Description
I asked this question on the Discord channel, but did not receive a satisfactory answer, so as promised, I am adding this as an issue.
======================================================
Hey Team, Flux-schnell seems to work quickly and fairly accurately. Problem is that no matter what I try, running FLUX.1-dev gives me results that look like a grid(close up of a carpet). I have tried several settings, but still get this blurry/garbled/garbage image:
Seed : 925456 | Steps : 2 | Guidance Scale : 7.5 | Img Width : 768 | Img Height : 768 | Sampler : k_euler | Model : FLUX.1-dev | Mode : txt2img_flux | Selected Style : none PROMPT: dog
Running on Mac Mini M4, 16gb of RAM. Thanks for advising.
{
"prompt": "dog",
"is_adv_mode": true,
"selected_sd_model": "FLUX.1-dev",
"img_width": 576,
"img_height": 576,
"num_imgs": 1,
"num_steps": 18,
"selected_sd_style": "none",
"scheduler": "karras",
"guidance_scale": 10,
"is_vn_gpu_seed": false,
"do_v_prediction": false,
"is_clip_skip_2": false,
"applet_name": "txt2img",
"available_text_emb_words": {},
"pipeline": "flux",
"flux_model_type": "dev",
"image_no": 0,
"seed": 751576
}
Here are my logs:
Logs
objc[32001]: Class _TtC3NNC10_DataFrame is implemented in both /Volumes/Sabrent-1tb/DiffusionBee/DiffusionBee253.app/Contents/Resources/core/ncc_sd.dylib (0x15fda11f0) and /Volumes/Sabrent-1tb/DiffusionBee/DiffusionBee253.app/Contents/Resources/core/flux_dylib.dylib (0x16e5bf568). One of the two will be used. Which one is undefined.
objc[32001]: Class _TtCV3NNC9DataFrame9_Iterator is implemented in both /Volumes/Sabrent-1tb/DiffusionBee/DiffusionBee253.app/Contents/Resources/core/ncc_sd.dylib (0x15fda1298) and /Volumes/Sabrent-1tb/DiffusionBee/DiffusionBee253.app/Contents/Resources/core/flux_dylib.dylib (0x16e5bf618). One of the two will be used. Which one is undefined.
objc[32001]: Class _TtCV3NNC9DataFrameP33_FBF375775A84C182F15995D3DBCE712C13WrappedMapper is implemented in both /Volumes/Sabrent-1tb/DiffusionBee/DiffusionBee253.app/Contents/Resources/core/ncc_sd.dylib (0x15fda1340) and /Volumes/Sabrent-1tb/DiffusionBee/DiffusionBee253.app/Contents/Resources/core/flux_dylib.dylib (0x16e5bf6c8). One of the two will be used. Which one is undefined.
objc[32001]: Class _TtCV3NNC9DataFrameP33_FBF375775A84C182F15995D3DBCE712C17WrappedManyMapper is implemented in both /Volumes/Sabrent-1tb/DiffusionBee/DiffusionBee253.app/Contents/Resources/core/ncc_sd.dylib (0x15fda13f0) and /Volumes/Sabrent-1tb/DiffusionBee/DiffusionBee253.app/Contents/Resources/core/flux_dylib.dylib (0x16e5bf780). One of the two will be used. Which one is undefined.
There is another user who was unable to generate a substantial image, with significant noise, but this is very different as every generation has poor output no matter the prompt.
There are no other reports of this exact issue, hence opening a ticket.
