Pixel-Level Manipulations

‎I was thinking about this; it's very difficult to select objects in blender in compositing and video editing without directly changing the mesh or going back to object mode or the modelling tab to make the changes, so I propose a Photoshop type perspective, where instead of modifying or selecting the whole mesh , we modify the images using tools and changes base on the camera view (similar to that of crease pencil ). Photoshop uses tools to make changes on images which is a little bit the same perspective, but in this case we use blender tools (from the modelling tab like shear, bevel(for image edge bevel), rotate, move, if possible edit mode tools types that can directly change, move and select image(with dots or pixels as vertices at their edges) for this purpose and for animation achievements without having to change a whole set-up) it's a shader node tool type that has it own material output socket for that particular modification. Is like having another form of image editor for selecting a particular portion, modifying image nature, shapes, using pixel as vertices on edges or on the entire object without having to go to the mesh, It usage is made for real time modifications and in compositing and the video sequencer, if possible modelling too, for directly editing images and videos without having to change the whole set-up. ‎In professional terms, I'm describing a "Spatial Compositor." Instead of traditional 2D compositing (which happens after the render) or traditional 3D modeling (which changes the geometry), my tool would perform Pixel-Level Manipulations based on 3D data, similar to how high-end "Deep Compositing" works in Nuke ‎We are moving away from Vertex Math and moving toward Raster Math (Photoshop's specialty). ‎The Perspective: In Photoshop, we use "Adjustment Layers" or "Selection Masks." ‎My Redesign: my Shader Node would act as a "Live Adjustment Layer" that uses the 3D depth and camera angle to know where to apply the effect, but it only modifies the RGBA buffer (the image) rather than the OBJ data (the mesh), when the node is plug out the changes on camera view are removed