Alternative 1: World texture (material preview mode)

It's possible to do it with a world texture:

World texture

Go to the shading tab, select the "World" and add an image texture:

• Coordinates must be from Window
• A mapping lets you adjust size and position
• Image texture must be set to "clip" instead of repeat

Visibility

It's visible from rendered and material preview modes, as long as you use "Scene World"

---

---

Alternative 2: Plane + Texture (solid mode)

Solid mode settings

For this to work in solid mode, we need to add a regular plane (mesh) and:

• Keep the plane at 0,0,0, facing up (you can scale in edit mode)
• Set its texture via nodes (without the BSDF node)
• Enable the "texture" visualization of the solid mode:

Dedicated camera

Then we will need to add a new camera, which we will call "ModelCamera". We will use this camera just for modeling. Set the local camera for the view:

This camera will allow us to keep the main camera unchanged

Camera navigation

Then we need to enter the camera view and allow navigation with zooming (click the padlock)

This will allow you to move around the same way you would normally, and the camera will follow. (Uncheck the padlock to make the camera frame bigger or smaller, check again to navigate)

Geonodes for plane

Then add these geonodes on the plane (remember the plane should be kept in 0,0,0, facing up to Z, you can scale the plane in "edit mode")

Note there is a "distance" input that controls how far from the camera you want the plane to be.

"Object info" here uses "Model Camera"

Result

You can move the position of the image by moving the plane in edit mode on X and Y

In addition to @unwave's answer, you can keep the Empty consistently locked to a specific position in the 3D Viewport by calculating the viewport's coordinates and updating the Empty’s location accordingly. This approach also allows you to move the Empty manually, and it will remain locked to its new screen-space position.

Here's a script that affects all empties whose names start with Empty_Reference:

import bpy import bpy_extras from mathutils import Vector depth = 10.0 empty_prefix = "Empty_Reference" class VIEW3D_OT_follow_viewport(bpy.types.Operator): bl_idname = "view3d.follow_viewport" bl_label = "Track Viewport Position" bl_options = {'REGISTER'} _timer = None _empties = {} # {object_name: last_location} _screen_positions = {} # {object_name: screen_pos} _suspend_frames = 0 def execute(self, context): self._empties = {obj.name: obj.location.copy() for obj in bpy.data.objects if obj.type == 'EMPTY' and obj.name.startswith(empty_prefix)} ctx = self.find_viewport_context(context) if not self._empties or not ctx: return {'CANCELLED'} region = ctx["region"] region_3d = ctx["region_3d"] region_3d.update() for name in self._empties: obj = bpy.data.objects.get(name) if obj: screen_pos = bpy_extras.view3d_utils.location_3d_to_region_2d(region, region_3d, obj.location) if screen_pos: self._screen_positions[name] = screen_pos wm = context.window_manager self._timer = wm.event_timer_add(0.01, window=context.window) wm.modal_handler_add(self) return {'RUNNING_MODAL'} def modal(self, context, event): if event.type == 'ESC': self.cancel(context) return {'CANCELLED'} if event.type == 'TIMER': ctx = self.find_viewport_context(context) if not ctx: return {'PASS_THROUGH'} region = ctx["region"] region_3d = ctx["region_3d"] region_3d.update() to_remove = [] for name in list(self._empties.keys()): obj = bpy.data.objects.get(name) if obj is None: to_remove.append(name) continue if obj.location != self._empties[name]: self._suspend_frames = 10 self._empties[name] = obj.location.copy() screen_pos = bpy_extras.view3d_utils.location_3d_to_region_2d(region, region_3d, obj.location) if screen_pos: self._screen_positions[name] = screen_pos for name in to_remove: self._empties.pop(name, None) self._screen_positions.pop(name, None) if self._suspend_frames > 0: self._suspend_frames -= 1 return {'PASS_THROUGH'} for name, screen_pos in list(self._screen_positions.items()): obj = bpy.data.objects.get(name) if obj is None: continue coord_2d = (screen_pos.x, screen_pos.y) view_vector = bpy_extras.view3d_utils.region_2d_to_vector_3d(region, region_3d, coord_2d) ray_origin = bpy_extras.view3d_utils.region_2d_to_origin_3d(region, region_3d, coord_2d) location_3d = ray_origin + view_vector * depth obj.location = location_3d self._empties[name] = obj.location.copy() return {'PASS_THROUGH'} def cancel(self, context): wm = context.window_manager wm.event_timer_remove(self._timer) def find_viewport_context(self, context): for window in context.window_manager.windows: screen = window.screen for area in screen.areas: if area.type == 'VIEW_3D': for region in area.regions: if region.type == 'WINDOW': space = area.spaces.active region_3d = space.region_3d return {"window": window, "screen": screen, "area": area, "region": region, "space": space, "region_3d": region_3d} return None def register(): bpy.utils.register_class(VIEW3D_OT_follow_viewport) def unregister(): bpy.utils.unregister_class(VIEW3D_OT_follow_viewport) if __name__ == "__main__": register() bpy.ops.view3d.follow_viewport()