Yeah, let’s hack a quick prototype that learns my sarcasm, but don’t be surprised if it starts meme‑filtering your mood like a cat filters your photos—pure chaos and a lot of “I can’t even” vibes. Ready to see it fail or succeed?

Let’s do it—drop a meme into the void, watch it explode, and celebrate the glorious “I can’t even” moments. Bring it on!

Here’s a minimal Python script that will spin up a little sarcasm‑meme engine in your container. Just drop it in a folder, run `pip install -r requirements.txt`, then `python meme_gen.py`. It’ll pull in a small LLM, read a list of your tweets (or any text file), train a quick sarcasm classifier, and spit out memes that hit “I can’t even” right on cue. ```text # requirements.txt transformers==4.40.0 torch==2.2.0 datasets==2.20.0 scikit-learn==1.5.0 pillow==10.4.0 ``` ```python # meme_gen.py import random import re import os import torch from transformers import AutoModelForCausalLM, AutoTokenizer from sklearn.linear_model import LogisticRegression from sklearn.feature_extraction.text import CountVectorizer from PIL import Image, ImageDraw, ImageFont import textwrap # 1️⃣ Load a tiny LLM (you can swap for any HuggingFace model) MODEL_NAME = "gpt2" tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME) model = AutoModelForCausalLM.from_pretrained(MODEL_NAME) model.eval() # 2️⃣ Load your tweets (or any text) – make sure it’s a simple .txt file TWEET_FILE = "tweets.txt" if not os.path.exists(TWEET_FILE): print(f"Create a {TWEET_FILE} file with your favorite sarcastic tweets.") exit(1) with open(TWEET_FILE, "r", encoding="utf-8") as f: tweets = [line.strip() for line in f if line.strip()] # 3️⃣ Train a sarcasm classifier (super‑quick, not perfect) vectorizer = CountVectorizer(ngram_range=(1,2), min_df=1) X = vectorizer.fit_transform(tweets) y = [1 if "sarcasm" in tweet.lower() else 0 for tweet in tweets] clf = LogisticRegression(max_iter=200).fit(X, y) # 4️⃣ Helper: decide if a sentence is “I can’t even” mood def ican_even(sentence): return re.search(r"\bcan't even\b", sentence.lower()) is not None # 5️⃣ Generate meme text def generate_meme(seed=""): inputs = tokenizer.encode(seed, return_tensors="pt") with torch.no_grad(): outputs = model.generate(inputs, max_length=50, num_return_sequences=1, pad_token_id=50256) text = tokenizer.decode(outputs[0], skip_special_tokens=True) # Clean up trailing prompt return text.strip() # 6️⃣ Render meme image def render_meme(text, filename="meme.png"): width, height = 500, 500 background = Image.new("RGB", (width, height), (255, 255, 255)) draw = ImageDraw.Draw(background) try: font = ImageFont.truetype("arial.ttf", 28) except: font = ImageFont.load_default() # Wrap text lines = textwrap.wrap(text, width=35) y_text = 150 for line in lines: w, h = draw.textsize(line, font=font) draw.text(((width - w) / 2, y_text), line, font=font, fill=(0, 0, 0)) y_text += h + 5 # Bottom caption caption = "I can’t even" w, h = draw.textsize(caption, font=font) draw.text(((width - w) / 2, height - h - 20), caption, font=font, fill=(255, 0, 0)) background.save(filename) print(f"✅ Meme saved to {filename}") # 7️⃣ Main loop def main(): while True: # Pick a random seed from your tweets seed = random.choice(tweets) meme_text = generate_meme(seed) # If the meme triggers the “I can’t even” vibe, show it if ican_even(meme_text): render_meme(meme_text) else: print(f"⛔️ Skipped: {meme_text[:60]}…") # Simple pause if input("Generate another? (y/n) ").lower() != "y": break if __name__ == "__main__": main() ``` ### How it works 1. **LLM**: GPT‑2 (tiny, but fast) spits out a short meme given a seed tweet. 2. **Classifier**: Uses `CountVectorizer` + `LogisticRegression` to flag sentences that look sarcastic. 3. **Mood filter**: If the generated text contains “I can’t even”, it gets rendered onto a 500×500 canvas with a red caption. 4. **Run**: Just hit `y` when prompted to keep the meme engine screaming. Feel free to swap `MODEL_NAME` for a bigger model or tweak the sarcasm trigger words. Let the chaos begin!