Begin with the largest value masses
Reduce the photograph to two values and you have a notan, which answers the only question that matters before anything is mixed: does the design hold up as an arrangement of light and dark. Most paintings that fail were already failing here, and the failure is invisible once colour is on the canvas.
From two, the value study runs up to forty five bands. The useful range for a block-in is between three and seven. More than that stops being a plan and becomes a picture of the photograph, which is the thing a block-in exists to avoid.
Keep value ahead of colour variety
Colour families are assigned after the value structure is settled, not alongside it. Two colours that read as the same value get the same number even where their hues differ, because on the canvas they will occupy the same role.
The palette is a ceiling rather than a target. Asking for more colours never produces two swatches you would mix identically, so raising the number widens the range available instead of adding work at the palette.
Edges carry the relationships
Hard, soft and lost edges are noted separately from the shapes themselves. A block-in that draws every boundary at the same weight has thrown away the information that says which shapes belong to each other and which are meant to separate.
A lost edge is a decision to let two masses merge. It is worth having marked on the plan, because it is the first thing that gets accidentally repainted back in.
Where a generated block-in stops helping
A block-in built from a photograph inherits the photograph's mistakes. A camera flattens value range in shadow and blows it out in light, and the guide reproduces both faithfully, because it has no way to know which parts of the picture you would have seen differently standing there.
So the plan is a starting position rather than a verdict. Where the photograph has gone dense and black, the guide will offer one mass, and the painting usually needs two. Reading the guide as a proposal is the difference between using it and copying it.
It is also the wrong tool for a subject you already know how to organise. The value of a block-in is highest on the pictures that are hard to see, and close to zero on the ones you could lay in from memory.
Transfer with checks, not assumptions
Transfer is where a good plan usually goes wrong. A grid drifts on a large canvas, and a projector throws a trapezoid unless the geometry is corrected for the angle it is standing at.
The projector view corrects for that with four corners you place on the four real corners of the surface, and a scale bar to measure before anything is drawn. Both are manual on purpose. The room is not something the software can see, and a check you perform is worth more than a calibration it claims to have done.
For anything above about a metre, the actual-size poster tiles are the more reliable route, because paper does not drift when somebody walks past the projector.