lllyasviel / lllyasviel/stable-diffusion-webui-forge

I can't reproduce an image that I had made /one month ago/ on the same forge, on the same pc

Open
#2,558 2 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
13k
Forks
1.7k
PR merge metrics
No merged PRs in 30d

Description

I spent 10 hours trying but I can't get out of it, the images are particularly different even if the settings are identical:

one month ago:
Pharfff, one-piece swimsuit, The photograph features an asian woman with long, straight hair, her lips are closed, she is looking at camera, she stands confidently on what appears to be a runway or stage, the background shows an audience and bright studio lights, she is wearing a black high-cut bodysuit with sheer panels across the front and sides, accentuated with a glossy finish and the "PHARFAITE" logo on the chest, her outfit exudes a sleek, futuristic style, accessorized with large dangling pearl earrings and a wristband, her makeup includes light blush, subtle eyeshadow, and nude lipstick, the mood is energetic and dynamic, capturing a moment of poised performance, she is wearing black heels and the photo is a full shot, highly detailed, realistic,
Steps: 35, Sampler: DPM++ 2M, Schedule type: SGM Uniform, CFG scale: 1, Distilled CFG Scale: 3.5, Seed: 7679530717, Size: 896x1152, Model hash: 3f97fdc57a, Model: flux1-dev, Denoising strength: 0.36, Hires CFG Scale: 1, Hires Distilled CFG Scale: 3.5, Hires upscale: 1.5, Hires steps: 20, Hires upscaler: 4x_NMKD-Siax_200k, Lora hashes: "Pharfff_flux1dev16_V256_V2.1: 7353a377d8b7", Version: f2.0.1v1.10.1-previous-560-g7eb824a0, Module 1: ae, Module 2: t5xxl_fp16, Module 3: clip_l

now
Pharfff, one-piece swimsuit, The photograph features an asian woman with long, straight hair, her lips are closed, she is looking at camera, she stands confidently on what appears to be a runway or stage, the background shows an audience and bright studio lights, she is wearing a black high-cut bodysuit with sheer panels across the front and sides, accentuated with a glossy finish and the "PHARFAITE" logo on the chest, her outfit exudes a sleek, futuristic style, accessorized with large dangling pearl earrings and a wristband, her makeup includes light blush, subtle eyeshadow, and nude lipstick, the mood is energetic and dynamic, capturing a moment of poised performance, she is wearing black heels and the photo is a full shot, highly detailed, realistic,
Steps: 35, Sampler: DPM++ 2M, Schedule type: SGM Uniform, CFG scale: 1, Distilled CFG Scale: 3.5, Seed: 7679530717, Size: 896x1152, Model hash: 3f97fdc57a, Model: flux1-dev, Denoising strength: 0.36, Hires Module 1: Use same choices, Hires CFG Scale: 1, Hires Distilled CFG Scale: 3.5, Hires upscale: 1.5, Hires steps: 20, Hires upscaler: 4x_NMKD-Siax_200k, Lora hashes: "Pharfff_flux1dev16_V256_V2.1: 7353a377d8b7", SGM noise multiplier: True, Version: f2.0.1v1.10.1-previous-636-gb835f24a, Module 1: ae, Module 2: t5xxl_fp16, Module 3: clip_l

version: f2.0.1v1.10.1-previous-636-gb835f24a  •  python: 3.10.6  •  torch: 2.3.1+cu121  •  xformers: 0.0.27  •  gradio: 4.40.0  •  checkpoint: 3f97fdc57a

the image changes drastically, i think it's the way the LoRa is loaded, even though i tried all the LoRa loading options and the image doesn't change. if anyone has any ideas i'd be happy to listen. i tried to reproduce another image i made 1 month ago without the time and that one comes out 99% the same. thanks for your time.

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

No source files or tests are identified. Start by reproducing the image with the supplied seed, model hash, LoRA hash, settings, and hardware/software versions, then compare the old and new metadata, including the SGM noise multiplier and Hires Module entry. Done means identifying a minimal difference that explains the output change or documenting a complete reproducible case.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.