Pyramid Flow 100M
A compact text-to-image model
1 min read
Pyramid Flow 100M generates 256x256 RGB images from English text prompts. The model runs on both CPU and CUDA using the Hugging Face Transformers library. It is built on the T5-base text encoder and SD-VAE decoder, which download automatically on first run. This model is ideal for developers and enthusiasts looking for lightweight local image generation.
Key features
- Generation of 256x256 pixel images from English text prompts
- Support for CPU and CUDA execution across various hardware setups
- Use of proven components: the T5-base text encoder and SD-VAE decoder
- Batch image generation capabilities when passing a list of prompts
What you can do
- Generate standalone images from text descriptions in local scripts
- Run batch image generation to compare results with a fixed seed
- Test diffusion pipelines on low-end or standard hardware configurations
Example
import torch
from transformers import AutoModel
repo = "TheHuggerFacer/Pyramid-Flow-100M"
device = "cuda" if torch.cuda.is_available() else "cpu"
model = AutoModel.from_pretrained(repo, trust_remote_code=True).eval().to(device)
image = model.generate_image(
"A clockwork owl perched on a brass telescope, studio photograph",
seed=42,
num_inference_steps=20,
guidance_scale=4.0,
)[0]
image.save("clockwork-owl.png")Read more
DetailsGeneration resolution: 256x256 RGB
Supported devices: CPU, CUDA
Dependencies: T5-base and SD-VAE download separately
Limitations: object count, spatial relationships, and fine details may be unstable










Comments