Fastcam sa3

In this work we have used a high-speed microscopy system composed by a high-speed camera (Fastcam SA3, Photron, Tokyo, Japan), an inverted microscope (IX71, Olympus, Tokyo, Japan), a syringe pump (Harvard Apparatus PHD ULTRA, Holliston, MA, USA), and a syringe with a volume of 10 mL (Terumo, Tokyo, Japan).
The GUVs’ flow measurements were performed in a microchannel with a hyperbolic constriction followed by a sudden expansion. This geometry is widely employed to perform separation of blood cells from plasma and to assess changes in cell deformability, due to its ability to achieve a controlled extensional flow at the center of the microchannel contraction and to measure large numbers of cells in one single run [5 (link),21 (link),22 (link)].
A soft-lithography technique was used to fabricate the microfluidic device. The length of the hyperbolic contraction region (C) was 600 µm, the width of the microchannel (L) was 400 µm, and the contraction width was 15 µm (see detail in Figure 1). Note that all microchannels have a depth of about 15 µm.

To visualize deformation of the water surface by Halobates, we used a projection of color bands^{79 (link)}. We illuminated the surface of the water using a home theater projector (Epson EX9200 Pro) and a custom strip pattern. The projector was focused towards the camera at an oblique angle to reproduce the stripped pattern on the water surface. This specific setup showed the water surface as a dark plane with intense reflection. The position and movement of the insect were recorded with a color high-speed camera (Photron Fastcam SA3), typically at 1000 frames per second (fps). The insects were illuminated by a backlight diffuser from the back of the camera using a Sumita metal halide cool light source (Sumita Optical glass, Inc.). This resulted in the insects being seen as black shadows projected on the camera.

Uniaxial tensile tests were performed at room temperature on a tensile machine model 5566 (INSTRON, Norwood, MA, USA) with a 10 kN load sensor. Specimens were sprayed with white and matt black paint to obtain a random grayscale pattern. A camera (Fastcam SA3, PHOTRON Inc., San Diego, CA, USA) recorded the tests at 50 fps to perform digital image correlation (DIC) [16 (link)]. All specimens were tested until failure.
The specimens were split into three groups (G1, G2, and G3). Taking into account the tensile machine limitations, two different velocities (0.5 and 5 mm/s, corresponding to 8 × 10⁻³ and 8 × 10⁻² s⁻¹ strain rates) were tested to investigate the strain rate sensitivity in the material between G1 and G2. In addition, the influence of sandblasting was analyzed between G1 and G3. Ten specimens were tested for each condition (Table 1).
Three stress relaxations were included in the plastic hardening part for G1 and G3 at 3.3, 6.4, and 9.4% of true strain (obtained from preliminary tests) to track viscoplasticity phenomena. The constant strain was held during the 30 s to observe short-term relaxation phenomena. Relaxations could not be performed on G2 specimens due to machine limits on velocity.

All experiments were conducted on an inverted microscope (Olympus IX71, Japan). Cell focusing and excess aqueous phase removal were visualized with a 5× objective lens, while droplet encapsulation was observed using a 10× objective lens. Experimental video clips were recorded using a high-speed camera (FASTCAM SA3, Photron, Japan) set at 2000 frames per second (fps) and a 0.02 ms exposure time.
The standard deviation analysis^{17 (link),32 (link)} employed in this study was based on lab-developed Python codes and was used to visualize the trajectory of cells/beads by stacking 1000 high-speed camera frames into one.
Droplet detection and cell/particle counting within the droplets were performed using lab-developed Python code. The Hough gradient method in the OpenCV library³³ was used to detect droplets and measure their diameters in the microchannel. To count the number of cells/beads in each droplet, a deep learning model (YOLOv8n)³⁴ was trained using the droplet datasets generated in this study and employed to analyze all videos. The detected bead/cell numbers were labeled in video clips and can be viewed in Movies S1 and S2, respectively, in the supplementary materials. The source code and trained model are available in the supplementary materials.