-
Notifications
You must be signed in to change notification settings - Fork 261
Add Gemini Nano support for prompt validation with fallback mechanism #171
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add Gemini Nano support for prompt validation with fallback mechanism #171
Conversation
Summary of ChangesHello @takahirom, I'm Gemini Code Assist1 ! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed! This pull request integrates Gemini Nano into the prompt validation workflow, mirroring the existing image validation strategy. Previously, prompt validation exclusively used Firebase AI, while image validation leveraged Gemini Nano with a Firebase AI fallback. This change introduces Gemini Nano as the primary validation method for prompts, with Firebase AI serving as a robust backup, ensuring consistency across validation processes and potentially enhancing performance or capabilities. Highlights
Using Gemini Code AssistThe full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips. Invoking Gemini You can request assistance from Gemini at any point by creating a comment using either
Customization To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a Limitations & Feedback Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here. You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension. Footnotes
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Code Review
This pull request is a great addition, bringing consistency to the app by adding Gemini Nano support for prompt validation with a fallback mechanism, similar to the existing image validation. The changes are well-structured, including updates to data sources, remote config, and tests.
I've left a couple of suggestions. One is a high-severity recommendation to improve the fallback logic to be more robust, ensuring it triggers not just when Nano is unavailable, but also when Nano's validation fails. This aligns better with the likely intent and improves reliability. The other is a minor suggestion for code conciseness.
Overall, excellent work on improving the app's on-device capabilities!
data/src/main/java/com/android/developers/androidify/data/ImageGenerationRepository.kt
Show resolved
Hide resolved
data/src/main/java/com/android/developers/androidify/data/GeminiNanoGenerationDataSource.kt
Outdated
Show resolved
Hide resolved
riggaroo
commented
Nov 17, 2025
@calren can you help with more information here for @takahirom
calren
commented
Nov 20, 2025
So cool that you're talking about this at DevFest Tokyo!
The goal of the original PR was to keep the image on the user's device for enhanced privacy. Since validatePromptHasEnoughInformation does not require sending the user's image to Firebase AI, we decided to keep using Firebase AI for it.
One consideration is that switching validatePromptHasEnoughInformation to on-device may require more tuning of the prompt. We did not extensively test whether the existing prompt for this use case was sufficient for on-device, or whether would we need to craft a new prompt to reach comparable quality in terms of speed and accuracy.
calren
commented
Nov 20, 2025
I see that a separate prompt prompt_text_verify_nano was added. Was this new prompt added after seeing that prompt_text_verify was not meeting expectations in terms of quality? @takahirom
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
should this value be something smaller like 2 since the output is either 'invalid' or 'valid'?
takahirom
commented
Nov 24, 2025
@calren
Thank you for providing the context! I understand that privacy is vital when uploading self-images. I had thought it was solely to optimize costs. I'm satisfied with the answers and have no particular issue with the merge. So we can close this PR.
Regarding the quality of prompts for the local model, I have no problem with the cloud prompts, and they are merely adjustments based on other examples. We can unify the prompt because I believe this is not a particularly challenging task for GenAI.
The best course of action, in my view, is to have CI validate or measure the effectiveness of prompts using Gemma 3n. It will definitely take some time, though.
Thanks for this awesome reference app! I'm going to talk about this app in DevFest Tokyo.
While exploring the codebase, I noticed that Androidify uses Gemini Nano for image validation (
validateImageHasEnoughInformation) with a fallback to Firebase AI, but prompt validation(
validatePromptHasEnoughInformation) only uses Firebase AI directly.I'm curious about the Gemini Nano strategy here. Is there a specific reason why prompt validation doesn't leverage Gemini Nano? If not, this PR adds Gemini Nano support for prompt validation to maintain
consistency with the existing image validation pattern.
Changes
promptTextVerifyNano()to RemoteConfigDataSource with validation promptvalidatePromptHasEnoughInformation()to GeminiNanoGenerationDataSourceImageGenerationRepositoryto try Gemini Nano first, then fallback to Firebase AILet me know if there's a specific reason this pattern wasn't used for prompt validation!