lllyasviel
commented
3 weeks ago you can git checkout 08f74875 to use "f2.0.1v1.10.1-previous-401-g08f74875"
However, as I just tested, it gives exactly same results to the latest version.
You can test if you will have different results by git checkout 08f74875. However, it is likely that you also get exactly same image as you have now. If not, report here.
I somehow remember that "Heun" sampler has some deterministic issues and may have different results on different GPU architecture. It is likely that your mentioned "Author" is on a different GPU architecture and that may be the reason. By the way, I cannot get same result to either of you, which may also because of this
So I'm trying to reproduce the output of this lora. (https://civitai.com/models/652699?modelVersionId=756149)
I've been communicating with the author and he is on an older version of Forge and is able to get MUCH more realistic skin texture with the exact same settings but I am not able to!
Author's output:
Author's prompt:
Amateur photography of three women sitting closely together on a sunny day. The woman on the left is White, with a plus-size build, light skin, long straight brown hair, and wearing a black tank top and large dark sunglasses, smiling confidently at the camera. The woman in the middle is Black, with a slim build, dark skin, long braided hair, and wearing a yellow floral dress with a flower crown, smiling brightly. The woman on the right is Asian with a slim build, light skin, shoulder-length wavy dark brown hair, wearing a turquoise tank top and sunglasses, smiling warmly. The scene is an outdoor gathering, possibly a festival or picnic, with people seated on lawn chairs and blankets on the grass. The background features trees, a parked car, and other attendees dressed in casual summer attire, with some wearing leis, suggesting a relaxed, festive atmosphere. Image quality tags: bright natural light, slight motion blur in the background, vivid colors, casual setting, cluttered background. on flickr in 2007, 2005 blog, 2007 blog
Steps: 20, Sampler: Heun, Schedule type: Beta, CFG scale: 1, Distilled CFG Scale: 4, Seed: 573886816, Size: 1024x1024, Model hash: 52cfce60d7, Model: flux1-dev-Q8_0, Lora hashes: "amateurphotov2-000049: 771781fd6719", Beta schedule alpha: 0.6, Beta schedule beta: 0.6, Version: f2.0.1v1.10.1-previous-401-g08f74875, Module 1: ae, Module 2: clip_l, Module 3: t5xxl_fp8_e4m3fn
My output:
My prompt:
Amateur photography of three women sitting closely together on a sunny day. The woman on the left is White, with a plus-size build, light skin, long straight brown hair, and wearing a black tank top and large dark sunglasses, smiling confidently at the camera. The woman in the middle is Black, with a slim build, dark skin, long braided hair, and wearing a yellow floral dress with a flower crown, smiling brightly. The woman on the right is Asian with a slim build, light skin, shoulder-length wavy dark brown hair, wearing a turquoise tank top and sunglasses, smiling warmly. The scene is an outdoor gathering, possibly a festival or picnic, with people seated on lawn chairs and blankets on the grass. The background features trees, a parked car, and other attendees dressed in casual summer attire, with some wearing leis, suggesting a relaxed, festive atmosphere. Image quality tags: bright natural light, slight motion blur in the background, vivid colors, casual setting, cluttered background. on flickr in 2007, 2005 blog, 2007 blog
Steps: 20, Sampler: Heun, Schedule type: Beta, CFG scale: 1, Distilled CFG Scale: 4, Seed: 573886816, Size: 1024x1024, Model hash: 52cfce60d7, Model: flux1-dev-Q8_0, Lora hashes: "amateurphotov2: 771781fd6719", Beta schedule alpha: 0.6, Beta schedule beta: 0.6, Version: f2.0.1v1.10.1-previous-414-gdf598c4d, Module 1: ae, Module 2: clip_l, Module 3: t5xxl_fp8_e4m3fn
As you can see, I have matched each and every setting meticulously by contacting the author and asking him for the prompt, but there is still a vast difference in outputs and the only difference is that the author is on an older commit!
I would really appreciate if this is addressed, thanks in advance.