blog

Deborah Fowler, Ph.D.

Billy Dongyoon Jang

Thanks to Travis Harkleroad for showing me the substep menu option, which I set to 0.25 to calculate subframes, and set the DOP substeps to 4. He also told me to use the animated=1 attribute instead of a deforming collsiion object, and to avoid concave hulls. I re-did many of my collision objects accordingly.

Feedback we received: Todd: It's looking better. There's something off about that car animation and how it interacts with the chocolate chips, seems a bit unnatural. The chips also get really still, you probably want some tumbling off motion afterwards. There's a spot by the lid in that same shot where there's an opportunity for the car to kick up some powder or something that's missing there, following this theme. The chocolate shader is pretty dark, usually it's a warm brown to subtle purples to emphasize the tastiness.

After scaling up the scene by a factor of 27 (and the time scale), the sims take about 10 minutes! I also cranked the constraint iterations in the bullet solver to around 16000 to cover my semi-complex objects. I lowered my substeps from 12 to 5. I also added in relevant set dressing objects.

Feedback we received: Todd: The collision issues may be involved with scale; somewhere around 5cm the bullet solver just doesn't work as expected, so scale up time and space. Constraint Iterations also are very important for collision calculations. Sometimes multiple duplicates of the collision objects can enforce the weight. Also avoid deforming objects, use geometrywrangle and manipulate attributes to stabilize the motion. Kyle: The top down shot looks good for the added shot, I'd like to see the animation for it next week. It's also the best color graded shot, so you should match the others more closely to that shot. A lot of motion blur and depth of field will help. That's better in the render, but if it's too expensive it could be done in compositing.

Lots of collision woes. I'm using an altered base from my flour (plan A) technique. A reduced 3D scan of our kitchen counter location is my ground plane. I also tried using volume based collisions. These sims took around 8 or 9 hours!! Deborah guided me in using the bullet solver with a multisolver so I could manipulate finer attributes with a geometry wrangle node. I use polyreduced geo for the sim, import those back to the SOP level with a dop I/O set to represent objects with points, and copy the heavier geometry to those.

Feedback we received: Todd: So you've got three main things going on; the main hit, slumping, and air material in the flour hit. I'm still not sold on the main hit. Shoot a bunch of reference video of the car hitting flour at different spots. The grain solver doesn't have great volume conservation. Kyle: There hasn't been enough progress on this project. Get the layout in full edits of the project. Billy: Render objects in place. Get in more detail and roughness of objects. You're headed in the right direction.

With a Drift Threshold of 0.14 and Kinetic Scale of 0.7 and particle separation of 0.0008, I achieved a high res sim that held up but was too stiff. Once I added the tentative animation alembic, the car volume was waking the particles but not colliding with them. Pre-alembic I was using a method of turning the geometry active and not active based on the object's keyframe positions, but this setup did not translate to the .abc. For the render tests, adding point Alphas increased render times by 20 to 40 percent. But on the render farm, each HD frame of semi-transparent particles (~8M) took just less than 10 minutes. The .abc collider almost fixed! At least it’s not crashing the program anymore! All I had to do (maybe) was uncheck “Create Active Object” I think it is barely colliding perhaps because of a collision padding value I changed during the troubleshoot process. This sim has about 4.5M particles and took 3.5 hrs.

Feedback we received: Todd: Combining resolutions is a good idea. Consider having the top stuff live with a smaller stiff mound underneath. Upres sooner than later so you can see what you can get away with. Billy: It's 60% done so all that's left is execution. You have a good layout so far. The packing peanut needs more shape detail. Start getting detailed asset textures from references, perhaps with Substance. Achieve subtle details, like a stretched specularity on the mixer. To tackle the reflections issue in the glass bowl, try texturing a counter proxy.

The first clip is a FLIP version of a flour pile with very high viscosity attributes animated with a volume. For troubleshooting the collapse of the pile I used sleeping (the blue particles above) and auto-wake in the POP Solver so that the pile only starts simulating after contact with the car volume. There's still excess collapse a few frames after contact. My focus has been on Internal Collision Stiffness, Kinetic Scale, and Drift Threshold parameters in the Grain Update DOP as well as the pop solver's substeps. I may still need to increase Drift Threshold and decrease Scale Kinetic values. The pile started exploding (I'm not sure why), which I fixed by turning down the Max Neighbor Query from 2.5 to 1.5 in the grain update. To save on calculations and perhaps lift up the pile (per Professor Fowler's suggestion) I hollowed a bit of the source volume and used that hole as a static collision object. For some reason this static object appears to be ignored in the simulation. I began setting up a pyro simulation advecting from particle data and plan on isolating fast-moving particles into a group to advect pyro with a set-up similar to the one below.

