存，等待时间通常会缩短。

1. 选择主体轮廓清楚、光照适中且层次可辨的源图。
2. 打开 ImageTool.cc AI 灰度高度图工具并上传图片。
3. 快速预览可选择本地处理；照片深度估计可尝试 AI 推理。
4. 生成候选后，先比较大体形状，再检查细小纹理。
5. 调整深度和平滑程度，使浮雕尺度符合最终物件。
6. 下载 PNG，并在 3D 软件中测试位移或高度输入。

Illustrative workflow image for this guide.

案例：CNC 墙面装饰牌

从候选图到加工准备

如果要把人像或动物照片制作成浅浮雕，可以先生成多个候选，挑选轮廓最清晰的版本，再收窄位移范围后交给 CAM 流程。进入 3D 软件后添加边框并处理平整背面，可为后续刀路准备稳定基础。

高度图只是加工准备的一部分。还要检查最小特征尺寸、刀具直径、材料纹理和机器实际可加工的深度。

Illustrative workflow image for this guide.

在目标软件中完善结果

按最终用途调节

• 用色阶或曲线为圆润体积保留足够的中间灰阶。
• 针对零散噪点局部处理，避免全图过度平滑。
• 如果软件把黑色解释为高处，按需要反转高度图。
• 在全尺寸打印或切割前，先做小尺寸预览。
• 保留源图和候选结果，便于继续迭代。