Last updated on 16 March 2026
New apps are developed constantly, but among the many apps we come across and test, only a few bring something new or unique to the table. Such ideas usually deserve more than a simple app entry with just a few notes.
I previously wrote about PiccyBot, the app that first brought video description to Android, and Be My AI, a feature of the Be My Eyes app that was the first to provide a reliable image description service with follow-up questions.
ScribeMe is another addition to the AI apps designed for blind users, with the ambition of standing out among apps that are gradually becoming similar in their features and capabilities. What sets ScribeMe apart is its focus on smart recognition of PDF content, along with its Live Assist feature, which alerts the user to changes in the surroundings without the need to continuously prompt it.
This article includes my impressions of ScribeMe. The app version at the time of writing is 2.0.17. Since the app relies on external AI services for its functionality, changes in those services may affect the results produced by its features. Some observations in this article therefore relate to the AI models that serve as the backbone of the app.
Table of Contents
What Is ScribeMe
ScribeMe is an app catering to visually impaired users that offers several AI-powered features:
- Smart scanning: Scans PDF and PowerPoint files and presents the text and other content in an accessible way.
- Live Assist: An AI assistant that attempts to monitor changes in the environment without the need to be continuously prompted.
- Snap nScribe: An image description feature that allows follow-up questions and supports face recognition.
The app’s UI is intuitive and simple, with the main tabs located at the bottom of the screen. Each feature also includes a brief explanation on how to use it when selected.
The free version of the app is very limited and mainly acts as a preview of the available features. To benefit from the full functionality and remove the limitations of the free version, a monthly or yearly subscription is required.
PDF and PowerPoint Accessible Scanner
This feature is the main focus of the app. It was the first feature the app was built around, and it is currently located in the Home tab, which is the default tab when the app opens. The scanner uses an online AI service with the goal of extracting all available text from a file and displaying it as formatted text. It also provides descriptions of images and other elements, such as logos.
To scan a file, choose either the PDF or PowerPoint tab and then press Upload File to select the file you want. Once selected, press Scan to start the scanning process. Alternatively, you can share a file from another app to ScribeMe, then press Scan to initiate the process. After scanning is complete, the results are presented on the screen, with the option to ask additional questions about the document or export a formatted Word (.docx) version. This allows you to share the file with other apps or save it to your device using a file manager.
Free users can upload one file per day, with only the first 10 pages scanned. Pro subscribers can upload up to 100 files per month with no daily limit, and the 10-page limit is increased to 400 pages. The maximum PDF file size is not specified.
My Impressions of the Feature
The accuracy of the scanning results impressed me greatly. Text extraction is excellent, and the additional information provided—such as how certain text is formatted or descriptions of included images—makes the results richer and more useful.
This is not the only app that provides PDF scanning, but it is the only app I know of that attempts to describe included images while also providing contextual information such as the color of certain text. The text extraction is also among the most accurate I have encountered. The ability to export the results as a Word document is very useful, especially if you want to keep the text or make edits.
I tested the app with a university certificate, a passport, a school holiday schedule, and a copy of a children’s book. The tested languages were Arabic and English. The waiting time is reasonable and varies depending on the size of the file. It should be noted that my tests were only with PDF documents and not PowerPoint presentations.
While viewing the scanned results, the Ask AI button can be used to ask additional questions about the document. It is worth mentioning that the AI receives the extracted text rather than the original file. Because of this, it is not possible to request improved descriptions of images that were included in the document.
However, several issues and observations are worth noting.
Currently, there is no indication through the screen reader that the app is processing the file except sound and vibration feedback. If these are disabled in settings, it becomes impossible to know whether the file is being processed. Even when sound and vibration are enabled, there is no indication of the processing stage, such as whether the file is uploading or being analyzed. After pressing Scan, only the file name is shown without any additional status information. Once the scanning process is complete, the file reader opens automatically to display the results.
If the app is left while a scan is in progress, a notification is sent when the results are ready. However, during my testing, tapping the notification did not open the scan results and instead returned me to the main interface.
Recent files can be accessed through the File History section located in the Profile tab. Here I also encountered an issue where a scanned file sometimes did not appear immediately in the history and only became visible after another file was scanned. In the File History list, long-pressing a file shows the delete confirmation.
Although this is not a problem for me personally, it should be noted that it is currently not possible to instruct the app to extract only the available text without attempting to generate additional content such as image descriptions. This may be problematic if the goal is simply to obtain an accessible version of the document’s text.
Another limitation worth mentioning is that there is currently no way to purchase additional file scans if the monthly limit of 100 files is exceeded.
Overall, I am quite satisfied with the text extraction performance of this feature.
Live Assist
Live Assist is an AI assistant that uses the camera feed to answer questions and describe surroundings or identify objects. What sets Live Assist apart from other mainstream assistants is its ability to continuously monitor the camera feed and report changes without the need for repeated prompts.
Although both Gemini and ChatGPT allege they can report changes directly, neither can do so at the time of writing. Live Assist circumvents this limitation by continuously sending snapshots of the camera view for analysis, allowing it to provide live announcements of changes, such as detecting a specific item or describing shifts in the environment.
