liveBook · Manning

Foundation model

As explained in Local Image Generation

A base model is the pretrained model selected as the foundation for further adaptation or fine-tuning.

Think of a foundational model as the starting canvas for fine-tuning: choose a base model, adjust it with training data, and save the resulting model for later use. section 15.3.3

When configuring a fine-tuning run:

  • Select Stable Diffusion 1.5 as the foundation, or begin with another fine-tune, including one created during an earlier iteration. section 15.3.3 section 15.5.2
  • Set the destination where the finished model will be saved; model files can require around 4GB or more. section 15.3.3
  • Keep float32 when possible, or switch to float16 to reduce disk usage while accepting some information loss and usually negligible quality reduction. section 15.3.3

Improving a base model through iterations

A Stable Diffusion 1.5 base generation may produce a disfigured human face at the start. After 10 epochs it improves noticeably, after 20 epochs it improves substantially, and by 30 epochs it becomes cleaner while beginning to reflect a different visual style. Training can then stop, use the result as a new base checkpoint, and continue with adjusted training data. section 15.5.2

Figure 15.9 We’ll be using base Stable Diffusion 1.5

Inspect the model-selection view showing Stable Diffusion 1.5 chosen as the base model. section 15.3.3

How Local Image Generation covers it

How deep it goes
Explains model selection, output destinations, and precision trade-offs, then stops short of detailed model internals.
Examples
Uses Stable Diffusion 1.5, alternative fine-tunes, and checkpoint iterations whose human images improve across 10, 20, and 30 epochs.
What it assumes
Assumes a fine-tuning workflow and model settings, then teaches base-model selection and precision choices directly.
What you can do after
Operate/configure a fine-tuning run by selecting a base model, output path, and weight and output data types.
How it connects
Connects fine-tuning with Stable Diffusion 1.5, model checkpoints, training data, epochs, and output data types.

Also explained in

Compare how 30 books explain foundation model →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Local Image Generation. The section links point to the parts of the book it draws on.