Do I need to specify guidance_scale =1.0 during inference?
rt. What's the expected guidance scale and true_cfg_scale for inference? I use guidance_scale=3.5 and true_cfg_scale=1.5 and the results are inferior to civitai showcases. The texture's are there but it still looks quite oily and plastic (like the original flux.1-dev).
do you use ComfyUI or ForgeWebUI? just use the model like standard Flux.1 Dev normally. or you can pull Civitai posted image to ComfyUI workshopsapce, it include workflow.
thanks for the info. I'm using diffusers pipeline and seems the sampler is rather crude. Based on your experience, what parameter is most crucial for photorealism? Is the deis sampler or max_shift/base_shift? (In diffusers there's only FlowMatchinDIscreteSampler which resembles Euler a, and default max_shift=1.15)