Feedback we received: Todd: Don't tackle it all in one big sim. The trick is bridging together multiple simulations. Grains are notorious for not scaling. Think of the airfield you'll need to get the floating particles effect. I wouldn't render out a volume but rather multiple layers of semi-opaque particles. Kyle: The reshoots and glass bowl look great. Start placing objects in the scene. Billy: Focus on the layout and previs with camera tracking to get an idea of which objects you need and where they will be placed in the scene first so you don't waste time shading things you may not use.

Once I increased resolution, the shape started deflating. After testing different parameters, the notable ones seem to be Scale Kinetic and Drift Threshold in the Grain Update. But neither of those values tweaked solved the full issue as much as increasing the solver substeps did. Other solutions I attempted were Solid Grains, a constraint applied based on the source shape. This started to give me an accurate shell but still fell apart at higher resolutions. I also tried auto sleep options which I think can work but needs to be tweaked more.

Feedback we received: Kyle: The 3D Scan is awesome. Exclude the shots with a person because it looks like a student film. I'm not a fan of the party hats and the objects don't look like they belong (they're too arranged). Utilize the round table or the stove to extend the length of some of the shots. Clean up the set a bit. Get something see-through to sell the compositing. Billy: (echoing with Kyle) Reshoot without any objects. Use a macro lens and perspective. Build in the rest of the set and objects in CG. Think about the story development, a cute car could be a part of a cuter story. With this much done in two weeks you could potentially get a lot done.

For these tests I scaled the scene down to be closer to our miniature space size (~10cm) and added impact to the side as in our shot plan. Some of the motion remains weird and wonky which I address below.

Here I refined the motion more mostly by changing the Scale Kinetic Friction to 2 in the POP Grains Behavior tab. I also set the Constraint Iterations to 40, Internal Collision Stiffness to 30k, Clumping Weight to 0.2 and Clumping Stiffness to 1. Changing the Drift Threshold to 0.012 in the POP Grains Solver tab helped stop much of the excessive slumping. With the particle separation at 0.0022, there are 76k points. In the second half of this test, I try out the Point Replicate Node to multiply that amount by 10 and try to add some noise. However it looks clumpy in a bad way, probably until I decrease the Particle Sep further.

Particle Separation here is 0.0016 (200k points) with a point replicate of 15. Going forward I hope to upres it, possibly excluding the still half of the flour pile. I want to test out POP Replicate on isolated points as well as pyro advection. Dale Bunten showed interesting results from FLIP foam using point replicate that looked similar to flour that I can explore further.

The beginning of this masterclass inspired me to use VOP point cloud filters (below) to isolate specific particles to advect pyro from later based on position and neighbors. I could also filter based on velocity or a hit attribute.

Feedback we received: Kyle: Think on the composition. Populate the space with props for intrigue, make it look great first. Refine the camera angles. Billy: (echoing with Kyle) You could either go set dressing and extension with CG or macro with CG props. Todd: Flour and grains are really hard. Develop the pipeline and prepare for a ton of VOPs. Think about a low res base sim to drive auxilliary detail POP sims. Go 8% with grains, then generate particles to render detail. You could even do this with FLIP. We run into this problem with snow, too. FEM is good. Make sure your initial states are solid. Don't discount RBDs.

Second R&D with Houdini Grains POP from heightfield (converted to VDB). Increased clumping and self-friction so it would hold together closer to flour.

First R&D with Houdini Grains POP from sphere using the Dry Sand shelf tool. Rectangle ram is a rigid body with Z-ward force applied, all atop a ground plane.

Our team zeroed in on a miniature kitchen setup, possibly a chase scene, and Deborah pointed us to this spot. After brainstorming and deliberation, we narrowed down to a Twizy car jumping out of a box bursting with packing peanuts (time permittng: fire pipe and chase cars) for shot one. For shot two we would continue along the counter, hit a pile of flour and fall into the sink (time permitting: add trailing behind tires, chase cars stopped by a cleaver). Extra shots may be of the driver (a toy? a gummy bear?) and or an establishing counter shot.











