Artefacts when running Krea 2

[Lusus-Naturae]

Custom Node Testing

• I have tried disabling custom nodes and the issue persists (see how to disable custom nodes if you need help)

Your question

I've been having an issue with image generation since yesterday. All images generated with Krea 2 come with a weird noise pattern. I've disabled all custom nodes and loras, but the issue persist even with the basic workflow. I've tried using regular fp8 over int8-convrot, but no luck. Attached are two examples:

Logs

** ComfyUI startup time: 2026-07-01 18:43:17.068
[2026-07-01 18:43:17.068] ** Platform: Linux
[2026-07-01 18:43:17.068] ** Python version: 3.14.6 (main, Jun 23 2026, 18:14:05) [GCC 16.1.1 20260430]
[2026-07-01 18:43:17.068] ** Python executable: /home/lucie/Bureau/ComfyUI/venv/bin/python
[2026-07-01 18:43:17.068] ** ComfyUI Path: /home/lucie/Bureau/ComfyUI
[2026-07-01 18:43:17.068] ** ComfyUI Base Folder Path: /home/lucie/Bureau/ComfyUI
[2026-07-01 18:43:17.068] ** User directory: /home/lucie/Bureau/ComfyUI/user
[2026-07-01 18:43:17.068] ** ComfyUI-Manager config path: /home/lucie/Bureau/ComfyUI/user/__manager/config.ini
[2026-07-01 18:43:17.068] ** Log path: /home/lucie/Bureau/ComfyUI/user/comfyui.log
[PRE] ComfyUI-Manager
[2026-07-01 18:43:19.347] Found comfy_kitchen backend eager: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['adaln', 'apply_rope', 'apply_rope1', 'apply_rope_split_half', 'apply_rope_split_half1', 'dequantize_int8_convrot_weight', 'dequantize_int8_convrot_weight_dtype', 'dequantize_int8_simple', 'dequantize_int8_simple_dtype', 'dequantize_mxfp8', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'gemv_awq_w4a16', 'int8_linear', 'quantize_and_rotate_rowwise', 'quantize_int8_convrot_weight', 'quantize_int8_rowwise', 'quantize_int8_tensorwise', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'quantize_svdquant_w4a4', 'scaled_mm_mxfp8', 'scaled_mm_nvfp4', 'scaled_mm_svdquant_w4a4', 'stochastic_rounding_fp8']}
[2026-07-01 18:43:19.347] Found comfy_kitchen backend triton: {'available': True, 'disabled': True, 'unavailable_reason': None, 'capabilities': ['adaln', 'apply_rope', 'apply_rope1', 'apply_rope_split_half', 'apply_rope_split_half1', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'int8_linear', 'quantize_and_rotate_rowwise', 'quantize_int8_rowwise', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8']}
