StableDiffusion.NET Once I draw a 1024*1024 image, the program crashes

The limit is 1024 * 768 My device is i9-13900KF, 4090 Is there any way to solve it?

Jan 10 '25 07:01 MuZzzlen

What exactly do you mean with crashing? Is there any error message? does the same model with the same settings work in the stable-diffusion.cpp example application?

Jan 12 '25 22:01 DarthAffe

When I input 1024*1024， it just disappeared without any warning or error.I don't know how to use C++，I am a novice in coding，sry

Jan 13 '25 01:01 MuZzzlen

I have the exact same issue, very simple steps:

Start up the example in this repo
Use Cuda12.win
Load for example https://huggingface.co/second-state/stable-diffusion-v1-5-GGUF
Load model, prompt anything (without changing resolution)
Click Create image-> Crash If I change the resolution for example 512x512, no crash.

The program '[30440] ImageCreationUI.exe' has exited with code 3221226505 (0xc0000409). This is the log from my app that has the same crash: STABLE DIFFUSION [Debug]: stable-diffusion.cpp:163 - Using CUDA backend

STABLE DIFFUSION [Debug]: stable-diffusion.cpp:163 - Using CUDA backend

STABLE DIFFUSION [Info]: stable-diffusion.cpp:195 - loading model from 'I:\LvsSdModels\stable-diffusion-v1-5-pruned-emaonly-Q4_0.gguf'

STABLE DIFFUSION [Info]: model.cpp:885 - load I:\LvsSdModels\stable-diffusion-v1-5-pruned-emaonly-Q4_0.gguf using gguf format

STABLE DIFFUSION [Debug]: model.cpp:902 - init from 'I:\LvsSdModels\stable-diffusion-v1-5-pruned-emaonly-Q4_0.gguf'

STABLE DIFFUSION [Info]: stable-diffusion.cpp:242 - Version: SD 1.x

STABLE DIFFUSION [Info]: stable-diffusion.cpp:275 - Weight type: q4_0

STABLE DIFFUSION [Info]: stable-diffusion.cpp:276 - Conditioner weight type: q4_0

STABLE DIFFUSION [Info]: stable-diffusion.cpp:277 - Diffusion model weight type: q4_0

STABLE DIFFUSION [Info]: stable-diffusion.cpp:278 - VAE weight type: q4_0

STABLE DIFFUSION [Debug]: stable-diffusion.cpp:280 - ggml tensor size = 400 bytes

STABLE DIFFUSION [Debug]: clip.hpp:171 - vocab size: 49408

STABLE DIFFUSION [Debug]: clip.hpp:182 - trigger word img already in vocab

STABLE DIFFUSION [Debug]: ggml_extend.hpp:1111 - clip params backend buffer size = 66.62 MB(VRAM) (196 tensors)

STABLE DIFFUSION [Debug]: ggml_extend.hpp:1111 - unet params backend buffer size = 1270.08 MB(VRAM) (686 tensors)

STABLE DIFFUSION [Debug]: ggml_extend.hpp:1111 - vae params backend buffer size = 159.68 MB(VRAM) (248 tensors)

STABLE DIFFUSION [Debug]: stable-diffusion.cpp:417 - loading weights

STABLE DIFFUSION [Debug]: model.cpp:1698 - loading tensors from I:\LvsSdModels\stable-diffusion-v1-5-pruned-emaonly-Q4_0.gguf

STABLE DIFFUSION [Info]: stable-diffusion.cpp:516 - total params memory size = 1496.38MB (VRAM 1496.38MB, RAM 0.00MB): clip 66.62MB(VRAM), unet 1270.08MB(VRAM), vae 159.68MB(VRAM), controlnet 0.00MB(VRAM), pmid 0.00MB(VRAM)

STABLE DIFFUSION [Info]: stable-diffusion.cpp:520 - loading model from 'I:\LvsSdModels\stable-diffusion-v1-5-pruned-emaonly-Q4_0.gguf' completed, taking 1.58s

