Preserve Text: Sharper Text with Fewer Artifacts (July 2023)

Improve the visual quality of text

The new Preserve Text filter specializes in making small or degraded text appear sharper and more natural than existing methods. It is also the first text enhancement AI model commercially available.

Preserve Text is designed to work in combination with other AI models in Topaz Photo AI, so you should specifically select text and let other models handle non-text output. Use the Low Resolution model for small text with low pixel counts, and use Noisy/Compressed for larger text with visual quality issues.

After you enable the filter, brush to select the text you’d like to enhance. If your image is mostly text (like a book page or document), you can use the “Select All” option. Because of the importance of selecting text correctly, Autopilot won’t automatically enable Preserve Text.

Increase text sharpness

Default Photo AI upscaling created pretty good results in the example above, but you’ll still notice some smudging around the last word. Preserve Text can help increase visual fidelity in both handwritten and printed text.

Remove AI-generated artifacts in text

Regular AI upscaling lacks an understanding of how text should look, but still attempts to generate detail and often does an unconvincing job. Preserve Text fixes incorrectly generated detail - such as around the “E” letters in the example above - and ensures that text appears natural to the human eye.

Fix issues caused by combining other models

Sometimes you may want to combine multiple AI functions to enhance an image, which can mangle text because of compounding artifacts. Preserve Text allows you to confidently stack other AI filters by maintaining the legibility of selected text.

You can read more about Preserve Text in the Topaz Photo AI docs. We hope you enjoy using this new filter, and please let us know what you think by commenting below! We’ll continue improving the feature based on your feedback.

Updated controls interface

It is now simpler and faster to view and edit the right controls panel:

The major changes from the last version include:

  • You can now more easily preview and edit Crop, Subject, and Face Selection.
  • The Autopilot list is hidden to increase emphasis on the filters. We will bring this back in a more space-efficient way soon.
  • The former Upscale and Enhance Resolution functions have been combined into the Upscale filter. Despite the name, you can still downscale or apply a “1x upscale” to enhance quality without upscaling.

We’re still very interested in hearing about any usability issues you experience with Topaz Photo AI. We will continue to improve the app’s design and usability based on your feedback.

Support for Nikon Z9 / Z8 High Efficiency raw files

Topaz Photo AI now natively supports the “High Efficiency*” and “High Efficiency” raw formats used in Nikon Z9 and Z8 cameras. Directly import your .NEF file into Topaz Photo AI for improved raw quality:

  • Standalone: drag and drop your .NEF file into the app.
  • Lightroom Classic: select your .NEF file, then go to File → Plug-Ins Extras → Topaz Photo AI.

This will allow Topaz Photo AI access to more image information than importing a tiff or jpeg, which is especially useful for better noise reduction. We will also soon release a significantly improved raw noise reduction and demosaicing model that will require this workflow.

Other improvements

Since June’s update post we’ve shipped many more improvements:

  • Added option to upscale by inches or centimeters in addition to pixels
  • Improved preview color management and sped it up with better caching
  • Updated splash screen that shows recently opened files and tutorials (standalone only)
  • Face Recovery no longer only applies to half a face, modifies clothes, or changes face color
  • Updated raw camera support, lens correction database, and fixed crop & orientation issues
  • Added 3x and 5x default scaling options to Autopilot Configuration in Preferences
  • Added upscale information to appended model names to exported files
  • Fixed navigator jumping to top-left on first zoom in/out
  • Photoshop plugin now correctly persists settings
  • Fixed Lightroom Classic sometimes opening multiple Photo AI instances
  • Computer will no longer sleep when batch processing on Windows
  • Fixed various crashes, errors, and other issues
  • Many minor interface and wording improvements

Read the release threads for v1.4.0, v1.3.12, v1.3.11, v1.3.9, v1.3.8, and v1.3.7 for the full change logs.

What’s next

We’re excited to bring the next round of improvements to you over the next few months:

  • Improve RAW Remove Noise with improved demosaicing, reduced artifacts, and better handling of strong noise.
  • Improve blur reduction (Sharpen) model quality.
  • Allow adding back original image detail after applying Remove Noise.
  • Improve the batch processing export flow, stability, and performance
  • Fix occasional Face Recovery eye color and blending issues.
  • Fix occasional blurry patches caused by tile depth issues.
  • Add Automate plugin that allows for upscaling within Photoshop.
  • (Exploratory) Generative object removal that runs on your local hardware

Please apply for our beta program if you’re interested in giving feedback on these changes before they’re released. If you’re interested in setting up a Zoom call with us to discuss your workflow, contact us to schedule a time.

