I'm running hard up against the gRPC broadcast receiver lagging behind until it errors out. Wondering if I'm holding it wrong or something of that nature

2025-07-31T08:16:52.893796772Z ERROR re_grpc_server: Error reading message from broadcast channel: channel lagged by 281974

Minimal repro: https://gist.github.com/hansihe/522103ad9c5d4e55e9bcfd99213f5b80 I can push a full crate to github too if that makes things easier to dig into, just let me know. I am attempting to use

re_renderer

directly within a custom

egui

application (using

wgpu

). With nothing queued in

ViewBuilder

I get nothing and see the background as expected. When I

queue_draw

a

GenericSkyboxDrawData

, I get a fully black viewport. Behavior of other renderers varies, but I never get any meaningful output. I captured a frame in RenderDoc to try to get more insight into what might be happening. The black screen seems to stem from the color value in the shader `nan`ing. Tracing it back in the debugger it seems to originate from the

re_renderer

frame uniform buffer

which is bound to descriptor set 0 being filled with nulls only. According to RenderDoc the buffer is only `clear`ed and barriered before it is used in the shaders for the

re_renderer

renderers. I imagine I am probably missing some important step in my code, does someone know what I might be doing wrong?

I am receiving a h264 video stream from a UDP socket and I would like to log it with the new VideoStream in C++. More specifically, I am receiving UDP packets (of length 1316 or a smaller multiple of 188) and send them to rerun with:

rec.log("video_stream",rerun::archetypes::VideoStream{}.with_sample(rerun::components::VideoSample{data}));

I had a first error similar to the one mentioned here: https://discord.com/channels/1062300748202921994/1397335975364919392 I applied the

uint32_t

fix and the next error was the ffmpeg version. Got the recommended version and the video stream doesn't get displayed:

Failed to decode video: FFmpeg error: [h264] [error] error while decoding MB 2 64, bytestream 16292

It gets the Dimensions of the video stream right (1920x1080) and the sent data chunks appear in the Video samples table but no video gets displayed. The video stream is created from that command:

ffmpeg -f rawvideo -pixel_format rgba -video_size {width}x{height} -framerate {frameRate} -i - -vf vflip -c:v libx264 -preset veryfast -tune zerolatency -f mpegts udp://localhost:1234?pkt_size=1316

and it is displayed correctly with

ffplay

.

Hello, I want to log data from two different sources (a python script and a c++ executable) in the same rerun recording. I would like the data to be saved to file as I log it with rerun (i.e. not all al the end) and also to be able to visualize it live in a single viewer. Is it possible with rerun? How would I implement this?

Hello, I have a list of joints from a robot encoder and I want to log them with rerun. Is it possible to log all of them in one call without looping on the list and call rr.log for each value?

I'm trying to render a surface patch generated from HiRISE DTM and orthoimages (e.g. https://www.uahirise.org/dtm/dtm.php?ID=PSP_002387_1985). I used the HiRISE DTM Importer blender plugin (https://hemelmechanica.nl/hirise-docs/quickstart.html) to generate the mesh from the DTM image and texture it use a matching grayscale orthoimage. Everything looks correct in Blender and also in https://gltf.report/ and https://gltf-viewer.donmccurdy.com/. However, in rerun, the model looks bright red as if its texture has the greyscale image in the R channel but G and B are just set to 0. Changing AlbedoFactor to rgb(255,0,0) has no effect on how the surface patch looks. But changing it to rgb(0,255,0) or rgb(0,0,255) or rgb(0,255,255) just makes the entire surface patch pitch black. I'm not sure what I'm doing wrong since when I import other GLB files in rerun (e.g. a glider model I found and a free rubber ducky model I found to use as a test), they look normal. https://cdn.discordapp.com/attachments/1398520508022132866/1398520508907126846/Screenshot_2025-07-25_at_9.01.37_PM.png?ex=6885a938&is=688457b8&hm=2de6e6ff4fe3488eae2d1a41a87b80ce707f6683f76b8bb3dcbe70be27ebf2df& https://cdn.discordapp.com/attachments/1398520508022132866/1398520509552918639/Screenshot_2025-07-25_at_9.02.07_PM.png?ex=6885a938&is=688457b8&hm=ce2b4e462c1542db2b5c763923f96d77cffd7df42356d0a8e433063d318009b0& https://cdn.discordapp.com/attachments/1398520508022132866/1398520509964095608/Screenshot_2025-07-25_at_9.03.30_PM.png?ex=6885a938&is=688457b8&hm=9c3609eac08dcf3c2f3a5568d8f32793ff84ccc0ec46fde8cbf400ef2b67d991& https://cdn.discordapp.com/attachments/1398520508022132866/1398520510509350963/Screenshot_2025-07-25_at_9.15.54_PM.png?ex=6885a938&is=688457b8&hm=7c68c1b8c144dbef323ffce38a4a1c8a000a248cdeb7f24035c439baaea3187a&

