search

openvinotoolkit / openvino.genai

Run Generative AI models with simple C++/Python API and using OpenVINO Runtime
Apache License 2.0
155 stars 174 forks source link

the quality of generated image is low by benchmark on stable diffusion v1.5 #443

Open aoke79 opened 6 months ago

aoke79 commented 6 months ago

I've run the benchmark to test stable diffusion v1.5, but the quality of generated image is low. I doubted there are something computing wrong in the process. I compared it with optimum-intel. and set the same prompt, steps (20), resolution (512x512), and got very different quality image. please have a look at.

xx_1 stable-diffusion-optimum-sdv1_5_p0_iter0_pid15516_output

andrei-kochin commented 6 months ago

@aoke79 please share the command you've run to get those images in optimum and in the benchmark?

Thank you!

aoke79 commented 6 months ago

python benchmark.py -d GPU --model "C:\AIGC\openvino\models\stable-diffusion-sdv1_5" --prompt_file "prompts/stable-diffusion.jsonl" -n 3 the model was converted from hugging-face runway stable diffusion.

e-ddykim commented 6 months ago

@aoke79 Please check guidance_scale in https://github.com/openvinotoolkit/openvino.genai/blob/master/llm_bench/python/prompts/stable-diffusion.jsonl. Could you try it with the value of 7.5?

Edward-Lin commented 5 months ago

I have set up the same parameters for both cases. it should not be the root cause.

Edward-Lin commented 4 months ago

I've udpated the environments, like below, but the generated image is still wrong, nothing better. python -m pip install --upgrade pip pip install --pre -U openvino --extra-index-url https://storage.openvinotoolkit.org/simple/wheels/nightly

please refer what I've attached for the details. ov_pip_list.txt stable-diffusion.jsonl.txt

geunhwan commented 4 months ago

@Edward-Lin Please check above comment. guidance_scale 7.5

Edward-Lin commented 4 months ago

@Edward-Lin Please check above comment. guidance_scale 7.5

Have you ever tried? it did not work for me.