Michael Limberger
Need me? Email mike@limberger.ca
AI
Understanding Framing
How much of the person
Framing is how much of the person shows up. This matters a lot when you do not have hires fix.
The ladder of shots
| Framing | What you see | Eye quality |
|---|---|---|
| close-up | Face only | Excellent |
| bust | Head and shoulders | Excellent |
| upper body | Head to mid-torso | Very good |
| cowboy shot | Head to mid-thigh | Good |
| thigh up | Head to thighs | Acceptable |
| full body | Everything | Needs hires |
The eye problem
SDXL struggles with eyes when they are tiny.
Full body at 832 by 1216? The face is maybe 5 to 10 percent of the image. Not enough pixels for clean, symmetrical eyes.
Hires fix normally solves this: upscale 2x, run a detail pass. On Apple Silicon, that detail pass often trashes the image instead of helping it.
Rule of thumb: no hires fix? Stay at cowboy shot or tighter.
Resolution by model
| Model | Base resolution | With hires (2x) |
|---|---|---|
| Bridgetoons | 672x1200 | 1344x2400 |
| Cyberrealistic | 832x1216 | disabled |
| Realism | 832x1216 | disabled |
| Arthemy | 1024x1344 | disabled |
Arthemy has the biggest base resolution, and with hires fix it handles full-body detail well.
Pick the shot
Need full body? Bridgetoons, period. Tight portrait or cowboy shot? Any model works.