BetterGI: Windows UI Automation for Genshin Impact, Reviewed

What BetterGI automates, and who the README is written for

BetterGI targets the repetitive parts of a Genshin Impact daily routine. The README lists real-time tasks that trigger on what is on screen: auto-pickup presses F when an interactable appears, auto-dialogue fast-clicks through story and selects options, quick teleport clicks a waypoint on the map, and semi-auto fishing casts and reels with AI recognition. Standalone tasks run longer loops: the Genius Invokation TCG bot clears PvE invitations and weekly guest challenges, auto-felling uses the Z key with the Tree of Auspiciousness item and relies on logging in and out to refresh wood, auto-domain cycles a domain from entry to the ancient tree reward, and there are bots for Stygian Onslaught, leylines, the music minigame, cooking, and artifact salvage. A one-dragon mode chains the daily routine together.

The intended user is a Windows player who already runs the game comfortably. The README states the recommended machine should handle medium settings at 60 fps, because image recognition is expensive. That is a real admission: this is not a lightweight overlay. The audience is also Chinese-first. The repository ships English, Simplified Chinese, Traditional Chinese and Japanese README translations, but the documentation site and the feature pages linked from the README are in Chinese, so an English-only reader gets the feature list and not much of the operating detail.

Screen capture and simulated input instead of memory reading

The architecture is visible in the repository layout. The solution is split into the main BetterGenshinImpact project plus three supporting libraries: Fischless.GameCapture for grabbing frames, Fischless.HotkeyCapture for global hotkeys, and Fischless.WindowsInput for synthesizing mouse and keyboard events. That split tells you the whole design: the program captures the game window, runs computer vision over the frames, decides on an action, and injects an input event. The README describes it as relying purely on visual algorithms and simulated operations, and states it does not modify game files or read or write game memory.

The README also mentions a mask window drawn over the game interface to show logs and recognition results. That is a useful debugging surface, because when a bot misreads the screen you can see what it thought it saw. The trade-off is inherent to the approach: anything that changes the pixels changes the recognition. The README warns that window size changes, resolution switches and monitor switches require restarting the software, that no screen filters are supported (HDR, Nvidia filters), that game brightness must stay at default, and that only 16:9 is supported, with 1920x1080 windowed recommended. Those are not incidental caveats. They are the boundary of the vision pipeline.

Installing BetterGI on Windows and running your first task

The README points downloads at the GitHub releases page, with a mirror on CNB for faster downloads in China, and says a first-time user should read the quickstart page. There is no package manager install and no build step required for normal use. The prerequisite is a 64-bit Windows 10 or later system with the .NET 8 runtime; the README says that if the runtime is missing, the program prompts you to download and install it on launch, linking to the .NET runtime download page.

There is no command-line install step in the README, so the procedure is the graphical one. Download the release archive from the releases page, extract it, and run the executable as administrator. The README explains why: the game itself runs elevated, so without matching privileges the program cannot simulate mouse clicks. On first launch, go to the start page, choose a screen capture method, then press start. The README's own summary of this step is that after choosing the capture method and clicking start, you can use the features.

Before starting any long task, set the game to a 16:9 window at 1920x1080, leave brightness at default, and turn off HDR and any Nvidia filters. If you change resolution or move the window to another monitor, restart BetterGI rather than expecting it to re-detect. Builders who want to compile the project themselves are pointed by the README at the compilation guide inside the BetterGenshinImpact directory; the README does not reproduce those steps.

Where BetterGI fails, and the account risk the README admits

The most important limitation is not technical. The README's FAQ asks directly whether the account will be banned and answers that it theoretically will not be, because BetterGI does not modify game files or touch memory, but then states that the terms of service list third-party software and simulated operations as grounds for a ban, and that detection remains possible. It advises using it quietly. Whatever you think of that framing, it is the project telling you the risk is real and unquantified. If an account ban would be unacceptable to you, this is the wrong tool regardless of how well the automation works.

