You can deploy this model and start earning money today!

nguyzn/
epic-rpg-captcha-solver

Name: epic-rpg-captcha-solver
Availability: InStock

keras

Views

No views yet

Epic Guard Captcha Solver

Hosts the bundled TensorFlow CNN (trained on Kaggle from the images in model/data/) behind a Gradio app. Send a captcha image URL — e.g. a Discord attachment — and get back the predicted item class name.

Input: an image URL (200x700 RGB captcha; resized internally).
Output: JSON with the top label, its confidence, the full class_names list, and per-class probabilities.
Classes (15): apple, banana, chip, coin, dragon scale, epic coin, epic fish, fish, golden fish, life potion, mermaid hair, ruby, unicorn horn, wolf skin, zombie eye.

Files in this Space

File	What
`app.py`	Gradio app + inference.
`classifier.keras`	Trained model (~205 MB, Git LFS).
`class_names.json`	Output-order class names.
`requirements.txt`	Pinned deps (TensorFlow 2.19.0 = training env).
`.gitattributes`	LFS rules for the model.

Deploying

The model is ~205 MB, so it must go through Git LFS or the push fails.

bash

# 1. Create the Space on huggingface.co (SDK: Gradio), then clone it:
git clone https://huggingface.co/spaces/<your-username>/epic-guard-captcha
cd epic-guard-captcha

# 2. Copy the Space files in (from this repo's spaces/ folder):
cp /path/to/repo/spaces/{app.py,requirements.txt,README.md,.gitattributes,class_names.json} .

# 3. Copy the trained model in (renamed to match app.py):
cp /path/to/repo/model/classifier.keras ./classifier.keras

# 4. Track the model with LFS BEFORE adding it (.gitattributes already does this):
git lfs install
git lfs track "*.keras"

# 5. Commit & push — LFS uploads the big file:
git add .gitattributes app.py requirements.txt README.md class_names.json classifier.keras
git commit -m "Add Epic Guard captcha solver Space"
git push

The Space will build (installing TensorFlow takes a few minutes the first time), then serve at https://<your-username>-epic-guard-captcha.hf.space.

Hardware: free CPU tier works. Inference on one captcha is fast; the cold start (loading a 205 MB Keras model) takes ~10-20 s on first request.

Calling it from code (e.g. the Discord bot)

The same predict function is exposed at the API endpoint /predict.

python

from gradio_client import Client

client = Client("https://huggingface.co/spaces/<your-username>/epic-guard-captcha")
result = client.predict(
    "https://cdn.discordapp.com/attachments/.../captcha.png",
    api_name="/predict",
)
print(result["label"], result["confidence"])
# -> apple 0.985

predict never raises: on a bad/unreachable URL it returns {"label": null, ..., "error": "..."} so the caller can fall back or ping a user.

Wiring it into the bot as a third backend

captcha_bypasser.py already dispatches across backends. To add this Space, add a solve_space(image_url) that calls the snippet above and returns result["label"] when result["confidence"] >= captcha_min_confidence, else None — matching the existing solve_roboflow / solve_local contract.

nguyzn/epic-rpg-captcha-solver