[2026-07-01 18:43:19.347] Found comfy_kitchen backend cuda: {'available': True, 'disabled': False, 'unavailable_reason': None, 'capabilities': ['adaln', 'apply_rope', 'apply_rope1', 'apply_rope_split_half', 'apply_rope_split_half1', 'dequantize_int8_convrot_weight', 'dequantize_int8_convrot_weight_dtype', 'dequantize_int8_simple', 'dequantize_int8_simple_dtype', 'dequantize_nvfp4', 'dequantize_per_tensor_fp8', 'gemv_awq_w4a16', 'int8_linear', 'quantize_and_rotate_rowwise', 'quantize_int8_convrot_weight', 'quantize_int8_rowwise', 'quantize_int8_tensorwise', 'quantize_mxfp8', 'quantize_nvfp4', 'quantize_per_tensor_fp8', 'quantize_svdquant_w4a4', 'scaled_mm_nvfp4', 'scaled_mm_svdquant_w4a4', 'stochastic_rounding_fp8']}
[2026-07-01 18:43:19.352] Checkpoint files will always be loaded safely.
[2026-07-01 18:43:19.752] Total VRAM 24090 MB, total RAM 48070 MB
[2026-07-01 18:43:19.753] pytorch version: 2.12.1+cu132
[2026-07-01 18:43:19.753] Set vram state to: NORMAL_VRAM
[2026-07-01 18:43:19.753] Device: cuda:0 NVIDIA GeForce RTX 3090 : cudaMallocAsync
[2026-07-01 18:43:19.753] Using async weight offloading with 2 streams
[2026-07-01 18:43:19.753] Enabled pinned memory 43263.0
[2026-07-01 18:43:20.740] Using pytorch attention
[2026-07-01 18:43:22.564] DynamicVRAM support detected and enabled
[2026-07-01 18:43:22.564] Python version: 3.14.6 (main, Jun 23 2026, 18:14:05) [GCC 16.1.1 20260430]
[2026-07-01 18:43:22.564] ComfyUI version: 0.27.0
[2026-07-01 18:43:22.564] comfy-aimdo version: 0.4.10
[2026-07-01 18:43:22.565] comfy-kitchen version: 0.2.16
[2026-07-01 18:43:22.566] comfyui-frontend-package version: 1.45.20
[2026-07-01 18:43:22.566] comfyui-workflow-templates version: 0.11.1
[2026-07-01 18:43:22.566] comfyui-embedded-docs version: 0.5.6
[2026-07-01 18:43:22.566] comfy-kitchen version: 0.2.16
[2026-07-01 18:43:22.566] comfy-aimdo version: 0.4.10
[2026-07-01 18:43:22.567] [Prompt Server] web root: /home/lucie/Bureau/ComfyUI/venv/lib/python3.14/site-packages/comfyui_frontend_package/static
[2026-07-01 18:43:22.568] Asset seeder disabled
[2026-07-01 18:43:22.568] [START] ComfyUI-Manager
[2026-07-01 18:43:22.601] [ComfyUI-Manager] Using GitPython backend
[2026-07-01 18:43:22.636] [ComfyUI-Manager] network_mode: public
[2026-07-01 18:43:23.266] No OpenGL_accelerate module loaded: No module named 'OpenGL_accelerate'
[2026-07-01 18:43:24.171] Skipping loading of custom nodes
[2026-07-01 18:43:24.173] Context impl SQLiteImpl.
[2026-07-01 18:43:24.173] Will assume non-transactional DDL.
[2026-07-01 18:43:24.213] Using RAM pressure cache.
[2026-07-01 18:43:24.214] Starting server

