Dance / Run - Blue Yellow
Dance / Run
I love to dance, and enjoy watching dancing performances, so I have been wanting to make some images about dancing for a while. I have made some images of ballet before — and if you follow my Instagram stories, you would’ve seen one of two before. If you haven’t, I posted one of them to the “One Shot” category.
The tricky thing about depicting ballet is that the AI engines will put them in ballet shoes, but often the female version instead of the male versions. And so the images always look awkward. If I render the dancers without shoes, then it creates another issue, which is that AI engines are terrible with feet.
So today as I tried to work on this project again, I decided to put the dancers in sneakers instead. Then I gave all of them running shorts. I wanted a relatively clean and coherent background but didn’t want just something plain, so I made some abstract paintings inside Midjourney and fed them to Stable Diffusion as part of two control nets: Ref only and Shuffle — a workflow I have come back to time and time again in order to control style and color palette.
The resulting images give the right balance of energy to the photorealistic paintings, while keeping the figures in focus. They actually looked a bit like an advertising campaign for Adidas!
In order to get very detailed brush strokes for the background, I rendered all of these images in fairly high settings (50 steps + 30 hi-res steps), but I think that the results are worth wait.
30 images on IG, 31 images on gymdreams8.com — I had to remove a few of these last minute so the number on the Instagram is not correct.
Text prompts with Midjourney (for Control Nets) + Text prompts with Stable Diffusion: 50 steps, DPM++ SDE (not Karras), CFG 5, 512x512, Virile Reality v2, Denoising 0.5. CN0 Reference Only. CN1 Shuffle. ADetailer mediapipe_face_full. Hires 2x (1024x1024), 30 steps, 4x_NMKD-Siax_200k. Post: Topaz Gigapixel HQ 4x (4096x4096), Adobe Lightroom color correction
Images
Images
Image Index
Technical Details
Text prompts with Midjourney (for Control Nets)
Text prompts with Stable Diffusion
- Steps: 50
- Sampler: DPM++ SDE
- CFG scale: 7
- Size: 512x512
- Model hash: 63f0231527
- Model: virileReality_v20
- Denoising strength: 0.5
- ADetailer model: mediapipe_face_full
- ADetailer prompt: handsome french face
- ADetailer confidence: 0.3
- ADetailer dilate/erode: 4
- ADetailer mask blur: 4
- ADetailer denoising strength: 0.4
- ADetailer inpaint only masked: True
- ADetailer inpaint padding: 32
- ADetailer version: 23.7.10
- ControlNet 0:
- preprocessor: shuffle
- model: control_v11e_sd15_shuffle, 526bfdae
- weight: 1
- starting/ending: (0, 0.5)
- resize mode: Crop and Resize
- pixel perfect: False
- control mode: Balanced
- preprocessor params: (1024, -1, -1)
- ControlNet 1:
- preprocessor: reference_only
- model: None
- weight: 1
- starting/ending: (0, 1)
- resize mode: Crop and Resize
- pixel perfect: False
- control mode: Balanced
- preprocessor params: (-1, 0.5. -1)
- Hires upscale: 2
- Hires steps: 30
- Hires upscaler: 4x_NMKD-Siax_200k
- Lora hashes:
- add_detail: 7c6bad76eb54
- epi_noiseoffset2: d1131f7207d6
- muscle2: 58b17ca61429"
- Version: v1.3.2
- Post:
- Topaz Gigapixel HQ 4x, 4096x4096