jakedahn/flux-autumn-green

Flux Autumn Green

Prompt

screenprint tshirt design, a happy cat holding a sign that says "I LOVE REPLICATE", ATMGRN illustration style, green

Prompt

a woman, ATMGRN illustration style

Prompt

incredibly intricate and detailed illustrated abstract art iphone wallpaper, ATMGRN illustration style, green

Prompt

a fall landscape, ATMGRN illustration style

Prompt

A punk rock frog in a studded leather jacket shouting into a microphone while standing on a stump, ATMGRN illustration style, blue tint

Prompt

a penguin that is a car, ATMGRN illustration style, blue tint"

Trained on Replicate using:

https://replicate.com/ostris/flux-dev-lora-trainer/train

Trigger words

You should use ATMGRN to trigger the image generation.

Training details

This model was trained on Replicate, here: https://replicate.com/ostris/flux-dev-lora-trainer/train

The training set is comprised of 14 images generated on MidJourney using the --sref 2795713976.

You can find the entire training set, including auto-generated captions, and training images in the ./training_set directory.

Below are the training parameters I used, which seem to work fairly well for illustration/cartoony Flux LoRAs.

NOTE: This is 3200 training steps in total. The reason the steps parameter is 800, is because I did a batch_size of 4.

{
  "steps": 800,
  "lora_rank": 24,
  "batch_size": 4,
  "autocaption": true,
  "input_images": "training_set/2024-09-23-autumn-green.zip",
  "trigger_word": "ATMGRN",
  "learning_rate": 0.0003,
  "autocaption_suffix": "ATMGRN style"
}

Use it with the 🧨 diffusers library

from diffusers import AutoPipelineForText2Image
import torch

pipeline = AutoPipelineForText2Image.from_pretrained('black-forest-labs/FLUX.1-dev', torch_dtype=torch.float16).to('cuda')
pipeline.load_lora_weights('jakedahn/flux-autumn-green', weight_name='lora.safetensors')
image = pipeline('cat with a hat, ATMGRN illustration style').images[0]

For more details, including weighting, merging and fusing LoRAs, check the documentation on loading LoRAs in diffusers