Some styles aren’t something we can describe in words, and it isn’t something the model would be trained on. We need a way to provide an image and tell the model “do something like this”.
We can use T2I-Adapter on top of Stable Diffusion to allow for a style image as an additional input.
Using the technique listed in Style Transfer From Text In Stable Diffusion, we just add one more step to provide a style image.
Example prompt: Sad Keanu in the style of minimalist line art
Model: AOM3+orangemix vae Sampler: Euler a, 20 steps 3 ControlNets:
- depth w/ original image
- openpose_full w/original image
- t2iadapter_style w/ style image
Prompt:
[basic prompt (e.g. man in jacket sitting)] lora:add_detail:-0.5 cartoon fanart style lora:cartoon_fanart_style:0.5 Simple line lora:20.50:0.5
Notice that the style image impacts both the color and content of the scene.
| Style Image | Output |
|---|---|
| [none] |  |
 |  |
 |  |
 |  |
.png) |  |