To enable Live Assist, select its tab from the main UI and press Go Live to start interacting. You can mute the microphone, toggle the flashlight, and end the session when done. If you don’t speak, Live Assist keeps reporting what it sees. Free users have 4 minutes of daily access, while pro users can use it without time outs—but don’t take that literally and try to keep it running for hours.
Observations
I can confirm that Live Assist’s live reporting of changes works in practice, unlike Gemini or ChatGPT apps. The continuous sending of camera snapshots allows the assistant to report changes it detects or confirm the presence of requested items. For example, I asked it to notify me when a person appeared in view, and it successfully did. I also prompted it to report when the person raised their hand, which it also did correctly. When I tried identifying objects, it provided updates as I moved a package, noting when it could detect it better or worse, as well as reporting other items like a seat and a plant.
It is important to note that reporting is not instantaneous. While not slow, there is a short delay due to image uploading, processing, and response generation. You should expect a few seconds between an event occurring in the camera view and the assistant reporting it. Consequently, despite its usefulness, this feature should not be relied upon to identify obstacles in unfamiliar or potentially dangerous environments.
Occasionally, due to the continuous camera snapshots, the assistant may answer a question based on a previous image rather than the current view. In such cases, it is better to wait a moment before asking.
Regarding accuracy and reliability, the app performed on par with, and sometimes better than, ChatGPT or the free Gemini tier. For instance, Gemini keeps getting confused by a transparent glass balcony door at my house, insisting it is open when it is actually closed. Live Assist correctly reported that the door was closed, recognizing the frame and seeing that there was no gap between the glass and the frame, concluding that the view of the outside was due to the glass’s transparency.
Live Assist can identify gender and skin color when asked, state hair color and clothing, and even report details like a person having a mustache. The app also recognizes facial expressions and gestures. The model used in Live Assist does not attempt to be judgmental or deduce moods from images, which I appreciate. It focuses on describing what it sees without offering unnecessary opinions. For instance, when it reported a person wrapping themselves in a blanket, I asked if they were cold, and it correctly replied that it could not determine this just from the fact that the person was under a blanket. I also noticed that the model is cautious about unclear images, refusing to identify products if the image is blurry or insufficient, rather than guessing incorrectly.
Like any AI assistant, Live Assist is not free from issues and misidentifications. A prominent problem concerns directions. While testing live reporting, I asked the assistant how to reach detected items. One example involved a photo frame hanging on a wall. When the assistant recognized it, I placed my hand on the wall and asked it to guide me to the frame. It initially told me to move my hand to the right, when in fact I needed to move left. It kept insisting I should go right. After ignoring its guidance and moving left to reach the frame, I informed it that the correct direction was left, and it provided the right guidance on the second try, which I did without ending the ongoing session. Also, when it detected that my hand touched the frame, it automatically notified me.
This misidentification of orientation was not an isolated incident; it happened more than once. This experience suggests that the assistant should not be relied upon in unfamiliar environments or places that may contain hazards. It’s worth noting that other AI assistants have exhibited similar issues. How the camera is held and how the AI analyzes images play a critical role in locating items and determining their positions. If this data is not available or interpreted incorrectly, it can lead to wrong guidance, making it unreliable for navigating or reaching specific objects safely.
Despite the generally good accuracy, Live Assist sometimes misidentifies objects. For example, a rubber band in my nephew’s hand was repeatedly identified as earphones because his hand was close to his ear while holding it. A chicken luncheon can was identified as a tuna can, probably because the can’s circular shape is similar to typical tuna cans.
Regarding package handling, the cautious nature of the model was very evident. With several boxes and cans, the assistant didn’t immediately assume the product. It insisted that the shown image wasn’t enough, providing only descriptions of what it saw without deducing what the item actually was. Even when I asked it to infer the item from a recognized brand name, it refused to do so. I was only able to identify a coffee box because I recognized the brand name that the assistant provided.
During my tests, I repeatedly asked Live Assist to guide me on how to move the camera or the item to get a clearer view, reminding it that I am blind. It sometimes offered easy-to-follow guidance, but other times it failed to help me get a clear picture and would settle for providing a description of what it could see instead.
Text recognition is where Live Assist struggled the most. English was set as the assistant’s language in its settings, but I showed it products with Arabic text included. It correctly identified the text as Arabic but repeatedly declined to read it. This reluctance wasn’t consistent across tests. When shown a page from an Arabic book, it acknowledged the text was Arabic, and after insisting, it successfully read it, allowing me to understand it despite the accented English reading.
During my tests, I noticed that it was often necessary to prompt the assistant several times before it would read the entire text. When presenting pages from an English book, Live Assist sometimes summarized or gave an overview instead of reading the full text. It did read an entire page after several prompts, but it wasn’t a straightforward “read this text” process. On another occasion, it claimed a sticker was covering part of the text and could not read it, even though the sticker wasn’t obstructing anything. Rotating the document allowed it to read the text, confirming that my correction about the sticker was accurate.
