StreamDiffusionTD 0.2.3🚀🖼️ [ Changelog ]

I have a problem is : TensorRT is not found in the virtual environment

Eagerly waiting for this week's update!

Hey! been trying the streamdiff tool. It's pretty amazing. got an issue though. tensorrt doesn't work after changing the resolution. default 512*512 resolution works fine though. is there any workaround for this?

Hey guys, is there a video about huggingface / other models / lora / checkpoints downloaded from different sites (civitai, ecc) ?? I'm really curious about experimenting, and the base SD model from the component works but I'd love to customize more with other models

Is it possible to choose which GPU it uses if I have more than one

Was wondering if there is going to be an update for compability with new cuda and python versions?

Is this project abandoned? If not, when is the next release?

"Add Python, CUDA, and Git to your system's PATH during installation." What does this mean? I just installed it normally to my computer system. What else do i need to do? I didn't see any questions like that when i was installing all three components... Please help me

Hello :) Where can I install new models?

Loving this component so far! Great work. I'm interested in training a custom LoRA - any advice on how to get started? Also, can you please share any other LoRA known to work well with this TOX?

Hello, I am trying to change the 'SD model id/ path' on the stream setting page to that of a model stored locally on my computer, however it is unable to run giving me an error that it needs to be a URL path to a huggingface model. Any direction of this would greatly be apreciated

When is the new version coming out? What's the newest and highest resolution model that I can run with the current (and upcoming) release, with tensorrt enabled?

it works now but i loocking for a solution the model is taken so many GB when i hit pulse "startstream"

New version? We can't use new models with this one. Will the new version support new models?

Hi Mate! Thank you for version 0.2.3. However, I'm encountering an issue with TensorRT—it seems to be missing in the virtual environment. I need to resolve this. Do you have any tips or tricks to fix it? Please let me know. Thanks! have a nice day!

Thank you, artist. To use the 0.2.3 version, do I not proceed to the Cuda 12.6 version, but to use the latest version, do I need to change it to 12.1 and install it? Or will it proceed with 12.6 version without any problems?

Hello, this is a wonderfull TOX! Quick question: Is it possible to use the new SD 3.5 large turbo model?

Is this actually still a thing. Do you still work on this?

Does anyone know how I can join the Discord Group? I paid for patreon so that I could use Stream Diffusion but it wont even install because of something called TensorRT. I googled it but now I have to know how to use coding to install from github. I dont understand any of it.

I'm trying to use this version but the only way can get it to work is using 144x144 pixels, and tensorrt is kinda weird, when it works it says that it does not have enough memory, when it doesn't the images start to generate (just in 144x144px)

Hi everyone, I'm trying to download streamdiffusion I'm just missing the tesorRT, the problem is that when I try to download it it gives me this error: from huggingface_hub import HfFolder, cached_download, hf_hub_download, model_info ImportError: cannot import name 'cached_download' from 'huggingface_hub' (C:\StreamD\StreamDiffusion\venv\lib\site-packages\huggingface_hub\__init__.py). Has this happened to anyone? Do you know how to fix it? Because without downloading this it doesn't even let me start the prompts. If I try to start one it still tries to download the tesorRT without success.

hey guys! i was using the operator pretty smoothly, but it seems to be stuck when building the tensorrt engine. I know it takes a while but ive waited now way longer than i used to. tried building an engine for 512 x 512 and then 960 x 512 but no luck in either.

Seems like Hugging Face might change API links. There is problem then installing TensorRT on a new location, (ImportError: cannot import name 'cached_download' from 'huggingface_hub' , Did you mean: 'hf_hub_download') also there is new error, happend when streamdiffusion install ( subprocess-exited-with-error × git clone --filter=blob:none --quiet https://github.com/dotsimulate/StreamDiffusion.git ) - when you install from a this tox into new location also.

Hellooo :) How can i add a negative prompt? :)

Hi there! Sorry if this is a silly question, but how can I output an image with the same dimensions as my input? My output size is currently fixed at 512x512 px, but I’d like to output an image at 1920x1080.

Has anyone figured out how to enable the safety checker? Trying to keep it very strict SFW

I'm failing at getting StreamDiffusionTD to behave with either SDXL 1.5 or SDXL-Turbo, with xformers. I managed to patch StreamDiffusion myself for a Linux build of it, and it works over there, but it looks like the LCM-LoRA doesn't get loaded (I also tried giving a custom one): I need a very large step count for the model to produce an image corresponding to the input image, and it doesn't appear to follow the prompt. Swapping back to SD-Turbo makes it all work again. Are there instructions somewhere for how to configure the node to work with SDXL 1.5 or SDXL-Turbo?

did any of you succeed with making a LoRA work? Could you please link it to me and tell me your specs? 🙏🏼 thanks

Does anyone have issues loading any kind of LoRA? I'm running on a Windows 11, RTX 4080 SUPER, python 3.10.9, cuda 12.1 a SD1.5 model with 3 steps and V2V. Every LoRA has a slightly different type of error; this LoRA here [https://civitai.com/models/109043/skin-and-hands-malefemale-from-polyhedron] should be able to support SD1.5, but I still get a long error starting with this: Error loading LoRA 'polyhedron_new_skin_v1point1': Error(s) in loading state_dict for UNet2DConditionModel: size mismatch for down_blocks.0.attentions.0.proj_in.lora_A.polyhedron_new_skin_v1point1.weight: copying a param with shape torch.Size([128, 320, 1, 1]) from checkpoint, the shape in current model is torch.Size([128, 320]). Anyone that could help?

