Generate HDR images using SD
APACHE-2.0 License
Generate HDR images using StableDiffusionXL
How? By:
How fast is it? For standard 20 steps and with threshold of 0.2, it would run 16 base steps and then 3 x 4 steps for each different exposure for total of 28 steps - so 20% overhead.
Primary mode of operation for batch operations is via wrappers that
VENV
requirements
hdr.py
scriptWrappers:
hdr.sh
hdr.bat
...or do it manually and run hdr.py
script
python hdr.py --help
--dtype DTYPE torch dtype
--device DEVICE torch device
--model MODEL sd model
--width WIDTH image width
--height HEIGHT image height
--steps STEPS sampling steps
--seed SEED noise seed
--cfg CFG cfg scale
--sampler SAMPLER sd sampler
--prompt PROMPT prompt or prompts file
--negative NEGATIVE negative prompt or prompts file
--image IMAGE init image(s)
--strength STRENGTH denoise strength
--output OUTPUT output folder
--format FORMAT hdr file format: png,hdr,dng,tiff,all
--exp EXP exposure correction
--gamma GAMMA gamma adjustment for dng/hdr
--timestep TIMESTEP correction timestep
--save save interim images
--ldr create 8bpc hdr png image
--json save params to json
--debug debug log
--offload offload model components
Experimental mode for, well, experiments Based on streamlit framework which is not installed by default as its not required for batch operations
source venv/bin/activate pip install streamlit streamlit run hdr.py
[!TIP] you can pass standard args to hdr.py by using additiona
--
separator example: streamlit run hdr.py -- --model sdxl/model.safetensors