Image-to-Normal Map Generators Compared on Six Textures

How we designed the comparison

A normal-map comparison is only useful when every tool receives equivalent source data and the outputs are judged in the same way. The final test uses the same crop and source resolution for every generator, records the software version and settings, and applies no post-processing to the displayed outputs.

Every normal is presented in the OpenGL convention, then rendered on the same neutral material under the same lighting. We judge low-frequency form, edge direction, preservation of useful surface detail, artifacts, and the amount of cleanup needed. We do not judge how dramatic the purple thumbnail looks.

• Six identical input textures, covering different surface structures and detail scales.
• The same output resolution, normal convention, crop, render material, and lighting.
• Default settings where possible; every manual adjustment documented beside the result.
• No selective repainting, sharpening, or cleanup before the comparison images.
• Full-resolution files retained so readers can inspect artifacts beyond the web preview.

The six-texture comparison boards

Each comparison board contains the source texture and the generated normals side by side, ordered from Lamron Mago2 through Substance 3D Sampler and ShaderMap 4 to NormalMap.ai. A second row of close-up crops makes fine detail and artifacts easier to inspect.

Texture 1

Texture 2

Texture 3

Texture 4

Texture 5

Texture 6

What each workflow offers

Speed descriptions are relative observations from our test machine and connection. The final article will include the measured time, version, settings, and output resolution for every run.

NormalMap.ai: fast browser processing, limited shape inference in our test

NormalMap.ai is quick to open and needs no desktop installation. Its upload workflow processes images locally in the browser with GPU and WebGL support, while the site also offers a separate text-to-texture generator. That makes it convenient for a fast preview or a machine where installing software is not practical.

Across our six images, its normals behaved much like a luminance- or height-derived conversion: visible color changes frequently became relief, while the larger physical structure remained flat or ambiguous. The results required the most correction before we would use them in production.

The provider describes parts of the product as AI-assisted. We did not audit its implementation, so this comparison does not claim to prove which algorithm powers the upload converter; it reports the behavior visible in the test outputs.

ShaderMap 4: more controls, but still dependent on image cues

ShaderMap 4 is a fast offline Windows tool with normal editing, map filters, and controls that make it easier to tune an image-derived result. In our test it recovered more of the medium and large shapes than the simplest browser conversion, but several forms still followed brightness rather than the likely physical surface.

It remains useful when local processing, predictable controls, and a one-time license matter more than automatic semantic shape prediction. ShaderMap 4 Basic is free for non-commercial use; the paid Personal and Commercial editions unlock the relevant licensed use and additional Pro features.

Substance 3D Sampler: genuine AI inference with an established material workflow

Adobe documents Sampler’s AI Powered Image to Material as a neural-network workflow trained across fabrics, organic materials, and indoor and outdoor surfaces. It predicts normal, height, and roughness while reducing shadows and highlights, so it goes beyond treating brightness as height.

Its results were clearly more useful than the filter-led workflows in our sample. However, we also saw pixelated artifacts, too much small-scale information on some textures, and broad shapes that did not consistently match the apparent structure of the source. Sampler is still the more complete desktop material application and may be the better value for artists who use its wider authoring workflow or the other Substance tools.

Lamron Mago2: the strongest overall result in this comparison

Mago2 is Lamron’s higher-quality image-to-normal model. According to our internal training documentation, it was trained on a proprietary dataset built from high-quality scanned surfaces. The goal is to learn the relationship between visible material features and plausible geometry instead of turning every light or color change into relief.

That difference showed most clearly in the large forms. Across the six selected textures, Mago2 more consistently predicted the main surface direction while retaining enough fine information to keep the material alive. It was not the fastest option, but its outputs needed the least corrective work in our visual review.

Mago2 runs in the cloud, so the upload and generation require an internet connection. The advantage is that processing does not depend on the user having a compatible local GPU: an image can be submitted from a phone, laptop, or desktop, and completed results remain available in the generation history for later download.

What the cost comparison actually looks like

Adobe’s German pricing currently lists Substance 3D Texturing at €23.99 per month on an annual plan billed monthly and Substance 3D Collection at €56.99 per month. These are two product tiers, not old and new prices: Texturing includes Painter, Designer, Sampler, and 3D Assets, while Collection adds more Substance applications. Comparing either subscription only with normal-map generation therefore understates the value of Adobe’s wider suite.

Mago2 costs 20 Lamron credits per result. At the displayed credit-pack prices, 500 credits for €5.94 cover 25 Mago2 generations at about €0.24 each. The 7,000-credit pack costs €47.59 and covers 350 generations at about €0.14 each. Purchased credits do not expire and there is no monthly fee.

Practical note: Prices include German VAT where shown and were checked on September 8, 2026. Promotions, regional taxes, checkout pricing, and future model costs can change the calculation. Lamron’s 50 free monthly credits cover two 20-credit Mago2 generations, with 10 credits left over.

Web access or offline control?

Lamron’s browser workflow is useful when the source is already on a phone or when a local GPU is unavailable. A surface photographed on a hike can be uploaded immediately, or queued from a laptop on a train, without installing a desktop suite. The trade-off is the need to upload the source and wait for cloud processing.

ShaderMap and Sampler keep the main workflow on the desktop and can be used without an active internet connection after installation and licensing requirements are satisfied. Adobe also provides a Capture mobile workflow that can preview materials and send them to Sampler, although the connected functionality is tied to the relevant Substance subscription. Privacy requirements, file size, connection quality, batch volume, and existing licenses may matter more than raw generation time.

Our verdict

For these six inputs and the recorded settings, Lamron Mago2 produced the best overall normal maps in our review. NormalMap.ai was the fastest way to get a browser preview but produced the least usable shape information in this sample. ShaderMap offered better adjustment and an offline workflow. Substance 3D Sampler delivered credible AI-generated material data and a much broader authoring suite, but its normal predictions were less consistent than Mago2 on the selected textures.

The practical choice still depends on the job. Use the full-resolution grid to inspect the forms that matter to your own materials, consider how much manual cleanup each result needs, and test shortlisted tools on sources representative of your production work.