Thanks to Travis Harkleroad for showing me the substep menu option, which I set to 0.25 to calculate subframes, and set the DOP substeps to 4. He also told me to use the animated=1 attribute instead of a deforming collsiion object, and to avoid concave hulls. I re-did many of my collision objects accordingly.

Feedback we received: Todd: It's looking better. There's something off about that car animation and how it interacts with the chocolate chips, seems a bit unnatural. The chips also get really still, you probably want some tumbling off motion afterwards. There's a spot by the lid in that same shot where there's an opportunity for the car to kick up some powder or something that's missing there, following this theme. The chocolate shader is pretty dark, usually it's a warm brown to subtle purples to emphasize the tastiness.

After scaling up the scene by a factor of 27 (and the time scale), the sims take about 10 minutes! I also cranked the constraint iterations in the bullet solver to around 16000 to cover my semi-complex objects. I lowered my substeps from 12 to 5. I also added in relevant set dressing objects.

Feedback we received: Todd: The collision issues may be involved with scale; somewhere around 5cm the bullet solver just doesn't work as expected, so scale up time and space. Constraint Iterations also are very important for collision calculations. Sometimes multiple duplicates of the collision objects can enforce the weight. Also avoid deforming objects, use geometrywrangle and manipulate attributes to stabilize the motion. Kyle: The top down shot looks good for the added shot, I'd like to see the animation for it next week. It's also the best color graded shot, so you should match the others more closely to that shot. A lot of motion blur and depth of field will help. That's better in the render, but if it's too expensive it could be done in compositing.

Lots of collision woes. I'm using an altered base from my flour (plan A) technique. A reduced 3D scan of our kitchen counter location is my ground plane. I also tried using volume based collisions. These sims took around 8 or 9 hours!! Deborah guided me in using the bullet solver with a multisolver so I could manipulate finer attributes with a geometry wrangle node. I use polyreduced geo for the sim, import those back to the SOP level with a dop I/O set to represent objects with points, and copy the heavier geometry to those.

Feedback we received: Todd: So you've got three main things going on; the main hit, slumping, and air material in the flour hit. I'm still not sold on the main hit. Shoot a bunch of reference video of the car hitting flour at different spots. The grain solver doesn't have great volume conservation. Kyle: There hasn't been enough progress on this project. Get the layout in full edits of the project. Billy: Render objects in place. Get in more detail and roughness of objects. You're headed in the right direction.

With a Drift Threshold of 0.14 and Kinetic Scale of 0.7 and particle separation of 0.0008, I achieved a high res sim that held up but was too stiff. Once I added the tentative animation alembic, the car volume was waking the particles but not colliding with them. Pre-alembic I was using a method of turning the geometry active and not active based on the object's keyframe positions, but this setup did not translate to the .abc. For the render tests, adding point Alphas increased render times by 20 to 40 percent. But on the render farm, each HD frame of semi-transparent particles (~8M) took just less than 10 minutes. The .abc collider almost fixed! At least it’s not crashing the program anymore! All I had to do (maybe) was uncheck “Create Active Object” I think it is barely colliding perhaps because of a collision padding value I changed during the troubleshoot process. This sim has about 4.5M particles and took 3.5 hrs.

Feedback we received: Todd: Combining resolutions is a good idea. Consider having the top stuff live with a smaller stiff mound underneath. Upres sooner than later so you can see what you can get away with. Billy: It's 60% done so all that's left is execution. You have a good layout so far. The packing peanut needs more shape detail. Start getting detailed asset textures from references, perhaps with Substance. Achieve subtle details, like a stretched specularity on the mixer. To tackle the reflections issue in the glass bowl, try texturing a counter proxy.

The first clip is a FLIP version of a flour pile with very high viscosity attributes animated with a volume. For troubleshooting the collapse of the pile I used sleeping (the blue particles above) and auto-wake in the POP Solver so that the pile only starts simulating after contact with the car volume. There's still excess collapse a few frames after contact. My focus has been on Internal Collision Stiffness, Kinetic Scale, and Drift Threshold parameters in the Grain Update DOP as well as the pop solver's substeps. I may still need to increase Drift Threshold and decrease Scale Kinetic values. The pile started exploding (I'm not sure why), which I fixed by turning down the Max Neighbor Query from 2.5 to 1.5 in the grain update. To save on calculations and perhaps lift up the pile (per Professor Fowler's suggestion) I hollowed a bit of the source volume and used that hole as a static collision object. For some reason this static object appears to be ignored in the simulation. I began setting up a pyro simulation advecting from particle data and plan on isolating fast-moving particles into a group to advect pyro with a set-up similar to the one below.