[2026-07-01 18:43:24.214] To see the GUI go to: http://127.0.0.1:8188
[2026-07-01 18:44:14.112] got prompt
[2026-07-01 18:44:14.128] Using pytorch attention in VAE
[2026-07-01 18:44:14.129] Using pytorch attention in VAE
[2026-07-01 18:44:14.220] VAE load device: cuda:0, offload device: cpu, dtype: torch.bfloat16
[2026-07-01 18:44:14.580] CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cpu, dtype: torch.float16
[2026-07-01 18:44:14.583] Requested to load Krea2TEModel_
[2026-07-01 18:44:14.675] Model Krea2TEModel_ prepared for dynamic VRAM loading. 8463MB Staged. 0 patches attached. Force pre-loaded 249 weights: 627 KB.
[2026-07-01 18:44:19.139] Found quantization metadata version 1
[2026-07-01 18:44:19.139] Detected mixed precision quantization
[2026-07-01 18:44:19.139] Using mixed precision operations
[2026-07-01 18:44:19.139] Native ops: int8_tensorwise , emulated ops: nvfp4, float8_e5m2, float8_e4m3fn, mxfp8
[2026-07-01 18:44:19.152] model weight dtype torch.bfloat16, manual cast: torch.bfloat16
[2026-07-01 18:44:19.155] model_type FLUX
[2026-07-01 18:44:19.198] Requested to load Krea2
[2026-07-01 18:44:19.278] Model Krea2 prepared for dynamic VRAM loading. 12864MB Staged. 0 patches attached. Force pre-loaded 160 weights: 2824 KB.
[2026-07-01 18:45:07.856] 
100%|█████████████████████████████████████████████████████████████████████████| 8/8 [00:44<00:00,  6.34s/it]
100%|█████████████████████████████████████████████████████████████████████████| 8/8 [00:44<00:00,  5.53s/it]
[2026-07-01 18:45:11.341] Requested to load WanVAE
[2026-07-01 18:45:11.344] 0 models unloaded.
[2026-07-01 18:45:11.361] Model WanVAE prepared for dynamic VRAM loading. 241MB Staged. 0 patches attached. Force pre-loaded 60 weights: 61 KB.
[2026-07-01 18:45:12.954] Prompt executed in 58.84 seconds
[2026-07-01 18:45:24.257] got prompt
[2026-07-01 18:45:24.376] Model Krea2TEModel_ prepared for dynamic VRAM loading. 8463MB Staged. 0 patches attached. Force pre-loaded 249 weights: 627 KB.
[2026-07-01 18:45:26.286] 0 models unloaded.
[2026-07-01 18:45:26.350] Model Krea2 prepared for dynamic VRAM loading. 12864MB Staged. 0 patches attached. Force pre-loaded 160 weights: 2824 KB.
[2026-07-01 18:45:31.951] FETCH ComfyRegistry Data [DONE]
[2026-07-01 18:45:32.144] [ComfyUI-Manager] default cache updated: https://api.comfy.org/nodes
[2026-07-01 18:45:32.161] FETCH DATA from: /home/lucie/Bureau/ComfyUI/user/__manager/cache/1514988643_custom-node-list.json [DONE]
[2026-07-01 18:45:32.202] [ComfyUI-Manager] All startup tasks have been completed.
[2026-07-01 18:46:15.053] 
100%|█████████████████████████████████████████████████████████████████████████| 8/8 [00:44<00:00,  6.35s/it]
100%|█████████████████████████████████████████████████████████████████████████| 8/8 [00:44<00:00,  5.62s/it]
[2026-07-01 18:46:18.645] 0 models unloaded.
[2026-07-01 18:46:18.662] Model WanVAE prepared for dynamic VRAM loading. 241MB Staged. 0 patches attached. Force pre-loaded 60 weights: 61 KB.
[2026-07-01 18:46:19.995] Prompt executed in 55.74 seconds
[2026-07-01 18:47:25.344] got prompt
[2026-07-01 18:47:25.445] Model Krea2TEModel_ prepared for dynamic VRAM loading. 8463MB Staged. 0 patches attached. Force pre-loaded 249 weights: 627 KB.
[2026-07-01 18:47:28.637] 0 models unloaded.
[2026-07-01 18:47:28.690] Model Krea2 prepared for dynamic VRAM loading. 12864MB Staged. 0 patches attached. Force pre-loaded 160 weights: 2824 KB.
[2026-07-01 18:48:16.415] 
100%|█████████████████████████████████████████████████████████████████████████| 8/8 [00:44<00:00,  6.51s/it]
100%|█████████████████████████████████████████████████████████████████████████| 8/8 [00:44<00:00,  5.51s/it]
[2026-07-01 18:48:19.868] 0 models unloaded.
[2026-07-01 18:48:19.886] Model WanVAE prepared for dynamic VRAM loading. 241MB Staged. 0 patches attached. Force pre-loaded 60 weights: 61 KB.
[2026-07-01 18:48:21.179] Prompt executed in 55.83 seconds
[2026-07-01 18:48:46.587] got prompt
[2026-07-01 18:48:46.668] Model Krea2 prepared for dynamic VRAM loading. 12864MB Staged. 0 patches attached. Force pre-loaded 160 weights: 2824 KB.
[2026-07-01 18:49:34.831] 
100%|█████████████████████████████████████████████████████████████████████████| 8/8 [00:44<00:00,  6.44s/it]
100%|█████████████████████████████████████████████████████████████████████████| 8/8 [00:44<00:00,  5.58s/it]
[2026-07-01 18:49:38.268] 0 models unloaded.
[2026-07-01 18:49:38.284] Model WanVAE prepared for dynamic VRAM loading. 241MB Staged. 0 patches attached. Force pre-loaded 60 weights: 61 KB.
[2026-07-01 18:49:39.514] Prompt executed in 52.92 seconds

Other

No response

[github-actions]

This issue is being marked stale because it has not had any activity for 30 days. Reply below within 7 days if your issue still isn't solved, and it will be left open. Otherwise, the issue will be closed automatically.