How are class images generated if use_tx2img is set to false?

[whydna]

Where do the class images come from?

[78Alpha]

It uses the DiffusersPipeline from the normal diffusers. Putting a code snippet as the version I am running is older.

from diffusers import DiffusionPipeline, AutoencoderKL
if not self.use_txt2img:
            self.accelerator = accelerator
            if accelerator is None:
                try:
                    accelerator = Accelerator(
                        gradient_accumulation_steps=config.gradient_accumulation_steps,
                        mixed_precision=config.mixed_precision,
                        log_with="tensorboard",
                        logging_dir=os.path.join(config.model_dir, "logging")
                    )
                    self.accelerator = accelerator
                    self.del_accelerator = True
                except Exception as e:
                    if "AcceleratorState" in str(e):
                        msg = "Change in precision detected, please restart the webUI entirely to use new precision."
                    else:
                        msg = f"Exception initializing accelerator: {e}"
                    print(msg)
            torch_dtype = torch.float16 if shared.device.type == "cuda" else torch.float32

            self.image_pipe = DiffusionPipeline.from_pretrained(
                config.pretrained_model_name_or_path,
                vae=AutoencoderKL.from_pretrained(
                    config.pretrained_vae_name_or_path or config.pretrained_model_name_or_path,
                    subfolder=None if config.pretrained_vae_name_or_path else "vae",
                    revision=config.revision,
                    torch_dtype=torch_dtype
                ),