Feedback we received: Todd: Don't tackle it all in one big sim. The trick is bridging together multiple simulations. Grains are notorious for not scaling. Think of the airfield you'll need to get the floating particles effect. I wouldn't render out a volume but rather multiple layers of semi-opaque particles. Kyle: The reshoots and glass bowl look great. Start placing objects in the scene. Billy: Focus on the layout and previs with camera tracking to get an idea of which objects you need and where they will be placed in the scene first so you don't waste time shading things you may not use.

Once I increased resolution, the shape started deflating. After testing different parameters, the notable ones seem to be Scale Kinetic and Drift Threshold in the Grain Update. But neither of those values tweaked solved the full issue as much as increasing the solver substeps did. Other solutions I attempted were Solid Grains, a constraint applied based on the source shape. This started to give me an accurate shell but still fell apart at higher resolutions. I also tried auto sleep options which I think can work but needs to be tweaked more.

Feedback we received: Kyle: The 3D Scan is awesome. Exclude the shots with a person because it looks like a student film. I'm not a fan of the party hats and the objects don't look like they belong (they're too arranged). Utilize the round table or the stove to extend the length of some of the shots. Clean up the set a bit. Get something see-through to sell the compositing. Billy: (echoing with Kyle) Reshoot without any objects. Use a macro lens and perspective. Build in the rest of the set and objects in CG. Think about the story development, a cute car could be a part of a cuter story. With this much done in two weeks you could potentially get a lot done.

For these tests I scaled the scene down to be closer to our miniature space size (~10cm) and added impact to the side as in our shot plan. Some of the motion remains weird and wonky which I address below.

Here I refined the motion more mostly by changing the Scale Kinetic Friction to 2 in the POP Grains Behavior tab. I also set the Constraint Iterations to 40, Internal Collision Stiffness to 30k, Clumping Weight to 0.2 and Clumping Stiffness to 1. Changing the Drift Threshold to 0.012 in the POP Grains Solver tab helped stop much of the excessive slumping. With the particle separation at 0.0022, there are 76k points. In the second half of this test, I try out the Point Replicate Node to multiply that amount by 10 and try to add some noise. However it looks clumpy in a bad way, probably until I decrease the Particle Sep further.

Particle Separation here is 0.0016 (200k points) with a point replicate of 15. Going forward I hope to upres it, possibly excluding the still half of the flour pile. I want to test out POP Replicate on isolated points as well as pyro advection. Dale Bunten showed interesting results from FLIP foam using point replicate that looked similar to flour that I can explore further.

The beginning of this masterclass inspired me to use VOP point cloud filters (below) to isolate specific particles to advect pyro from later based on position and neighbors. I could also filter based on velocity or a hit attribute.

Feedback we received: Kyle: Think on the composition. Populate the space with props for intrigue, make it look great first. Refine the camera angles. Billy: (echoing with Kyle) You could either go set dressing and extension with CG or macro with CG props. Todd: Flour and grains are really hard. Develop the pipeline and prepare for a ton of VOPs. Think about a low res base sim to drive auxilliary detail POP sims. Go 8% with grains, then generate particles to render detail. You could even do this with FLIP. We run into this problem with snow, too. FEM is good. Make sure your initial states are solid. Don't discount RBDs.

Second R&D with Houdini Grains POP from heightfield (converted to VDB). Increased clumping and self-friction so it would hold together closer to flour.

First R&D with Houdini Grains POP from sphere using the Dry Sand shelf tool. Rectangle ram is a rigid body with Z-ward force applied, all atop a ground plane.

Clarissa called a few commercial locations and did some excellent scouting, and we settled on my girlfriend's kitchen for its natural light, spacial layout, and ease of access.

Our team zeroed in on a miniature kitchen setup, possibly a chase scene, and Deborah pointed us to this spot. After brainstorming and deliberation, we narrowed down to a Twizy car jumping out of a box bursting with packing peanuts (time permittng: fire pipe and chase cars) for shot one. For shot two we would continue along the counter, hit a pile of flour and fall into the sink (time permitting: add trailing behind tires, chase cars stopped by a cleaver). Extra shots may be of the driver (a toy? a gummy bear?) and or an establishing counter shot.