I was running streamdiffusion smoothly but fow some reason it just started to crash whenever I start streaming. It connects to the server without problem but for some reason every time I press stream, it crashes and says it is due to touchdesigner/project1/streamdiffusionSD/numphy_share_out. Is it only happening for me, or do I do something wrong???

Hi Dot i want to ask what is this error, i change pc, try to insall 0.2.3 and this happend " cloning into 'C:/ StreamDiff/StreamDiffusion'... error: RPC failed; curl 92 HTTP/2 stream 5 was not closen cleanly : CANCEL (err 8) error: 6465 bytes of body are still expected fetch-pack: unexpected disconnect while Reading sideband packet fatal: early EOF fatal: fetch-pack:invalid index-pack output "

I'm running streamdiffusion on AWS - is it possible to use that server inside touchdesigner?

So I have some performance related issues. Running v2.3 and latest td on RTX3090 and getting only 8ish fps regular and maybe 5fps in v2v/controlnet (dont know if these are normal fps numbers but I think thats pretty low for rtx3090). When I enable tensorrt I got 15fps (although it looked like far more) so great! But ONLY the first time, now every other time after I start it WITH tensortrt enabled I only get 8ish again. Weird thing was, I didn't even close td, I just stopped the stream and started it again later without changing anything. Since then I obviously did reboots and such to make sure that wasn't it, ran it without any other apps open (although when I got higher fps I had a lot of stuff open and running). Should I remove rensorrt and redo the optimizations or just redo the whole streamdiffusion installation? What is the best approach?

Is it no longer possible to control the Denoise Schedule of Input 1?

I'm getting this error "TensorRT not found in the virtual environment" even though I installed it from the operator

Hi is there any document on how to download hugging face models and use only the newly downloaded models for StreamDiffusionTD?

Hello! What’s the best way to troubleshoot? Start Stream brings up a terminal window for half a second before the window closes and status remains Not Streaming.

Trying to get this working on Mac and keep running into this error: Preparing Stream... Process Process-4: Traceback (most recent call last): File "/Library/Frameworks/Python.framework/Versions/3.11/lib/python3.11/multiprocessing/process.py", line 314, in _bootstrap self.run() File "/Library/Frameworks/Python.framework/Versions/3.11/lib/python3.11/multiprocessing/process.py", line 108, in run self._target(*self._args, **self._kwargs) File "/Users/ethan/SD223/StreamDiffusion/streamdiffusionTD/main_sdtd.py", line 369, in image_generation_process pinned_input_buffer = torch.empty((height, width, 3), dtype=torch.uint8, pin_memory=True) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ RuntimeError: Need to provide pin_memory allocator to use pin memory. Please help

Hello, there is an issue with memory allocation if I try to use any other nvidia stuff, like bg removal, stream diffusion will lock out the node, mediapipe gpu will act the same if I try segmentation. I think this might be an issue for the streamdiffusion git repo. I got a 4070ti and the gpu memory is around 6.3/12gb with the tox running an sdxl turbo with control net at 512x512px. cannot create Cuda stream - an illegal memory access encountered.

Hello, everything is working perfectly, thank you very much. However, I have a question: how can I increase the amount of VRAM I am using? I have a 4060 with 16GB, but I am currently only utilizing 4GB to run the Tox

I Love this update, just hit 56fps on 512x512 SDTurbo with TensorRT Thank you! so Incredible! Please support FLUX.1-schnell soon, this would really take my projects to the next level. and it seems the 4090 has enough headroom for it! Error : KeyError: 'time_embed.0.weight' Failed to load C:\Projects\StreamDiffusionTD-0.2.3\models\Model\FLUX.1-schnell\flux1-schnell.safetensors / SD1.5 + LCM LoRA model from both local and Hugging Face due to: 'time_embed.0.weight'

how can I get better fps?, sometimes get 5. fps but after and mostly get 0. etc. fps thanks man

Banger!! Thanks a lot !!!!

Amazing. Thank you 🙏

Amazing. Thanks for adding the FPS counter:) cool stuff

Nice! Thank you! :D

is this Mac compatible?

Woo hoo! An update that arrives when I actually have time to test it. Thank you for all your hard work!