Facing: Front ↔ Back Orientation Slider
Facing is a bidirectional Concept Slider for controlling whether people orient toward or away from the viewer.
Ever find yourself prompting “face visible” over and over, but your base checkpoint keeps turning your subject away? Or have the opposite problem—wanting a subject to look out over the amazing city you just put in front of them instead of turning toward the viewer to mug for the camera? Ever hand a subject a task only to see them avoid it entirely to stare at you?
Facing aims to solve that problem.
Positive weights encourage front-facing / viewer-facing subjects. Negative weights do two things: from -1 to -3 the LoRA progressively encourages profile, turned-away, and rear-facing solutions, but the cool quirk is what happens at negative weights between -0.25 and -1.0.
Weights between -0.25 and -1.0 tend to break checkpoints' strong prior for camera-facing and gaze, allowing your subjects to turn to whatever task you've assigned them. A person building a campfire looks at what they're doing instead of staring at the camera dumbly with a piece of wood in their hand. Two people gazing into each other's eyes don't break eye contact to acknowledge a camera present. A fortuneteller gazing into her crystal ball will actually gaze into her crystal ball.
So, the useful part isn't simply generating a picture of someone's back. Facing can change orientation while preserving surprisingly complex poses, activities, scenes, and groups. Standing subjects are easy, but the slider also works with sitting, kneeling, crouching, crawling, running, and multi-person compositions.
Facing is not a geometric rotation control. Different poses and prompts have different transition points, and the change is often stepped rather than smoothly interpolated. As negative strength increases, the checkpoint may try several plausible side/profile orientations before settling into a fully rear-facing solution.
That behavior is intentional to the concept: the slider applies directional pressure while the base checkpoint determines a plausible pose and orientation.
Typical behavior
Positive: stronger front/viewer-facing orientation
0: normal checkpoint behavior
Negative: profile → turned away → rear-facing
A good starting range is approximately -1 to +1. For strongly rear-facing subjects, try progressively stronger negative values. Depending on the pose, a complete rear-facing solution may require roughly -1.5 to -3, with difficult poses sometimes responding at still stronger weights.
Testing found substantial stability even at strong weights; don't be afraid to explore beyond the usual ±1 LoRA range.
Pose-dependent examples
In testing, rear-facing endpoints appeared approximately around:
standing: -1.75 to -2
crouching: -1.75 to -2
sitting on a stool: ~ -1.5
kneeling: -2.75 to -3
crawling: -3 or stronger
These are examples, not fixed settings. Prompt, seed, checkpoint, composition, and pose all matter.
Groups
Facing also works surprisingly well on groups. Increasing negative pressure can progressively turn individual people away until much or all of the group becomes rear-facing.
This makes it useful for compositions such as crowds, teams, people watching events, groups looking toward scenery, and other situations where SDXL would normally prefer to turn everyone toward the camera.
Usage tips
Keep the prompt describing the scene and activity you actually want. You often don't need to fight the checkpoint with a pile of orientation language.
Start at -1, then increase negative strength until the desired orientation appears. If the subject stops at profile or three-quarter rear view, keep moving negative—the transition can occur in distinct steps.
For front-facing correction, use positive weights.
Because Facing affects orientation rather than simply inserting a memorized pose, the exact endpoint varies naturally with the scene.
Recommended: sweep the weight when finding the best value for a new pose rather than assuming one universal strength.
Base tested: SDXL / Juggernaut XL Ragnarok.