STABLE DIFFUSION [Info]: stable-diffusion.cpp:550 - running in eps-prediction mode

STABLE DIFFUSION [Debug]: stable-diffusion.cpp:594 - finished loaded file

STABLE DIFFUSION [Debug]: stable-diffusion.cpp:1535 - txt2img 1024x1024

STABLE DIFFUSION [Debug]: stable-diffusion.cpp:1230 - prompt after extract and remove lora: "Anime girl jumping"

STABLE DIFFUSION [Info]: stable-diffusion.cpp:682 - Attempting to apply 0 LoRAs

STABLE DIFFUSION [Info]: stable-diffusion.cpp:1235 - apply_loras completed, taking 0.00s

STABLE DIFFUSION [Debug]: conditioner.hpp:329 - parse 'Anime girl jumping' to [['Anime girl jumping', 1], ]

STABLE DIFFUSION [Debug]: clip.hpp:311 - token length: 77

STABLE DIFFUSION [Debug]: ggml_extend.hpp:1062 - clip compute buffer size: 1.40 MB(VRAM)

STABLE DIFFUSION [Debug]: conditioner.hpp:457 - computing condition graph completed, taking 96 ms

STABLE DIFFUSION [Debug]: conditioner.hpp:329 - parse '' to [['', 1], ]

STABLE DIFFUSION [Debug]: clip.hpp:311 - token length: 77

STABLE DIFFUSION [Debug]: ggml_extend.hpp:1062 - clip compute buffer size: 1.40 MB(VRAM)

STABLE DIFFUSION [Debug]: conditioner.hpp:457 - computing condition graph completed, taking 12 ms

STABLE DIFFUSION [Info]: stable-diffusion.cpp:1368 - get_learned_condition completed, taking 162 ms

STABLE DIFFUSION [Info]: stable-diffusion.cpp:1391 - sampling using Euler A method

STABLE DIFFUSION [Info]: stable-diffusion.cpp:1428 - generating image: 1/1 - seed 10990

STABLE DIFFUSION [Debug]: stable-diffusion.cpp:798 - Sample

STABLE DIFFUSION [Debug]: ggml_extend.hpp:1062 - unet compute buffer size: 8360.93 MB(VRAM)

STABLE DIFFUSION [Debug]: ggml_extend.hpp:1062 - unet compute buffer size: 8360.93 MB(VRAM) The program '[23928] LyricVideoStudio.win.exe' has exited with code 3221226505 (0xc0000409).

Edit: I got bit better error message when I used the pverload with parameters (_diffusionModel!.TextToImage(prompt, newTp.Parameters)):