It’s worth reiterating that other live assistants aren’t free from similar issues, such as misidentifications or text hallucinations. For example, the free Gemini version sometimes generated text that wasn’t present.
Live Assist likely relies on a Gemini-based model, although I don’t know which one exactly. The model performs very well and is not inferior to competing alternatives, especially when combined with the continuous detection feature that enables live monitoring and reporting.
Interaction with Live Assist is only through voice; it’s not possible to type prompts. I didn’t face issues with the assistant understanding my speech, but, similar to other live features, when other people are talking nearby, the assistant struggles to isolate your voice. This can cause the assistant to stop mid-sentence, thinking it should listen for a prompt due to the surrounding noise.
During the conversation, Live Assist generally keeps its focus on the image it sees. When asked a question that cannot be answered from the provided image, it states that it cannot find an answer. Additionally, it follows Gemini’s refusal to give medical advice. For example, when I asked about an identified ingredient on a medicine’s package, it initially declined. After I clarified that I only needed general information, it confirmed that the ingredient was used for treating diabetes, while reminding me that it was not providing medical advice and that consulting a doctor is necessary.
That said, despite the main focus being the camera view, it is possible to ask general questions. However, the answers tend to be brief rather than detailed. Currently, there is no way to review past conversations. Adding a conversation history could be very useful if users need to go back and pull specific information.
In terms of stability, throughout my tests I was able to start the assistant and interact with it without issues most of the time. However, I noticed that the assistant sometimes stopped working after a few minutes of use. This happened several times, though due to the not very stable nature of my internet connection, the issue may have been related to that.
Regardless of whether my connection speed was a factor, what I observed was that the assistant would not resume working until I ended the conversation. In some cases it worked again immediately after starting a new conversation, while in others I had to wait a bit before it responded again. In all cases, no error messages were shown; the assistant simply stopped responding or describing anything.
Screen Sharing
In the Live Assist tab, press Share Screen to start sharing your screen with the assistant. Once sharing begins, the assistant can recognize what is displayed and answer questions about the screen content. When starting screen sharing, you are asked whether you want to share a specific app or the entire screen.
This feature is not working on my phone, as I do not receive any feedback after starting screen sharing. The developers are aware of the issue and are looking into it.
Snap nScribe
There is no shortage of good image description services, with even TalkBack including a reliable built-in one. What makes ScribeMe’s image descriptions worth checking is the face recognition feature. To describe an image, you can either share it from other apps or use the Snap nScribe tab, where you can take photos directly or open the gallery to upload saved ones.
To save a face, choose Recognize People and assign a name to the person. You can save up to 10 faces using an accessible camera with guidance on how to take a photo of the face, the regular camera, or by selecting a photo from the gallery. Once a face is saved, whenever it is detected in a described photo, the assigned name is spoken automatically. I tested it with one face, and the app successfully recognized the person in a photo alone as well as in another photo where the person was with someone else.
Note that face identification only works in image descriptions and is not available in Live Assist. I will not discuss the overall quality of the descriptions in this review, as similar features exist in many other apps. It is enough to say that the descriptions are quite good. It is also possible to ask follow-up questions about the photo after receiving the description.
A word about subscription plans
ScribeMe’s subscription plans are not on the cheap side. In my country, the monthly plan costs $20, similar to what you pay for ChatGPT Plus or Gemini Pro subscriptions. This puts the app in direct competition with the aforementioned mainstream services, which could put it under pressure to deliver high-quality, reliable services that justify paying for it instead of Gemini or ChatGPT.
We should not forget that ScribeMe is a specialized app that excels in certain areas, whereas mainstream subscriptions usually offer a more rounded, general package.
It is unfair to demand that an app developer cover costs from their own pockets, or to challenge their right to earn a profit. At the same time, striking a balance is key to a project’s success.
Offering more flexible plans could be a solution worth exploring. This could include paying for file scans as needed, paying for a number of Live Assist minutes, or providing a smaller plan with more limitations that could suit users who might not be able to pay for the current plans.
Final Remarks
ScribeMe is not just another app in the crowded market of apps with similar functionality. It is an app that seeks to earn its place among innovative apps with original ideas. It leverages AI capabilities to address real-life challenges, such as the many inaccessible PDFs a blind person encounters while studying or working. The app also takes a step forward in making AI assistants more useful through continuous camera view uploads, allowing the assistant to report changes in real time. There is no denying that issues and shortcomings exist, some of them are due to current AI models, which are not yet fully reliable despite the enormous improvements we’ve witnessed.
Starting a project is one thing, but maintaining it and ensuring sustainable development is another, as many challenges lie ahead. For ScribeMe, the desire to improve is clear, and the team’s determination is noticeable. Funding the project, through external support or subscriptions, and keeping up with advances in AI models are essential to keep the service running smoothly and competitively.
I try to be very objective in my reviews, but I admit I am more personal when it comes to supporting developers with a vision who, in my opinion, aim to make a meaningful change. ScribeMe deserves a chance. If it can assist you and you are able to subscribe, or can help by reporting bugs or sharing the app, you take part in pushing the project forward and supporting an initiative that attempts to have a positive impact on the lives of visually impaired people.
Comments