Uploading custom assets to OctoAI's Asset Library

OctoAI empowers you to customize images by leveraging assets like checkpoints, LoRAs, and textual inversions. You can either use:

  • Public assets in the OctoAI Asset Library
  • Upload your custom asset to the OctoAI Asset Library (private by default, and optionally public)
  • Or use OctoAI fine-tuning to create new assets - see Fine-tuning Stable Diffusion

This tutorial explains how to upload your own private assets to the Asset Library.

  1. First download the OctoAI CLI by following the instructions in CLI Installation. Check that it is properly installed by running the following in your terminal:
bash
$$ octoai asset --help
>Manage OctoAI assets
>
>Usage:
>  octoai asset [command]
>
>Available Commands:
>  create      Create asset
>  delete      Delete an asset
>  get         Get asset
>  list        List assets
>
>Flags:
>  -h, --help   help for asset
>
>Global Flags:
>      --log-level string   Set log level: fatal, error, warning, info, debug, trace (default "info")
>
>Use "octoai asset [command] --help" for more information about a command.
  1. Run octoai login to cache your token and authenticate to your account.
  2. We can now use the octoai asset create subcommand to upload assets (you can run octoai asset create --help to learn more on the options).
    1. --engine denotes whether this is an asset for SDXL or SD1.5
    2. --upload-from-file flag denotes the path of the file on your local machine that you’re trying to upload.
    3. --upload-from-url flag is an alternative to upload-from-file allowing you to upload an asset from a public URL; OctoAI will fetch and upload the file.
    4. —type can be lora, checkpoint, or textual_inversion (VAEs coming soon)
    5. --format denotes the format of your asset, which can be safetensors or pt.
    6. --datatype can be fp16, fp32, int4, or int8. For image gen, it should almost always be fp16, but for LLMs and other modalities your asset may have other datatypes.
    7. --name is a name for your asset. You can only use each asset name once.
    8. --transfer-api defaults to sts which is the fastest way to upload a large asset.
    9. If you are uploading a textual inversion, make sure to use the -w flag to denote the default trigger word for the asset. That trigger word can later be used at generation time to activate the inversion. For LoRAs, trigger words are optional.

As an example, you could use this command to upload a checkpoint from your local machine:

octoai asset create \
  --upload-from-file LCM_Dreamshaper_v7.safetensors \
  --name Dreamshaper_v7 \
  --engine image/stable-diffusion-v1-5 \
  --format safetensors \
  --data-type fp16 \
  --type checkpoint \
  --description "Dreamshaper v7"

You can alternatively upload the file via public URL using upload-from-url:

octoai asset create \
  --upload-from-url https://huggingface.co/SimianLuo/LCM_Dreamshaper_v7/resolve/main/LCM_Dreamshaper_v7_4k.safetensors?download=true \
  --name Dreamshaper_v7 \
  --engine image/stable-diffusion-v1-5 \
  --format safetensors \
  --data-type fp16 \
  --type checkpoint \
  --description "Dreamshaper v7"