Model is outdated

[aishatabubakarx]

How to change the model, i.e the model it is working on is outdated from 2024, there is a newer better model but it refuse to communicate with the newest model. is there a way to configure this

[Strvm]

How to change the model, i.e the model it is working on is outdated from 2024, there is a newer better model but it refuse to communicate with the newest model. is there a way to configure this

Hi @aishatabubakarx, what question are you asking it to find this out?

I went on https://meta.ai and got the same response as in this package:

[aishatabubakarx]

I mean this model uses llama 3 but the current model is llama 4. My main use was to generate images and the quality was significantly less than when used directly on the website, so i found out they use llama 4 but this model send message to llama 3 only. also I cant choose dimension, it stays at 1:1, Can you help with this too?

I can also share how my code is done, the main.py and automate api.

[aishatabubakarx]

The new model not as important as how to change image dimesion to 16:9 so all generated images come out in that dimension. I will be eternally greatful for your help

[Strvm]

I mean this model uses llama 3 but the current model is llama 4. My main use was to generate images and the quality was significantly less than when used directly on the website, so i found out they use llama 4 but this model send message to llama 3 only. also I cant choose dimension, it stays at 1:1, Can you help with this too?

I can also share how my code is done, the main.py and automate api.

And with code:

meta = MetaAI()
    resp = meta.prompt(
        "what llama model are you currently running and when were you last updated?",
        stream=False,
    )
    print(resp)

{'message': "I'm running Llama 4, and I was trained on data up to August 2024, but I'm not sure when I was last updated. You might find more info online if you look into it.\n", 'sources': [], 'media': []}

Maybe they provide a better model if you are logged in? The log-in process is currently broken in this project.

[aishatabubakarx]

Okay. Thank you. what about for the dimension for the image, can i configure all output to 16:9

[siddht4]

Actually you are using the newer model ; Meta AI has replaced the older one with the newer . The APi & calls are same ; so you wont notice any differnce . you can try this by 1. asking something though this meta_ai_api ; 2. directly inside Meta AI ; 3. through the Meta Ai in your whatsapp. You will see similar responses , confirming all the same head

[aishatabubakarx]

Thank you siddht4. You are right. I figured it out and fixed all the issues I had.

[siddht4]

Also the default size of the images are 1280 * 720 that is provided by META {Aspect ratio : 16:9}; other dimensions do exists but there is no cheat sheet stating dimensions for all ; so its more like a hit or a miss. still you can use ffmpeg

Basic Resize:

ffmpeg -i input.jpg -vf scale=1080:1080 output.jpg

Scale while keeping aspect ratio (padding with black bars):

ffmpeg -i input.jpg -vf "scale=1080:1080:force_original_aspect_ratio=decrease,pad=1080:1080:(ow-iw)/2:(oh-ih)/2" output.jpg

or PIL/ pillow to convert it however you like

from PIL import Image

def resize_image(input_path, output_path, width, height):
    with Image.open(input_path) as img:
        # Using LANCZOS for high-quality downsampling
        new_img = img.resize((width, height), Image.Resampling.LANCZOS)
        new_img.save(output_path)

# Example: Convert to Square (1080 x 1080)
resize_image("meta_default.jpg", "meta_square.jpg", 1080, 1080)