Is there a way I can programmatically set the viewpoint for a Spatial3DView using the PythonAPI so that I don't have to manually pan/zoom/rotate a pointcloud with the mouse everytime to see the view I need?

Hi, I'm trying to get the VideoStream support working on C++. The example on python works correctly, but when trying to port this over to C++ I keep getting:

re_viewer_context::component_ui_registry] Failed to deserialize component of type rerun.components.VideoCodec: Failed to deserialize "rerun.components.VideoCodec"

I've put the RUST_LOG up to debug on both the sender and the receiver but it gives me no other information other than the deserialization error. Any idea what I might be doing wrong? I've attached the code I'm working with. This also dumps the h264 data to a file which does reproduce correctly when using ffplay so I don't think this is an encoding error https://cdn.discordapp.com/attachments/1397335975364919392/1397335975645941882/on-the-fly.cpp?ex=688b3d49&is=6889ebc9&hm=f06d6a86384a371ebdcb0a5c7b02a7556716b8bace7b699e20bee65ea0ef7161&

I tested this by creating an application in C++ using the API: - Reads a blueprint (.rbl file) - Creates a recording stream - Attempts to spawn the viewer - Logs some data. When I compile and run this application, it normally spawns the viewer and functions as expected. But when I run this same application as a root user (using sudo), it gets stuck at reading the blueprint stage itself, and cannot proceed forward, unable to spawn the Viewer. If in my application, a root access is required to be used for other purposes, are there any environment variables I need to pass so that the GUI (Rerun viewer) is able to spawn correctly? Supporting evidence is attached. https://cdn.discordapp.com/attachments/1397186188128817222/1397188114430038036/image.png?ex=6880d054&is=687f7ed4&hm=504cba18bd88a0bb8610bcda35b82d9d1ab99b0acc0ef1e55a0d6ef779b8ee93& https://cdn.discordapp.com/attachments/1397186188128817222/1397188114937544834/image.png?ex=6880d055&is=687f7ed5&hm=9142e882a49cf4ddbe87f89c795155c5dad7fb6053aad5e9ba9c0af738984b6b&

When we make the pointcloud panel vertically small (wide and short panel) the points at either end of the viewer become abnormally big. This goes away when the viewer's aspect ratio is more 1:1 (square). Please let me know if this is a bug or configuration issue? I'm using python rerun-sdk 0.24.0 You can reproduce this even on the rerun nuscenes example: https://rerun.io/viewer?url=https://app.rerun.io/version/0.24.0/examples/nuscenes_dataset.rrd All you'll have to do is vertically squish the lidar panel and zoom in so that there are points in the left and right edges of the lidar viewer. https://cdn.discordapp.com/attachments/1397008726090580112/1397008727021850736/image.png?ex=68802943&is=687ed7c3&hm=15c74c5fb58d00906ea87ae9690e4803b87552e1e5195678cdc72bd501ba5266& https://cdn.discordapp.com/attachments/1397008726090580112/1397008728321949756/image.png?ex=68802943&is=687ed7c3&hm=ab2e9ce469b04d766f5e39418c86389a13667c7cc527f5bfa747cf56a3302fd5&

I'm trying to write my own

Renderer

using the

re_renderer

crate and the shader I want to use requires some `texture2d`s. From looking through the Rerun code I can see examples like:

ctx.gpu_resources.bind_groups.alloc(
                &ctx.device,
                &ctx.gpu_resources,
                &BindGroupDesc {
                    label,
                    entries: smallvec![
                        BindGroupEntry::DefaultTextureView(position_data_texture.handle),
                        BindGroupEntry::DefaultTextureView(color_texture.handle),
                        BindGroupEntry::DefaultTextureView(picking_instance_id_texture.handle),
                        draw_data_uniform_buffer_binding,
                    ],
                    layout: point_renderer.bind_group_layout_all_points,
                },
            )

however,

BindGroupEntry

is private. For the variant

BindGroupEntry::Buffer

there is

create_and_fill_uniform_buffer

, but I can't find anything that lets me pass my own texture handles to shaders. Am I missing something?

After upgrading my rerun dependency from 0.23 -> 0.24, my project does not compile anymore because

ashpd

fails to compile. Here's a part of the errors from

cargo build

. Does anybody know how to solve this? I use rust 1.88 on Linux.

error[E0432]: unresolved import `futures_util::AsyncReadExt`
   --> /home/xyz/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/ashpd-0.10.2/src/desktop/secret.rs:26:5
    |
26  | use futures_util::AsyncReadExt;
    |     ^^^^^^^^^^^^^^^^^^^^^^^^^^ no `AsyncReadExt` in the root
    |
note: found an item that was configured out
   --> /home/xyz/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/futures-util-0.3.31/src/lib.rs:320:47
    |
320 |     AsyncBufRead, AsyncBufReadExt, AsyncRead, AsyncReadExt, AsyncSeek, AsyncSeekExt, AsyncWrite,
    |                                               ^^^^^^^^^^^^
note: the item is gated behind the `io` feature
   --> /home/xyz/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/futures-util-0.3.31/src/lib.rs:316:7
    |
316 | #[cfg(feature = "io")]
    |       ^^^^^^^^^^^^^^

error[E0432]: unresolved import `futures_util::AsyncReadExt`
   --> /home/xyz/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/ashpd-0.10.2/src/helpers.rs:4:5
    |
4   | use futures_util::AsyncReadExt;
    |     ^^^^^^^^^^^^^^^^^^^^^^^^^^ no `AsyncReadExt` in the root
    |
note: found an item that was configured out
   --> /home/xyz/.cargo/registry/src/index.crates.io-1949cf8c6b5b557f/futures-util-0.3.31/src/lib.rs:320:47
    |
320 |     AsyncBufRead, AsyncBufReadExt, AsyncRead, AsyncReadExt, AsyncSeek, AsyncSeekExt, AsyncWrite,
    |                                               ^^^^^^^^^^^^
...

NOTE: I saw [this](https://users.rust-lang.org/t/issue-with-futures-lite-dependency-resolution-in-ashpd-read-to-string-read-to-end-errors/125669/1) already, but it didn't apply to me

I have some questions about Revy and Rerun. I use Revy, which I want to make some changes to it but I'm new to rerun, so I want to ask some questions about the rerun side. Maybe also for people who know rerun, like for a long time, the questions sound a little bit dumb, so sorry for that in advance. My first question is, I want to leverage the reflection system and the apptyperegistry like Bevy Inspector Egui does and want to create a general-purpose like rerun Rust type component that presents like an enum or struct, which is a vec of fields. Is it possible, like in bevy, that I can register the same type but with a different component ID? In rerun, I think it is called component name or component descriptor, I'm not sure right now. I saw that rerun 24 came out, which introduces tag components. Would that be better for structuring the data? Because now we could have like pure data type components (bool, ints and floats) and have the Rust type or Bevy components be an archetype with the corresponding name, and use the data type components to be a part of the archetype. So, an archetype would represent the Bevy component. Would that be better for querying, structuring and projecting the data? Why I am suggesting the idea is that how the structure of revy is implemented now, you would have to write logger for every single type you have or it just creates a text component with the Ron serialized data in it, which I think is not good for projecting data and querying, but I can also be corrected on that. And creating a logger function for every type you have is a little bit tedious because I create a tool, only a tool, and I have already like 200 to 300 types. If you create a real game with Bevy, you would like maybe have more than hundreds of types, 1000 to 2,000 types, because every single thing is a data type in Bevy. So, writing a logger function for every single type is not a small barrier to entering Revy for logging the data.

Does the log function block until the data is stored to file? I'm writing a control loop and don't want to block the loop if the log function is taking a long time to run. If it doesn't do this already, I can use a crossbeam channel to send messages from my main loop to the logger thread.

For context, I'm running rerun inside of an egui app, similar to here: https://github.com/rerun-io/rerun/tree/e572612a9fa13c0b3168610f119bc8b7a4034f83/examples/rust/extend_viewer_ui. This gives me a little more access to rerun internals. I'm trying to add some quick buttons to my custom sidebar to essentially toggle the visibility of items in a 3D space on/off. From what I can tell, I'll need to log to the blueprint to the correct path to be able to do this. There seems to be a handful of API options to do the logging, - in `SystemCommand`: https://docs.rs/rerun/latest/rerun/external/re_viewer_context/enum.SystemCommand.html#variant.UpdateBlueprint - using a custom

RecordingStream

set to log to the blueprint directly - using

RecordingStream::send_blueprint

Unfortunately, all of these require the

StoreId

of the active blueprint and I can't seem to find a way to get it. The closest seems to be

StoreHub::active_blueprint_id

, but I don't see anyway to access that. Any suggestions here for how I can get that? Or another route I could take? Thanks!

Hey, I'm exploring using rerun to visualize geometry algorithms (and help debug edge cases), I have a rough implementation prototype for parry at https://github.com/Vrixyz/parry/pull/6. Is there a way to hide some values, or specify that a specific log should be displayed only on that frame ? Thx! I'm using rerun's

.set_time_seconds

and

.log

for attached demo. - I'd like to show input points only at first frame, then only facets, then only valid facets (but maybe another need would be to persist some data multiple frames) https://cdn.discordapp.com/attachments/1393153897325002814/1393153897517809706/parry_step_debugging.mp4?ex=6872232c&is=6870d1ac&hm=aa1692e9fa99b41628eef4de6cc50ed6cc2852c3d76467211b9f0a3643038e48&

I am trying to live plot the pose estimation of the camera. I have two issues. 1. I am not able to follow the camera ( track ) 2. If I try to reset the origin of the 3D window, it won’t let me with mouse. Only way I can see the out of view pose path is by zooming out. Any suggestions on how I can fix this ? https://cdn.discordapp.com/attachments/1392843579793215549/1392843580175028235/IMG_9541.jpg?ex=6871022b&is=686fb0ab&hm=2b673d2e401cd34ee443377e5a1c90e512f0da79808801d252a77d4e2dfe295a&

Hello, With the ability to log custom component, as mention here: https://rerun.io/docs/concepts/entity-component#adding-custom-data, it would be great to be able to change the color element use by the viewer directly inside the viewer. Basically, use any component of the entity type that make sense (Float, Int, Array, ...) Currently I didn't find a way to do this, I needed to precompute my color Component in my code. In my currently use case, I'm mostly interested in the

Point3D

archetype

Hi team, I am new to rerun and trying to make a live visualizer tool for robot teleoperation recording - with 2 cameras, tactile sensors, robot states, point cloud, etc. I subscribe to all the ROS2 topics and

rr.log

according to the respective type. It works fine on the direct host machine with the native GUI but has a lag when I use the web_viewer mode with

grpc

. My team prefers to run everything from inside dockers (combined with another ROS2 functionality) and I am having a hard time figuring it out. I don't want to use web_viewer since that brings in ~3 seconds lag on the visualization. Is there any way to run the native viewer through the docker?

Hi, I would like to log the motion of a parallel mechanism, similar to the example shown [at this link](https://github.com/Rhoban/placo-examples/blob/master/kinematics/planar_2dof_trajectory.py). My current robot is a Stewart platform, but this example illustrates the concepts well. While it is possible to rotate the two motors, the connection is not maintained. Additionally, directly moving the end effector is not feasible. Do you have any ideas on how to accomplish this? https://cdn.discordapp.com/attachments/1389638923768172694/1389638924951097546/image.png?ex=68660259&is=6864b0d9&hm=b6494c11dd36a0c7dfd26e264dce7718d01a99d2ede0986a011dfcfb27fd4cf4&

