Open aalok013 opened 2 months ago
-input-id-images-dir
expects a directory which contains input images. Try
--input-id-images-dir ./
in the command line.
BTW, can you point me to the safetensor version of the photomaker model file that gave the error?
I tired changing it to relative path. It's seems to identify the image now. But it's throwing the error.
Command:
./sd\sd.exe -M txt2img -m ./models/Stable-Diffusion/juggernaut_reborn.safetensors -p "a girl img" --sampling-method euler_a --steps 20 --schedule discrete -W 384 -H 384 -b 1 --cfg-scale 7 -s -1 --clip-skip 1 --embd-dir ./models/Embeddings/ --lora-model-dir ./models/Lora/ -t 0 --rng cuda -o F:\SD\sd-cpp-output/txt2img/39_photo.png --vae-on-cpu --color --stacked-id-embd-dir ./models/PhotoMaker/photomaker-v1.q8_0.gguf --input-id-images-dir ./models/PhotoMaker/image --normalize-input
Error:
D:\a\stable-diffusion.cpp\stable-diffusion.cpp\ggml\src\ggml.c:4702: GGML_ASSERT(ggml_can_repeat(b, a)) failed
And I tired using the original safetensors file as well. But that's giving another error.
D:\a\stable-diffusion.cpp\stable-diffusion.cpp\ggml\src\ggml.c:5792: GGML_ASSERT(ggml_nelements(a) == ggml_nelements(b)) failed
But this error is different from previous error I got when using safetensors file. It seems there is some problem when I'm using absolute path for photomaker.
I downloaded the model mentioned in the guide. Here is the url
I found the issue. It seems the problem is the model. I'm using 1.5 and I have to use SDXL. Although the results are not great, I'm able to run it now. But we still have an issue when using absolute path instead of relative path for PhotoMaker.
Thanks for the help.
I'm unable to run photomaker as it is unable to find input image. I'm using bellow command. And I'm using gguf model as safetensors model of photomaker is giving me another error
GGML_ASSERT(ggml_nelements(a) == ggml_nelements(b)) failed
./sd\sd.exe -M txt2img -m ./models/Stable-Diffusion/juggernaut_reborn.safetensors -p "a guy img" --sampling-method euler_a --steps 20 --schedule discrete -W 384 -H 384 -b 1 --cfg-scale 7 -s -1 --clip-skip 1 --embd-dir ./models/Embeddings/ --lora-model-dir ./models/Lora/ -t 0 --rng cuda -o ./output/txt2img/39_photo.png --vae-on-cpu --color --stacked-id-embd-dir ./models/PhotoMaker/photomaker-v1.q8_0.gguf --input-id-images-dir ./mage.jpg --normalize-input
I see the following log in console. And I tried using image without extension as shown in the guide as well. But it's not working either.