this is a version of the “generate images & weights” but using on diffusers 0.11.1

This is a prototype Trainer that does return a model, as well as generates a list of images from a json with inputs (name, prompts, …) you provide.

Each of the images that either pass the nsfw safety check or have disable_safety_check set to True are returned

This is a prototype Trainer that does return a model, as well as generates a list of images from a json with inputs (name, prompts, …) you provide.

Each of the images are returned

This is a prototype Trainer that doesn’t return a model, instead, it only generates a list of images from a json with inputs (name, prompts, …) you provide.

Each of the images are returned (unlike the previous version which zipped up the images)

DO NOT USE THIS WITH THE DREAMBOOTH TRAINING API!

This is a prototype Trainer that doesn’t return a model, instead, it only generates a list of images from a json with inputs (name, prompts, …) you provide.

Those images are zipped up and returned instead of weights.

DO NOT USE THIS WITH THE DREAMBOOTH TRAINING API!

This is a prototype of a trainer that can be used to trigger a rebuild of dreambooth-inference model without re-running the training.

You send in weights with this trainer version, and it will simply return the same url. This allows you to use the dreambooth-api to re-build the inference model - to get any improvements (such as img2img)

this is a prototype version that lets you provide a hf_model_name and revision start training from it

This prototype trainer only works on public HF models.

This is a prototype to start training from the results from an existing training.

By providing initial_weights that are the output.zip url generated from your first dreambooth training, you can continue training. note: this is confusing - because replicate’s dreambooth api doesn’t expose weights in an easy to find way. You need to find the “training” prediction that generated the weights. And send that url.

Please share feedback with me if this works for you… dreambooth training on an existing dreambooth training isn’t known to produce good results.

This is a prototype trainer_version for the dreambooth api that support training on multiple concepts based on filename.

This prototype takes the instance_prompt and combines it string extracted from instance_data image filenames.

If instance_prompt is “a photo of” and the your instance_data data.zip has filenames, it will use the image with the prompt:

• foo bar.jpg -> “a photo of foo bar”
• bar_baz.jpg -> “a photo of bar baz”
• “foo baz-123.jpg” -> “a photo of foo baz”

If this works/fails - please provide feedback on the github PR