Font size conflict between document class and tikzpicture

Question

Here are my files reduces for the MWE :

file 1 :

\documentclass[a4paper,12pt]{article}
\usepackage{tikz}
   
\begin{document}

\input{Chaine_Fonctionnelle}

\end{document}

file 2 :

\documentclass[border=2pt]{standalone}
\usepackage[utf8]{inputenc}
\usepackage{tikz}


\begin{document}

\input{Chaine_Fonctionnelle}
 
\end{document}

Chaine_Fonctionnelle.tex :

% \tikzset{font=\fontsize{12pt}{14pt}\selectfont}
\begin{tikzpicture}[scale=.8]
        \node[line width=2pt, minimum width=5cm, text width=3.8cm,minimum height=5cm, rectangle] (Ali) at (0,0) {};
        \draw[line width=2pt] (Ali.east)--(Ali.west);
        \node[font=\large\bfseries, anchor=south, above = 1cm,align=center,inner sep=-18pt] at (Ali.south) {Redresseur  \\ (pont de diodes)\\ + filtre \\ (Condensateur)};
\end{tikzpicture}
    

here is the version from the article :

here is the version from the standalone :

I've tried a few things that showed me what causes the differences. For example I can remove the 12 points from the first line of the article, then the two images look the same, but I would like a solution that doesn't imply a change at this top line.

I've tried local changes like the commented line, but to no avail.

I thought the hierarchy was putting local indication above global indication, but I can't seem to overrule that 12 point from the documentclass command.

Can you help me ?

First Edition ---------------- :

Thank you for your answers and comments, I'll try make a few changes try and avoid some problems : avoiding negative value for inner sep, for that I'll anchor the text to the middle of the line, but the problem doesn't go away :

New file Chaine_Fonctionnelle.tex :

% \tikzset{font=\fontsize{12pt}{14pt}\selectfont}
\begin{tikzpicture}
        \node[draw,line width=2pt, minimum width=5cm, minimum height=4.2cm, rectangle] (Ali) at (0,0) {};
        \draw[line width=2pt] (Ali.east)--(Ali.west) node[midway] (MidAli) {};
        \node[font=\large\bfseries, anchor=north, align=center] at (MidAli) {Redresseur  \\ (pont de diodes)\\ + filtre \\ (Condensateur)};
\end{tikzpicture}

Here are the result, first from the standalone then from the article compilation :

You can see the two compilations still give different results.

Answer 1

Try using the following instead. This code uses the positioning library and places the node below the centre of the reference node, rather than above its bottom. It dispenses with the inner sep specification.

Note that scale=0.8 is doing nothing whatsoever. Try scale=0.1 and notice that the standalone image has precisely the same dimensions.

\begin{filecontents}[overwrite]{\jobname-1.tex}
\begin{tikzpicture}%[scale=.8] this has no effect
  \node[line width=2pt, minimum width=5cm, text width=3.8cm,minimum height=5cm, rectangle] (Ali) at (0,0) {};
  \draw[line width=2pt] (Ali.east)--(Ali.west);
  \node[font=\large\bfseries, anchor=north, below =.25cm of Ali.center,align=center]  {Redresseur\\(pont de diodes)\\+ filtre\\(Condensateur)};
\end{tikzpicture}
\end{filecontents}
\documentclass[a4paper,12pt]{article}
% \documentclass[border=2pt]{standalone}
\usepackage{tikz}
\usetikzlibrary{positioning}
\begin{document}
\input{\jobname-1}
\end{document}

EDIT

From my comment at the request of the OP.

One of the main advantages of using tikz/pgf or similar is that it ensures the fonts and spacing in your diagrams are consistent with those in the document.

There is no practical way to make such diagrams independent of the surrounding environment. To do that you'd have to essentially do class and package setup inside the tikzpicture. But a major reason to use \input (and TikZ) is to ensure the picture is sensitive to the context e.g. that the fonts are consistent with those used in the document, that the size of skips and font-dependent dimensions such as em and ex match etc.

If that's not what you want, compile your images separately and use \includegraphics. If you don't want the context-sensitivity, there's no disadvantage to \includegraphics. That's often the right choice for things like logos, but generally sub-optimal for things like diagrams.