cjwbw / distilgpt2-stable-diffusion-v2

Descriptive stable diffusion prompts generation using GPT2

  • Public
  • 392 runs
Demo API Examples Versions (d53a494f)

Run time and cost

Predictions run on Nvidia T4 GPU hardware. Predictions typically complete within 90 seconds. The predict time for this model varies significantly based on the inputs.

weights from: https://huggingface.co/FredZhang7/distilgpt2-stable-diffusion-v2 cog implementation: https://github.com/chenxwh/cog-themed-diffusion/tree/distilgpt2

DistilGPT2 Stable Diffusion V2

This model was trained on 2,470,000 descriptive stable diffusion prompts on the FredZhang7/distilgpt2-stable-diffusion checkpoint for another 4,270,000 steps.

Compared to other prompt generation models using GPT2, this one runs with 50% faster forwardpropagation and 40% less disk space & RAM.

Major improvements from v1 are: - 25% more variations - faster and more fluent prompt generation - cleaned training data * removed prompts that generate images with nsfw scores > 0.5 * removed duplicates, including prompts that differ by capitalization and punctuations * removed punctuations at random places * removed prompts shorter than 15 characters