Thanks to Travis Harkleroad for showing me the substep menu option, which I set to 0.25 to calculate subframes, and set the DOP substeps to 4. He also told me to use the animated=1 attribute instead of a deforming collsiion object, and to avoid concave hulls. I re-did many of my collision objects accordingly.

Feedback we received: Todd: It's looking better. There's something off about that car animation and how it interacts with the chocolate chips, seems a bit unnatural. The chips also get really still, you probably want some tumbling off motion afterwards. There's a spot by the lid in that same shot where there's an opportunity for the car to kick up some powder or something that's missing there, following this theme. The chocolate shader is pretty dark, usually it's a warm brown to subtle purples to emphasize the tastiness.

After scaling up the scene by a factor of 27 (and the time scale), the sims take about 10 minutes! I also cranked the constraint iterations in the bullet solver to around 16000 to cover my semi-complex objects. I lowered my substeps from 12 to 5. I also added in relevant set dressing objects.

Feedback we received: Todd: The collision issues may be involved with scale; somewhere around 5cm the bullet solver just doesn't work as expected, so scale up time and space. Constraint Iterations also are very important for collision calculations. Sometimes multiple duplicates of the collision objects can enforce the weight. Also avoid deforming objects, use geometrywrangle and manipulate attributes to stabilize the motion. Kyle: The top down shot looks good for the added shot, I'd like to see the animation for it next week. It's also the best color graded shot, so you should match the others more closely to that shot. A lot of motion blur and depth of field will help. That's better in the render, but if it's too expensive it could be done in compositing.

Lots of collision woes. I'm using an altered base from my flour (plan A) technique. A reduced 3D scan of our kitchen counter location is my ground plane. I also tried using volume based collisions. These sims took around 8 or 9 hours!! Deborah guided me in using the bullet solver with a multisolver so I could manipulate finer attributes with a geometry wrangle node. I use polyreduced geo for the sim, import those back to the SOP level with a dop I/O set to represent objects with points, and copy the heavier geometry to those.

Feedback we received: Todd: So you've got three main things going on; the main hit, slumping, and air material in the flour hit. I'm still not sold on the main hit. Shoot a bunch of reference video of the car hitting flour at different spots. The grain solver doesn't have great volume conservation. Kyle: There hasn't been enough progress on this project. Get the layout in full edits of the project. Billy: Render objects in place. Get in more detail and roughness of objects. You're headed in the right direction.

With a Drift Threshold of 0.14 and Kinetic Scale of 0.7 and particle separation of 0.0008, I achieved a high res sim that held up but was too stiff. Once I added the tentative animation alembic, the car volume was waking the particles but not colliding with them. Pre-alembic I was using a method of turning the geometry active and not active based on the object's keyframe positions, but this setup did not translate to the .abc. For the render tests, adding point Alphas increased render times by 20 to 40 percent. But on the render farm, each HD frame of semi-transparent particles (~8M) took just less than 10 minutes. The .abc collider almost fixed! At least it’s not crashing the program anymore! All I had to do (maybe) was uncheck “Create Active Object” I think it is barely colliding perhaps because of a collision padding value I changed during the troubleshoot process. This sim has about 4.5M particles and took 3.5 hrs.

Feedback we received: Todd: Combining resolutions is a good idea. Consider having the top stuff live with a smaller stiff mound underneath. Upres sooner than later so you can see what you can get away with. Billy: It's 60% done so all that's left is execution. You have a good layout so far. The packing peanut needs more shape detail. Start getting detailed asset textures from references, perhaps with Substance. Achieve subtle details, like a stretched specularity on the mixer. To tackle the reflections issue in the glass bowl, try texturing a counter proxy.

