Premiere 2021

Unless stated otherwise, images were processed using Fiji⁵³ (link) and Matlab 2020b (Mathworks) using custom codes. Figures were assembled in Adobe Illustrator 27.3.1. Movies were edited in Adobe Premiere 2021.
Spatial drift during acquisition was corrected using a custom GPU-accelerated registration code based on cross correlation between successive frames. Drift was measured on one channel and applied to all the channels in multichannel acquisitions. Code is available on our github page (https://github.com/deriverylab / GPU_registration).
For representation purposes, all confocal images of Par clusters in colocalization figures (Figures 1, 2, 3, S1, and S4–S8) were processed with a Wavelet “à trous” filter (custom GPU-accelerated MATLAB port of a code originally developed by Fabrice Cordeliere for the “Improve Kymo” ImageJ plugin⁶⁴ (link) and the raw image was averaged with the filtered one to generate the figure panel. Code is available on our Github page (https://github.com/deriverylab / GPU_wavelet_a_trous). Note that the colocalization analysis (see below) was performed on the raw 3D stacks.

Images were processed using Fiji^{70 (link)} and MATLAB 2020b (MathWorks) using custom codes that are available on request. For visualization purposes, the PopRed lookup table from the J. Manton collection (https://github.com/jdmanton/ImageJ_LUTs) was applied to most monochrome images after the dynamic range was adjusted between minimum and maximum grey values of each image (note that the dynamic range was not kept identical between images when presenting different conditions). Figures were assembled in Adobe Illustrator 2021. Videos were edited in Adobe Premiere 2021.
Where appropriate, spatial drift during acquisition of videos was corrected using a custom GPU-accelerated registration code based on cross-correlation between successive frames. For representation purposes, a wavelet ‘à trous’ denoising filter was applied to Extended Data Fig. 8c (custom GPU-accelerated MATLAB port of a code originally developed by F. Cordeliere for the Improve Kymo ImageJ plugin^{81 (link)}). The raw images were averaged with the filtered video. Both codes are available at our GitHub page (https://github.com/deriverylab), as well as the codes for quantification of protein condensation and nuclear segmentation described below.

Photographic and video footage was filmed using a Nikon (Melville, USA) D3300 DSLR Camera with 18–55 mm and 35 mm lenses. Micrographic footage was captured using this camera in addition to Wild M240 Makroskop and Olympus BH2 optical microscopes. Additional footage was provided by Vera Meyer and Stephan Starke (TU Berlin). All footage was edited using After Effects 2021 and compiled using Premiere 2021, both applications from Adobe Creative Cloud (San Jose, USA). Drawings and illustrations were created using Savage Interactive Procreate (Hobart, Australia) and Adobe Illustrator 2021 (San Jose, USA). Animations were created and edited using After Effects 2021. Music was composed and generated using a Yamaha P115 Electronic Keyboard (Hamamatsu, Japan) and recorded, edited, and mixed using Apple GarageBand for Mac (Cupertino, USA).