Python 3.12.12 : simple example with CompVis/stable-diffusion-v1-4 model on colab.

Because this year we need to start it as advanced and more prepared as we know and can do ...

Today I tested a simple source code with an interactive interface for text-to-image generation based on the CompVis/stable-diffusion-v1-4 model.

This is not an advanced model and you will have some dizzy images, but the learnning idea is the base of these colabs notebooks.

See the default example on my colab github project.

The colab notebook use this python version:

Python 3.12.12 (main, Oct 10 2025, 08:52:57) [GCC 11.4.0] on linux
Type "help", "copyright", "credits" or "license" for more information.

I'm mode advanced with some models like SDXL, image generation is not a priority at the moment ...

... and first image result is this:

Python 3.13.0 : Predicted XAU/USD with torch.

Testing the torch python package

import torch
import torch.nn as nn
import numpy as np

data = np.array([
    [1800.5, 1810.0, 1795.0, 1000, 1805.2],
    [1805.2, 1815.0, 1800.0, 1200, 1812.8],
    [1812.8, 1820.0, 1808.0, 1100, 1810.5],
    [1810.5, 1818.0, 1805.0, 1300, 1825.0],
    [1825.0, 1830.0, 1815.0, 1400, 1820.3],
    [1820.3, 1828.0, 1810.0, 1250, 1835.7]
])

X, y = torch.tensor(data[:, :4], dtype=torch.float32), torch.tensor(data[:, 4], dtype=torch.float32)
model = nn.Sequential(nn.Linear(4, 6), nn.ReLU(), nn.Linear(6, 4), nn.ReLU(), nn.Linear(4, 1))
optimizer = torch.optim.Adam(model.parameters())
loss_fn = nn.MSELoss()
for _ in range(3000):
    optimizer.zero_grad()
    y_pred = model(X).squeeze()
    loss = loss_fn(y_pred, y)
    loss.backward()
    optimizer.step()
prediction = model(torch.tensor([[1830.0, 1840.0, 1825.0, 1150]], dtype=torch.float32))
print("Predicted XAU/USD closing price:", round(prediction.item(), 2))

The result is :

python torch_001.py
Predicted XAU/USD closing price: 1819.57

Python 3.11.11 : Colab simple image to video with stabilityai - part 052.

Today, a simple test with artificial intelligence and stabilityai/stable-video-diffusion-img2vid-xt.

The result is not very good, but I believe the source code can be improved ...

pipe = StableVideoDiffusionPipeline.from_pretrained(
    'stabilityai/stable-video-diffusion-img2vid-xt',
    torch_dtype=torch.float16,
    variant='fp16'
)

pipe.enable_model_cpu_offload()

You can find the source code on my colab GitHUb projects - catafest_069.

Python 3.11.11 : Colab fixed catafest_0059 with NVIDIA models - part 051.

Fixed the colab notebook for test models from NVIDIA for waveglow and tacotron2.

Python 3.10.12 : Simple test with diffusers to create texture.

Today I tested on colab a simple python script to generate a texture using: diffusers, torch and gradio.

I upload the notebook on my colab repo from GitHub.

The script is simple and works good:

import gradio as gr
from diffusers import StableDiffusionPipeline
import torch

model_id = "dream-textures/texture-diffusion"
pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16)
pipe = pipe.to("cuda")

def generate_image(prompt):
    image = pipe(prompt).images[0]
    image.save("result.png")
    return image

iface = gr.Interface(
    fn=generate_image,
    inputs="text",
    outputs="image",
    title="Stable Diffusion Image Generator",
    description="Introduceți un prompt pentru a genera o imagine folosind Stable Diffusion."
)

iface.launch()

The result for the pbr winter terrain is this image: