-- Currently stencil feature works with 512x512 images only.
-- This commit relaxes this constraint and adds support for various
image sizes.
Signed-off-by: Abhishek Varma <abhishek@nod-labs.com>
* [WIP] Add ControlNet to SD pipeline
-- This commit adds ControlNet to SD pipeline.
Signed-off-by: Abhishek Varma <abhishek@nod-labs.com>
* [SD] Add ControlNet to img2img + fix bug for img2img scheduler
-- This commit adds ControlNet execution to img2img.
-- It restructures the addition of ControlNet variants.
-- It also fixes scheduler selecting bug for img2img pipeline.
Signed-off-by: Abhishek Varma <abhishek@nod-labs.com>
* add shark models for stencilSD
* Add Stencil controlled SD in img2img pipeline (#1106)
* use shark stencil modules
* adjust diffusers change
* modify to use pipeline
* remove control from unet
* pump stencils through unet
* complete integration in img2img
* fix lint and comments
* [SD] Add ControlNet pipeline + integrate with WebUI + add compiled flow execution
-- This commit creates a dedicated SD pipeline for ControlNet.
-- Integrates it with img2img WebUI.
-- Integrates the compiled execution flow for ControlNet.
Signed-off-by: Abhishek Varma <abhishek@nod-labs.com>
* [SD] Stencil execution
* Remove integration setup
* [SD] Fix args.use_stencil overriding bug + vmfb caching issue
-- This commit fixes args.use_stencil overriding issue which caused
img2img pipeline to pick wrong set of modules.
-- It also fixes vmfb caching issue to speed up the loading time
and pick right set of modules based on a mask.
Signed-off-by: Abhishek Varma <abhishek@nod-labs.com>
---------
Signed-off-by: Abhishek Varma <abhishek@nod-labs.com>
Co-authored-by: Abhishek Varma <abhishek@nod-labs.com>
Co-authored-by: PhaneeshB <b.phaneesh@gmail.com>
* [SD] Save missing metadata in case of img2img and outpaint
Signed-Off-by: Gaurav Shukla <gaurav@nod-labs.com>
* [SD] Update the canvas size for inpaint/outpaint
Signed-Off-by: Gaurav Shukla <gaurav@nod-labs.com>
* [SD] Update output gallery on each inference
Signed-Off-by: Gaurav Shukla <gaurav@nod-labs.com>
---------
Signed-off-by: Gaurav Shukla <gaurav@nod-labs.com>
python apps/stable_diffusion/scripts/outpaint.py --prompt="Face of a yellow cat, high resolution, sitting on a park bench" --img_path=test_imgs/overture-creations-5sI6fQgYIuo.png --import_mlir --hf_model_id="stabilityai/stable-diffusion-2-inpainting" --pixels=128 --mask_blur=8 --left --right --top --bottom --steps=20
disables usage of low_cpu_mem_usage=True in from_pretrained() calls.
Can be re-enabled by using flag --low_cpu_mem_usage
defaults to False to avoid spam as we don't include accelerate in our
requirements.txt
-- This commit adds custom vae support to SD wherein the user can
point to a model's checkpoint file whose Vae needs to be plugged
into the main model.
Signed-off-by: Abhishek Varma <abhishek@nod-labs.com>
-- This commit prepones loading of vmfbs, if present, for all sub-models.
-- It also involves restructuring the SD pipeline to achieve the loading
of vmfbs smoothly and postpones processing of checkpoint files only when
required.
Signed-off-by: Abhishek Varma <abhishek@nod-labs.com>
The stable diffusion codebase has been reorganized to make it more
modular so that the same script can be used for web as well as cli,
instead of duplicating the whole codebase.
Signed-off-by: Gaurav Shukla <gaurav@nod-labs.com>
