AI model

$1.50 / fine-tune

Fine-tuning is the process of taking a neural-network AI model which is responsible of generating an image, and training it with a new concept. The product of this process is a 4GB new model file which contains the knowledge of the new concept.

8 Images

$0.10 / prompt

Prompts are texts that are used together with the fine-tuned model to generate a set of images.

Each prompt defaults to 8 images

Video

$0.40 / 100 frames

A video at 24 frames per second. Use frame: prompt to generate a video in the prompt text box.

Login

Details

  • Flux1 training is $5.00 per 1000 steps. We default to 100 steps per training image. You may override the number of steps.
  • Flux1 inference is $0.23 per 8 images.
  • super-resolution or face correction will default to 4 images, or if num_images is specified, will double the cost.
  • --tiledupscalev2 - 7 cent per image
  • --tiled_upscale - 2 cent per image
  • --only_upscale - 2 cent for 512x512 image. Larger images are pro-rated.
  • Prompt Masking 1 cent.
  • --fix_bindi 1 cent.
  • Multi-person will consider each person as well as the base-prompt as a separate prompt for pricing purposes.
  • steps above 50 will increase cost respectively. e.g: 100 steps will cost 2x. Generally it is not advised to use more than 50 steps.
  • steps For Flux above 28 will increase cost respectively. e.g: 56 steps will cost 2x. Generally it is not advised to use more than 28 steps.
  • Minimum credit-card processing amount is $5.00
  • When making a payment, the amount is added to your account balance. Usage is deducted from your account balance.
  • Models are saved for 30 days since the moment they become available.
  • Extended model storage - $0.50 per model per month.
  • Nude is not allowed.
  • Astria does not provide support for using external tools with the checkpoints however you can find some tips in our Pros page.
  • For non API users, a CompVis CKPT format checkpoint and HuggingFace Diffusers TAR are provided.
  • For API users, if you require a CKPT - please contact support.