---
tags:
- text-to-image
- lora
- diffusers
- template:diffusion-lora
widget:
- output:
    url: images/ComfyUI_04526_.png
  text: >-
    A blue Cleanbot wearing an orange backpack and vacuuming the kitchen floor.
    The Cleanbot has happy eyes. The background shows a modern kitchen with
    stainless steel stovetop and white cabinets.
- output:
    url: images/ComfyUI_04542_.png
  text: >-
    A blue Cleanbot with happy eyes looking at a neverending line of houses in
    the suburbs that stretches off to the horizon.
- output:
    url: images/ComfyUI_04564_.png
  text: >-
    A purple Cleanbot standing in front of an ornate palace, holding a broom and
    ready to clean the intricate architecture.
- output:
    url: images/ComfyUI_04579_.png
  text: >-
    A sepia photograph of an old-fashioned cast iron Cleanbot with steampunk
    fittings and a wooden backpack, holding a straw broom, standing outside of a
    prairie homestead. 
base_model: black-forest-labs/FLUX.1-dev
instance_prompt: Cleanbot
license: cc-by-nc-sa-4.0
---
# Cleanbot

<Gallery />


## Trigger words

You should use `Cleanbot` to trigger the image generation.

You can use `happy eyes` for black and white eyes, or `glowing [$COLOR] eyes` for glowing eyes with an optional color (defaults to blue).


## Download model


[Download](/ssube/flux-cleanbot-lora/tree/main) them in the Files & versions tab.

## Training parameters

```bash
export PARAMS_D_COEF=1.5
export PARAMS_LEARNING_RATE=1.0
export PARAMS_LEARNING_SCHEDULER=constant_with_warmup
export PARAMS_MAX_STEPS=1500
export PARAMS_NETWORK_ALPHA=16
export PARAMS_NETWORK_DIM=8
export PARAMS_NOISE_OFFSET=0.1
export PARAMS_NUM_CYCLES=1
export PARAMS_OPTIMIZER_ARGS='decouple=True weight_decay=0.1 betas=0.9,0.999 use_bias_correction=False safeguard_warmup=False'
export PARAMS_OPTIMIZER_TYPE=prodigy
export PARAMS_SAVE_EVERY_N_STEPS=100
export PARAMS_SAMPLE_EVERY_N_STEPS=5000
export PARAMS_SPLIT_QKV=False
export PARAMS_TIMESTAMP_SAMPLING=shift
export PARAMS_WEIGHT_DECAY=0.1
```

```json
{
        "datasets": [{
                "resolution": 1024,
                "enable_bucket": true,
                "min_bucket_reso": 512,
                "max_bucket_reso": 1024,
                "bucket_reso_steps": 128,
                "bucket_no_upscale": true,
                "batch_size": 3,
                "subsets": [{
                        "image_dir": "/mnt/training/datasets/friendly-robot-v01/refluxed",
                        "caption_extension": ".txt",
                        "num_repeats": 10
                }]
        }]
}
```

## Training dataset

1. Generated a detailed description of the Cleanbot concept
2. Generated 8 images showing Cleanbots doing various tasks
3. Enhanced images using reflux process
4. Trained with 10 repeats