Hello everybody, first post here ! I am pretty new to Rerun, we are trying to use it in our company to do live visualization of time series signals for autonomous vehicles. The vehicle is publishing data every 100ms, and using the C++ SDK we collect the data back and stream it into a rerun viewer. The data collected consists of more than 600 signals that are independent from each other. Currently, we are logging the data by doing separate log calls on Scalars. From what I found on different discussions, this might not be the best idea in term of performance. Indeed, we have a huge latency building up into the viewer when we try to log all the signals. However, doing the same with a small number of variables (around 20) works well. I guess using individual Scalars call on multiple really doesn't scale well 😅 We are running it on x86_64 with Ubuntu 24.04. I have been searching around in the documentation or in Discord for a solution to that. From what I found, there is the idea of using the column API to batch the data. However, what the documentation or examples show is sending a range of value of the same signal as a column (https://rerun.io/docs/reference/types/archetypes/scalar#multiple-scalars-in-a-single-sendcolumns-call). Did I miss something otherwise in the explanation of the column API ? In my case, I would expect to send all the signals as a single column to update all the entities at once, so that would be a single signal for each entity. Any explanation on the mater would be greatly appreciated ! 🥹

Is there any documentation on webviewer callback? I see that they are available here, but just see a single example. Is there a list of callbacks with function call signature, etc anywhere? Is the callback structure stable or should we treat it as a private API that changes without warning? https://rerun.io/docs/howto/integrations/embed-web#callbacks

Callbacks 
The Viewer API also allows registering callbacks for certain events.

For example, here is how you would react to entities being selected in the Viewer:


viewer.on("selection_change", (event) => {
  for (const item of event.items) {
    if (item.type === "entity") {
      console.log(item.entity_path);
    }
  }
});

Previous to rerun version 0.23, I did run the rerun webviewer (served over the cli) and forwarded the ws port with nginx. That way I could access the webviewer with an open websocket connection e.g. via: http://172.31.11.40/mycluster/rerun/?url=ws%3A%2F%2F172.31.11.40%2Fmycluster%2Fws

location $BASE_PATH/rerun/ {
        proxy_pass http://localhost:9090/;
        proxy_http_version 1.1;
        proxy_set_header Upgrade $http_upgrade;
        proxy_set_header Connection "Upgrade";
        proxy_set_header Host $host;
        proxy_set_header X-Real-IP $remote_addr;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
        proxy_set_header X-Forwarded-Proto $scheme;
    }

    location = $BASE_PATH/ws {
        return 301 $scheme://$http_host$BASE_PATH/ws/;
    }

    location $BASE_PATH/ws/ {
        proxy_pass http://localhost:9877/;
        proxy_http_version 1.1;
        proxy_set_header Upgrade $http_upgrade;
        proxy_set_header Connection "Upgrade";
        proxy_set_header Host $host;
        proxy_set_header X-Real-IP $remote_addr;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
        proxy_set_header X-Forwarded-Proto $scheme;
    }

As of 0.23, the connection switched to grpc and I can't make it work the same way. How can I get rerun running inside the pod? My goal is to connect a vscode webviewer which can execute python code which logs to a rerun web viewer instance. Using rrd files would work but does not push data (I don't want to refresh the view).

Hi rerun team, I have a question regarding the use of the columnar api when the data have some gaps. So, in my case, I have several archetypes that I need to log (boxes, 3dpoints line series...) but all of them might have gaps in the recording. Now, with the row api, I'd just call clear on that entity when it is not available. Unfortunately, due to performance issues, I need to use the columnar api. How can I clear the content in this case? I came up with 2 solutions: - use the columnar api and log NaN values on the components of the archetype, which basically makes it disappear until a valid value is logged again. - log all valid entries with the columnar api, and use the row api to clear content on that entity using clear/cleared when needed both are not great, the first one seems like a hack, and the second one is a bit cumbersome to implement if there are many gaps. Can you think of some other way to achieve this? what would you suggest? Thanks!

Here is my current scenario: I have a CLI with an option to enable and disable Rerun logging. I also want to disable Rerun logging if I am unable to connect to a valid viewer. In order to do this I am relying on calls to

RecordingStream

to "early out" when disabled. However, I'm finding that

connect_tcp

doesn't seem to require a connection to a valid viewer for

connect_tcp(url).is_ok()

to return

ErrorCode::Ok

. For instance, if I don't open a Rerun viewer and run my application, I get the following:

CLI OPTION SET: Rerun viewer IP:PORT = 127.0.0.1:9876
[2025-06-18T18:15:12Z WARN  re_sdk_comms::buffered_client] Failed to send message after 3 attempts: Failed to connect to Rerun server at 127.0.0.1:9876: Connection refused (os error 111)
[2025-06-18T18:15:15Z WARN  re_sdk_comms::buffered_client] Dropping messages because tcp client has timed out.
[2025-06-18T18:15:15Z WARN  re_sdk_comms::buffered_client] Dropping messages because tcp client has timed out.
[2025-06-18T18:15:15Z WARN  re_sdk_comms::tcp_client] Tried to flush while TCP stream was still Pending. Data was possibly dropped.
Connected to 127.0.0.1:9876
Rerun logging enabled.