The first clip is a FLIP version of a flour pile with very high viscosity attributes animated with a volume. For troubleshooting the collapse of the pile I used sleeping (the blue particles above) and auto-wake in the POP Solver so that the pile only starts simulating after contact with the car volume. There's still excess collapse a few frames after contact. My focus has been on Internal Collision Stiffness, Kinetic Scale, and Drift Threshold parameters in the Grain Update DOP as well as the pop solver's substeps. I may still need to increase Drift Threshold and decrease Scale Kinetic values. The pile started exploding (I'm not sure why), which I fixed by turning down the Max Neighbor Query from 2.5 to 1.5 in the grain update. To save on calculations and perhaps lift up the pile (per Professor Fowler's suggestion) I hollowed a bit of the source volume and used that hole as a static collision object. For some reason this static object appears to be ignored in the simulation. I began setting up a pyro simulation advecting from particle data and plan on isolating fast-moving particles into a group to advect pyro with a set-up similar to the one below.

Feedback we received: Todd: Don't tackle it all in one big sim. The trick is bridging together multiple simulations. Grains are notorious for not scaling. Think of the airfield you'll need to get the floating particles effect. I wouldn't render out a volume but rather multiple layers of semi-opaque particles. Kyle: The reshoots and glass bowl look great. Start placing objects in the scene. Billy: Focus on the layout and previs with camera tracking to get an idea of which objects you need and where they will be placed in the scene first so you don't waste time shading things you may not use.

Once I increased resolution, the shape started deflating. After testing different parameters, the notable ones seem to be Scale Kinetic and Drift Threshold in the Grain Update. But neither of those values tweaked solved the full issue as much as increasing the solver substeps did. Other solutions I attempted were Solid Grains, a constraint applied based on the source shape. This started to give me an accurate shell but still fell apart at higher resolutions. I also tried auto sleep options which I think can work but needs to be tweaked more.

Feedback we received: Kyle: The 3D Scan is awesome. Exclude the shots with a person because it looks like a student film. I'm not a fan of the party hats and the objects don't look like they belong (they're too arranged). Utilize the round table or the stove to extend the length of some of the shots. Clean up the set a bit. Get something see-through to sell the compositing. Billy: (echoing with Kyle) Reshoot without any objects. Use a macro lens and perspective. Build in the rest of the set and objects in CG. Think about the story development, a cute car could be a part of a cuter story. With this much done in two weeks you could potentially get a lot done.

Here I refined the motion more mostly by changing the Scale Kinetic Friction to 2 in the POP Grains Behavior tab. I also set the Constraint Iterations to 40, Internal Collision Stiffness to 30k, Clumping Weight to 0.2 and Clumping Stiffness to 1. Changing the Drift Threshold to 0.012 in the POP Grains Solver tab helped stop much of the excessive slumping. With the particle separation at 0.0022, there are 76k points. In the second half of this test, I try out the Point Replicate Node to multiply that amount by 10 and try to add some noise. However it looks clumpy in a bad way, probably until I decrease the Particle Sep further.

Particle Separation here is 0.0016 (200k points) with a point replicate of 15. Going forward I hope to upres it, possibly excluding the still half of the flour pile. I want to test out POP Replicate on isolated points as well as pyro advection. Dale Bunten showed interesting results from FLIP foam using point replicate that looked similar to flour that I can explore further.

Feedback we received: Kyle: Think on the composition. Populate the space with props for intrigue, make it look great first. Refine the camera angles. Billy: (echoing with Kyle) You could either go set dressing and extension with CG or macro with CG props. Todd: Flour and grains are really hard. Develop the pipeline and prepare for a ton of VOPs. Think about a low res base sim to drive auxilliary detail POP sims. Go 8% with grains, then generate particles to render detail. You could even do this with FLIP. We run into this problem with snow, too. FEM is good. Make sure your initial states are solid. Don't discount RBDs.

Second R&D with Houdini Grains POP from heightfield (converted to VDB). Increased clumping and self-friction so it would hold together closer to flour.

First R&D with Houdini Grains POP from sphere using the Dry Sand shelf tool. Rectangle ram is a rigid body with Z-ward force applied, all atop a ground plane.

Clarissa called a few commercial locations and did some excellent scouting, and we settled on my girlfriend's kitchen for its natural light, spacial layout, and ease of access.

Our team zeroed in on a miniature kitchen setup, possibly a chase scene, and Deborah pointed us to this spot. After brainstorming and deliberation, we narrowed down to a Twizy car jumping out of a box bursting with packing peanuts (time permittng: fire pipe and chase cars) for shot one. For shot two we would continue along the counter, hit a pile of flour and fall into the sink (time permitting: add trailing behind tires, chase cars stopped by a cleaver). Extra shots may be of the driver (a toy? a gummy bear?) and or an establishing counter shot.