338|2025-02-06T19:02:47.5174260+00:00|ERROR|25|MainPage|AppDomain.CurrentDomain.FirstChanceException, stack: at StableDiffusion.NET.Native.g____PInvoke|10_0(sd_ctx_t* __sd_ctx_native, Byte* __prompt_native, Byte* __negative_prompt_native, Int32 __clip_skip_native, Single __cfg_scale_native, Single __guidance_native, Int32 __width_native, Int32 __height_native, Sampler __sample_method_native, Int32 __sample_steps_native, Int64 __seed_native, Int32 __batch_count_native, sd_image_t* __control_cond_native, Single __control_strength_native, Single __style_strength_native, SByte __normalize_input_native, Byte* __input_id_images_path_native, Int32* __skip_layers_native, Int32 __skip_layers_count_native, Single __slg_scale_native, Single __skip_layer_start_native, Single __skip_layer_end_native) at StableDiffusion.NET.Native.txt2img(sd_ctx_t* sd_ctx, String prompt, String negative_prompt, Int32 clip_skip, Single cfg_scale, Single guidance, Int32 width, Int32 height, Sampler sample_method, Int32 sample_steps, Int64 seed, Int32 batch_count, sd_image_t* control_cond, Single control_strength, Single style_strength, Boolean normalize_input, String input_id_images_path, Int32[] skip_layers, Int32 skip_layers_count, Single slg_scale, Single skip_layer_start, Single skip_layer_end) --> System.Runtime.InteropServices.SEHException (0x80004005): External component has thrown an exception. at StableDiffusion.NET.Native.g____PInvoke|10_0(sd_ctx_t* __sd_ctx_native, Byte* __prompt_native, Byte* __negative_prompt_native, Int32 __clip_skip_native, Single __cfg_scale_native, Single __guidance_native, Int32 __width_native, Int32 __height_native, Sampler __sample_method_native, Int32 __sample_steps_native, Int64 __seed_native, Int32 __batch_count_native, sd_image_t* __control_cond_native, Single __control_strength_native, Single __style_strength_native, SByte __normalize_input_native, Byte* __input_id_images_path_native, Int32* __skip_layers_native, Int32 __skip_layers_count_native, Single __slg_scale_native, Single __skip_layer_start_native, Single __skip_layer_end_native) at StableDiffusion.NET.Native.txt2img(sd_ctx_t* sd_ctx, String prompt, String negative_prompt, Int32 clip_skip, Single cfg_scale, Single guidance, Int32 width, Int32 height, Sampler sample_method, Int32 sample_steps, Int64 seed, Int32 batch_count, sd_image_t* control_cond, Single control_strength, Single style_strength, Boolean normalize_input, String input_id_images_path, Int32[] skip_layers, Int32 skip_layers_count, Single slg_scale, Single skip_layer_start, Single skip_layer_end) Exception thrown: 'System.Runtime.InteropServices.SEHException' in StableDiffusion.NET.dll An exception of type 'System.Runtime.InteropServices.SEHException' occurred in StableDiffusion.NET.dll but was not handled in user code External component has thrown an exception.

Edit x2: 1920x1024 does not crash, but it hogged all my memory from GPU (I have done 1920x1080 images before with Comfy without issues, with Flux 1 schell, but probably that's another topic) Literally everything else goes, like 1025x1024

Feb 06 '25 18:02 mpulukkinen

Looking at that stack trace I'm pretty sure that's an issue in stable-diffusion.cpp in which case I can't help. Please try the same model with the reference application over there (https://github.com/leejet/stable-diffusion.cpp/releases).

Feb 06 '25 20:02 DarthAffe

Actually right after commenting here, the integration started to crash on all resolutions, app restart did not work. I tried one other c# "wrapper" for the cpp code and that did not crash. I'll try with the latest cpp release later this today. How up-to-date this is with cpp version? I noticed the cpp version does not need everything to be converted to .gguf, so I'm thinking should I just use cpp dll's directly?

Feb 07 '25 05:02 mpulukkinen

OK, looks like there's issue with the latest .cpp: CUDA error: no kernel image is available for execution on the device current device: 0, in function ggml_cuda_compute_forward at D:\a\stable-diffusion.cpp\stable-diffusion.cpp\ggml\src\ggml-cuda\ggml-cuda.cu:2174 So, pretty much this: https://github.com/leejet/stable-diffusion.cpp/issues/554

Feb 07 '25 06:02 mpulukkinen

This is the last stable diffusion cpp release that works for me: https://github.com/leejet/stable-diffusion.cpp/releases/tag/master-b5f4932 After that, the cuda-dll-release size dropped dramatically and probably does not have all necessary files... (I tried with sd3.5)

Feb 07 '25 15:02 mpulukkinen

How up-to-date this is with cpp version?

The current Backend-Packages are using https://github.com/leejet/stable-diffusion.cpp/releases/tag/master-dcf91f9

OK, looks like there's issue with the latest .cpp: CUDA error: no kernel image is available for execution on the device

This should be fixed with https://github.com/leejet/stable-diffusion.cpp/commit/4fe83d52cf4506ea4fb87ccd5ed02efe69dc4167, but it seems like they dropped support for older architectures (< 2000 series) in https://github.com/leejet/stable-diffusion.cpp/commit/2535ad5a437478da00bc9b6d08dc6b36d1fc2e44

Feb 14 '25 09:02 DarthAffe