Two operational failure modes are documented. First, security software: the README warns that simulated operations may be intercepted by some security products and says to whitelist the program, naming 360 and Windows Defender with custom rules as known to block certain simulated clicks. Second, environment sensitivity: the resolution, aspect ratio, brightness and filter constraints above mean the vision pipeline breaks when the display setup changes, and the fix is a restart, not a setting.

There is also a scope limit worth stating plainly. This is a Windows desktop application. The README lists Windows 10 or later as a requirement, and nothing in the repository suggests a Linux, macOS, or mobile build. People searching for an APK are looking for something this project does not ship.

How BetterGI differs from memory-based Genshin assistants

The obvious alternative category is assistants that hook into the game process rather than watching the screen. The README's acknowledgements name several projects in this space, including genshin_impact_assistant, cvAutoTrack, HutaoFisher and Fischless, and the distinction between approaches is the interesting part. A memory-reading assistant can query game state directly, which makes it fast and largely independent of resolution, brightness or filters. BetterGI gives that up. In exchange, it never reads or writes game memory and never modifies game files, which is exactly the property the README leans on when answering the ban question.

That is a genuine engineering trade, not a marketing line. Vision-based automation is slower and more fragile, but its failure modes are visible: the mask window shows what was recognized, and a misread is a misread rather than a corrupted state. If your priority is reliability across arbitrary display configurations, a memory-based tool is the better fit. If your priority is staying out of the game process entirely, BetterGI's approach is the one that matches that constraint.

Release cadence, adaptation cost and the GPL-3.0 licence

The release history shows the maintenance cost this kind of project carries. Version 0.64.0 is titled as Genshin Impact 7.0 adaptation, and 0.65.0 as adaptation and optimization. Each game update can shift UI elements, and a vision pipeline that keys on those elements has to be retuned. The last push to the repository was on 2026-09-20, one day before the date of writing, so the project is being worked on now, but the pattern to expect is bursts of adaptation work around game patches rather than a stable interface you can set and forget.

The licence is GPL-3.0, stated in the README and present as a LICENSE file at the repository root. For an end user running the released binary, that is unremarkable. For anyone who wants to reuse the code in a closed-source product, GPL-3.0 is a copyleft licence and the source of derivative works must be made available under the same terms. If you plan to redistribute a modified build, read the licence text rather than relying on a summary; this is a description of the licence, not legal advice.

Upgrade cost is low in the normal case: download the new release and replace the old one. The README does not document a rollback path, so if a new version misbehaves on your machine, keep the previous archive.

What to check before you commit to running it

Confirm the runtime, confirm the display setup, and confirm your antivirus behaviour before you invest time in configuration. The three checks that matter are the presence of the .NET 8 runtime the README requires, a 16:9 window at 1920x1080 with default brightness and no filters, and a clean run of one short task such as auto-pickup with your security software active. If the pickups do not register, the README's whitelist advice is the first thing to try, not a configuration deep dive.

Beyond that, the practical question is which tasks you actually want. Enabling the one-dragon daily routine is a different commitment from enabling auto-dialogue, because it chains several bots and takes over more of the session. The README links each feature to its own documentation page, and those pages are the place to check what a given task does before turning it on.

Editorial conclusion

BetterGI suits Windows players who want to automate repetitive daily tasks and accept that the project reads pixels and clicks rather than touching game files, and who run 16:9 at 1920x1080 with no HDR or Nvidia filters. It is the wrong tool for anyone on Linux or macOS, anyone using a non-16:9 resolution, and anyone unwilling to grant administrator rights. Before adopting it, verify three things: that your machine runs Genshin Impact at medium settings and 60 fps, that your antivirus will not block the simulated clicks (the README names 360 and custom-rule Windows Defender as known blockers), and that you accept the account risk the README itself acknowledges.

Frequently asked questions

Official sources

1. babalae/better-genshin-impact on GitHub
2. License: GPL-3.0
3. Project website
4. README
5. Releases