Sink below the surface and you will be free.
All images are generated with Colab notebook included in the Quick Start Guide.
Prompt for underwater portrait
Positive prompt:
underwater photography portrait, dress, [amber heard: emma watson :0.6], floating hair, bubbles, sun light breaking through water surface, by ted grambeau
Negative prompt:
goggles, sun glasses, hand
| CFG scale | 7 |
| Sampling steps | 20 |
| Image size | 512×512 |
| Seed | -1 |
| Face restoration | Codeformer |
| Sampling method | DPM++ 2M Karas |
| Model | F222 |
| Recommended GUI | AUTOMATIC1111 |
Technical notes
- F222 model helped to generate body part relations correctly. It has high tendency of generating nudes though. The keyword dress is used to suppress nudes and swimming suits, so that the models are in unusual outfits. I specifically refrained from specifying what kind or color the dress is, so I can get some surprises.
- Perhaps the scene associates with beaches and swimming, it is quite easy to get swimming goggles and sunglasses. goggles and sunglasses are used in the negative prompt to suppress them.
- Hands are hard to fix so I used it as negative keyword to suppress it.
- Sunlight through water surface is used to both generate a stronger lighting effect and make the water surface visible, which is (to me) more visually interesting.
- keyword blending was used to blend two faces. The transition number needs to be adjusted carefully to get a good blend, and it can be slightly different for each image. The first actress not only sets the initial face, but also anchors the pose and the hair style in a particular way. You can experiment with different ones.
- Turning on face recovery is usually enough if the face is big enough. If the face is small, a second inpainting step at full resolution is needed.
- Ted Grambeau is an underwater photographer. I added him in the beginning to entice good aesthetics.