Camping at the Swiss Alps
I used the same Multipass technique in Camping and applied a photorealistic checkpoint model as my second step — Virile Reality 3 Beta 3 in this case.
Although SDXL can render photography very well, I intentionally used paintings as the medium for Step 1, mostly because of availability of training data. If you want to start with rich colors, imaginative compositions, and lighting conditions similar to what you see in paintings, then it would make sense to start that in Pass 1, instead of trying to do all of these in post-processing. So that’s what I did.
Here’s a step-by-step of what I did:
Step 1. SDXL
- Pass 1
- SDXL
- 1024x1024
- Oil painting for color, composition, with traditional art technique keywords
Images
Step 2. Virile Reality 3 Beta 3
- Pass 2
- Virile Reality 3 Beta 3
- 1024x1024
- Remove oil painting keywords and replace with prompts for photography
Images
Step 3. Upscale
- Upscale to 4096x4096
- Use an AI upscale model
- 4x_NMKD-Siax_200k
Images
Step 4. FaceDetailer
- Fix issues with the face with a FaceDetailer node
- Use prompts specifically for the face
Images
Step 5. Post-Process
Post-processing gives the finishing touch to an image. Often people assume lots of things are applied — but in truth, most of the time, I only adjust a few crucial things. Specifically, I extract more tonal contrast from the figure and brightened up his face because he is against the sun, but the difference is very subtle.
- Adjust for color contrast, tonal detail, and other things
- Adobe Lightroom
Images
Images
All of these images went through the same steps described above.
Images
Tech
- SDXL, 30 steps, Euler a, SDXL VAE
- virileReality_v30BETA3, DPM++ 2M SDE, 30 steps, 0.5 Denoise, vae-ft-mse-840000-ema-prune
- Upscale: 4x_NMKD-Siax_200k
- FaceDetailer: face_yolov8m.pt
- Adobe Lightroom color correction and tonal adjustments