---
tags:
- text-to-image
- stable-diffusion
- lora
- diffusers
- template:sd-lora
widget:
- text: <lora:4lph4bet-next_v040:1> 4lph4bet, typography, serif, simple, regular
  output:
    url: images/00057-2246810227.png
- text: <lora:4lph4bet-next_v040:1> 4lph4bet, typography, sans, fluid
  output:
    url: images/00068-2246810230.png
- text: <lora:4lph4bet-next_v040:1> 4lph4bet, typography, serif, round, brush
  output:
    url: images/00070-2246810228.png
- text: <lora:4lph4bet-next_v040:1> 4lph4bet, typography, serif,bold,playful
  output:
    url: images/00078-2246810228.png
- text: >-
    <lora:4lph4bet-next_v040:1> 4lph4bet, typography, sans, light, geometric,
    abstract
  output:
    url: images/00083-2246810229.png
- text: <lora:4lph4bet-next_v040:1> 4lph4bet, typography, retro, geometric
  output:
    url: images/00095-2246810228.png
- text: <lora:4lph4bet-next_v040:1> 4lph4bet, typography, retro, (geometric:.5)
  output:
    url: images/00112-2246810230.png
- text: >-
    <lora:4lph4bet-next_v040:1> 4lph4bet, typography, fairy, fantasy, floral,
    flower, art nouveau 
  output:
    url: images/00218-3253817665.png
- text: >-
    <lora:4lph4bet-next_v040:1> 4lph4bet, typography, brutalist, edgy, sharpy,
    boxy
  output:
    url: images/00236-57461198.png
base_model: runwayml/stable-diffusion-v1-5
instance_prompt: 4lph4bet, typography
license: cc-by-sa-4.0
---
# 4lph4bet-next_v040 Font Generation LoRA

<Gallery />

## Model description 

# 4lph4bet-next_v040 Font Generation LoRA

This LoRA was created using a custom dataset to finetune Stable Diffusion 1.5. This LoRA It will generate a grid of  88 characters in a consistent style and can be used for font generation. There is a [Script](https:&#x2F;&#x2F;github.com&#x2F;414design&#x2F;4lph4bet_processor&#x2F;) for post processing and font creating using [Calligraphr](https:&#x2F;&#x2F;www.calligraphr.com&#x2F;).

## Use 

Just add the LoRA to the prompt and us the following parameters as a baseline for further iteration:

* Trigger words: `4lph4bet`, `typography`
* Sample steps: `15–35`
* CGF-Scale: `5–9`

The model is trained on a variety of tokes to be able to specifically generate among other characteristics:

* `sans`, `serif`
* `light`, `regular`, `bold`
* `italic`

It is recommendes to use *Hires fix* for initial upscaling. Example parameters: 

* Upscaler: `R-ESRGAN 4x+ Anime6B`
* Hires steps: `10`
* Denoising strength: `0.2`
* Upscale by: `2`

## Post processing

Use this Python [Script](https:&#x2F;&#x2F;github.com&#x2F;414design&#x2F;4lph4bet_processor&#x2F;) to create a ready-to-use template for [Calligraphr](https:&#x2F;&#x2F;www.calligraphr.com&#x2F;). Before running the script it is recommended to upscale the font grid to 4096 by 4096 pixels. 

Using [stable-diffusion-webui-vectorstudio](https:&#x2F;&#x2F;github.com&#x2F;GeorgLegato&#x2F;stable-diffusion-webui-vectorstudio) for [stable-diffusion-webui](https:&#x2F;&#x2F;github.com&#x2F;AUTOMATIC1111&#x2F;stable-diffusion-webui) the grids can be vectorized directly to speed up use in traditional font creation workflows.

## License

This LoRA is provided  under a *CC BY-SA 4.0* license. 


## Download model

Weights for this model are available in Safetensors format.

[Download](/414design/4lph4bet-next_v040/tree/main) them in the Files & versions tab.