Open DavidDragonsage opened 1 day ago
"metadata_scheme": "fooocus",
The scheme of foocus does not support new models such as Flux, Kolors, and SD3m. So it should be 'simple' instead of 'fooocus' in the metadata scheme.
For better compatibility, the 'fooocus' option will be removed and the 'simple' and 'a1111' options will be retained in the future. 'simple' is compatible with 'fooocus' in the SDXL model
It has been resolved in the dev version and will be pushed to the official version soon. This was generated using the dev version:
Great news - thank you! 🙂
Checklist
What happened?
When generating an image using the RealvisXLv40_v40Bakedvae.safetensors base model and using the Fooocus Enhance style "win2_0916" creates a highly distorted image. The same image is produced correctly using "win2_0820".
This distortion occurs with all images I have attempted to create in win2_960 using the RealvisXLv40_v40Bakedvae base model and the Fooocus Enhance style. Image generation will be normal if the Fooocus Semi Realistic style is substituted for the Fooocus Enhance style.
I did not test this failure with various other XL base models, but RealvisXLv40_v40Bakedvae has never created these sort of problems before, and it is one of my favourite models.
Steps to reproduce the problem
1) Start SimpleSDXL win2_960 2) Find an image made with win2_820, created using the RealvisXLv40_v40Bakedvae base model and the Fooocus Enhance style 3) Load the image into the Metadata window and apply 4) Generate 5) Compare the images. If you check the image logs the parameters will be identical. However, the image created with win2_960 will be a junk image instead of a good image.
What should have happened?
The images created using both win2_960 and win2_820 should be identical. The first attached image was created in win2_820 and the second attachment was created in win2_960.
What browsers do you use to access Fooocus?
Mozilla Firefox
Where are you running Fooocus?
Locally
What operating system are you using?
Windows 10
Console logs
Additional information
It occurs to me that this anomaly may be related to the "[Bug]: Restore Detail to Flux1 Dev Image Generations #77" issue, in that it appears that the image generation algorithms have changed in some way between "win2_0820" and "win2_0916".
This is the image log. The logs are identical except for the version description: { "prompt": "full body long shot: a Norwegian mature woman (ambles:1.2) through the autumn with a (ambiguous:1.2) expression, 35mm lens, natural lighting, clearly defined facial features, sharp background, deep depth of field, (rim lighting:1.4)", "negative_prompt": "blurry background, bokeh, (NSFW, naked, nude, nudity:1.8)", "prompt_expansion": "full body long shot: a Norwegian mature woman (ambles:1.2) through the autumn with a (ambiguous:1.2) expression, 35mm lens, natural lighting, clearly defined facial features, sharp background, deep depth of field, (rim lighting:1.4), highly detailed, elegant, delicate, intricate, epic composition, cinematic light", "styles": "['Fooocus V2', 'Fooocus Enhance']", "performance": "Speed", "steps": 30, "resolution": "(896, 1152)", "guidance_scale": 7, "sharpness": 2, "adm_guidance": "(1.5, 0.8, 0.3)", "base_model": "RealvisXLv40_v40Bakedvae.safetensors", "refiner_model": "None", "refiner_switch": 0.6, "clip_skip": 2, "sampler": "dpmpp_2m", "scheduler": "karras", "vae": "Default (model)", "seed": "7492231340810660130", "lora_combined_1": "sd_xl_offset_example-lora_1.0.safetensors : 0.5", "backend_engine": "SDXL-Fooocus", "metadata_scheme": "fooocus", "version": "Fooocus v2.5.5 SimpleSDXL_v20240731.baa9f01" }