I have a largish-number og rrd files and would like to explore some of them based on some data I have available in a dataframe. As such, it would be convienent to have a function in the SDK or an option from the cli to not launch a new instance of the viewer, but ask the existing view to open a specific file

Is there a way to easily use the browser to view rerun streams? To save wifi bandwidth I want to run the viewer locally on my robot and ideally use a web page to see the data rendered on the robot when I need to. Otherwise, is there an alternative way? basically I would like to be able to opt in when to stream the data out of the robot, but keep recording it in the background. Thank you

Hello, I am observing a lot of latency often sporadically that gradually builds up. I already use a memory limit when spawning the viewer. But this does not help with latency building since this is during the logging time. I also noticed

--drop-at-latency

flag: https://rerun.io/docs/howto/visualization/limit-ram#dropatlatency in the documentation, but what is the way to incorporate this using C++ similar to memory limit spawn options?

When I include rr.save('adslfkj.rrd') I no longer see any data streamed live to the viewer. I can later open up the .rrd file but that's a bit annoying. Can I view the live data and write to file at the same time?

We are using the C++ sdk with c++20 std. Used valgrind massif to log the heap allocations. With rerun logging on I am seeing unbounded growth and with it off I am seeing a flat profile which is expected. We have rerun actively logging to the viewer over the network and are seeing data showing up. We are filling up 8 gigs of ram in less than 30 seconds when we are not profiling. I am using version 0.22.1 include(FetchContent) FetchContent_Declare( rerun_sdk URL https://github.com/rerun-io/rerun/releases/download/0.22.1/rerun_cpp_sdk.zip ) FetchContent_MakeAvailable(rerun_sdk) I am not sure if this issue is relevant but it has no reply https://github.com/mozilla/uniffi-rs/issues/2470 Any thoughts or suggestions are appreciated! https://cdn.discordapp.com/attachments/1380251130340315146/1380251131027914793/L824KW8tjiEcBAAAAAAAAAAAAAAAAAAAAAAAAAPgS2BYqflFizSdhKtDc9ZjHcd2T7LvrPRKPAgAAAAAAAAAAAAAAAAAAAAAAAACArU46UfM8Oc2zlcecJj4t998WkW4757ZpscSjAAAAAAAAAAAAAAAAAAAAAAAAAAAAp3TWsPYsx28KSIlHAQAAAAAAAAAAAAAAAAAAAAAAAAAAzqlxfGpmj81mc2jkQWyfgAAAABJRU5ErkJggg.png?ex=6853ad49&is=68525bc9&hm=cd889c185502cfa362d0d20906aced17b9269394de874d64d60bc6180a92a1b9& https://cdn.discordapp.com/attachments/1380251130340315146/1380251131816579142/L824KW8tjiEcBAAAAAAAAAAAAAAAAAAAAAAAAAPgS2BYqflFizSdhKtDc9ZjHcd2T7LvrPRKPAgAAAAAAAAAAAAAAAAAAAAAAAACArU46UfM8Oc2zlcecJj4t998WkW4757ZpscSjAAAAAAAAAAAAAAAAAAAAAAAAAAAAp3TWsPYsx28KSIlHAQAAAAAAAAAAAAAAAAAAAAAAAAAAzqlxfGpmj81mc2jkQWyfgAAAABJRU5ErkJggg.png?ex=6853ad49&is=68525bc9&hm=e16ec250f510150646db17784a3c1097574762af6a2e6255aae914fca4b51243& https://cdn.discordapp.com/attachments/1380251130340315146/1380251132651114627/5K2BmM6gAAAABJRU5ErkJggg.png?ex=6853ad49&is=68525bc9&hm=b83de04f18164540e1672bb3b2476cf94f5df705e4f87146c18875c0af59741a&