Thanks, and we hope you’ve been enjoying Topaz Photo AI! Please comment below with your thoughts; we look forward to hearing from you.

14 Likes

In the “we are good” example I like the upscaled one more than the preserved text one. :slight_smile:

The “Propeller Danger” example is exactly the solution I’ve been looking for.

2 Likes

Unfortunately it cannot cope with text on images that have been downsized.

1 Like

Preserve Text model is nice but can you please add in your roadmap to develop new photo enhancement model dedicated for video frame images ?
I believe this will be really useful for many users ! Many times i just get much better results using Video AI and Proteus model (for example) to upscale video frame image sequences then all the existing Photo AI models can provide . So i think we can get much better results in Photo AI with new specific model especially developed for such purpose !

Thank you in advance !

Thank you from the bottom of my heart for this “preserve Text” feature! I am a sign maker and need to use font finding software to figure out the dag-burn screenshots customers send me for making their signage. I will definitely tell the guys on the sign forums about this new feature of Topaz Photo AI.

3 Likes

A post was split to a new topic: External Editor - Fastone Image Viewer

I’m a fan of all.

4 Likes

Please stop these weekly releases as we strike problems each and every time. It seems quality control is suffering with annoying things like the latest release comes with Upscale settings on.

When are the RAW models going to be improved, currently they are a problem?

When is Autopilot going to do the job properly?

There are just so many issues that mean we are testing the Beta then we have to test the production release over again.

5 Likes

Once auto pilot select filter dropdown menu should also open , no need to click to open.

I don’t get why there are so much problems, that i don’t have.

Maybe because i did switch off the Autopilot for Enhance, Sharpen, Face and Masking.

The thing with denoising is that the auto values for it are much too high, i would set it to 4 no matter the image bc if the user want more denoising he should
change it by himself.

2 Likes

Are there any plans to introduce ability to use scanned color and monochrome negatives and correct for color? This would be a really must have feature since I am not a fan of Adobe and their subscription based service for lightroom

A post was merged into an existing topic: Topaz Photo AI v1.4.0

Where has the auto text recognition option for the preserve text option gone? I can not find it anymore. :eyes:

The Preserve Text feature is fantastic and something I’ve been looking forward to for a long time. Can you port it to Video AI?

1 Like

When PhotoAI does Text it may help to have a Gamut Line Rec 2020.
A wider Gamut Widder than sRGB.Since the letter would be a portion Text Letter’s area of a single letter and or letter can be enlarged six times.Recover text can have find Edges tool that turns image Green or Red(Background) and Blue (edges).Than a Denoise can be processed on Edges.Recover Text can have a load Font also .

@Eric can PhotoAI have an update to Process Meta Data From Sequence files (TIFF and JPEG)?

Hi (Guten tag!?),

I just saw your post & figured I’d try turning off the Sharpen pref (it’s the only A.P. pref I have on…and it’s set to all levels of sharpness; low to high) to see if I get different results.

Would you pls tell me how you turned it off? I cannot do so. At least I don’t see any “None” option like for the other A.P. setting preferences. If I try to scroll down in that Sharpen Prefs area, the Sharpen expand just closes. So if there’s something lower down, there’s a bug and the option for “None” is unreachable.

I’ll attach a screen snip showing there is no “none” under Sharpen for me - Win 11 Pro.

Thx in advance!
Fotomaker_PAI-Prefs_Sharpening-NoNONEoption

BTW, I just tried clicking on the checkmarks in the Prefs for each of the sharpening strengths - hoping they’d be toggles. I was able to uncheck them to hopefully have converted my Sharpening pref. to None for when I can test again (post surgery). What was throwing me was that when I had the Sharpening Pref category expanded to see the levels of sharpening listed, it looked as though there was something else below them that was mostly cut off. I’d assumed that what looked like a truncated top of letters was how to click on ‘None’.

Maybe it would be useful to have a “None” category in the Prefs for Sharpening (like Face Detect None option) & then the extra option to select which of the tiers of sharpening we want if Sharpening is clicked to On (almost like adding Neck & Hair for Face Detect enabled). Just a thought for the devs…