v0.6.0 - Latent Consistency Model Support

[saddam213]

Latent Consistency Models (LCM)

LCM ImageToImage and TextToImage now supported

Support has been added for Latent Consistency Models, these model dramatically reduce the amount of steps required to generate an image.

Plus a few more changes in this weeks build

1. Multiple pipeline support
2. Latent Consistency Model support
3. Improved BetaSchedule calculation
4. Support disabled classifier free guidance
5. Batch Prompt, multiple images per prompt now supported
6. DDIM, EulerDiscrete, KDPM2 and LCM schedulers added

Big thanks to @luosiallen for this awesome model, below is a link to the LCM project
https://github.com/luosiallen/latent-consistency-model

New Contributors

• @TheyCallMeHex made their first contribution in Added newly included DDIM scheduler to schedulers listed in the readme #5

Full Changelog: v0.5.0...v0.6.0

Nuget Packages

https://www.nuget.org/packages/OnnxStack.Core/0.6.0
https://www.nuget.org/packages/OnnxStack.StableDiffusion/0.6.0

---

This discussion was created from the release v0.6.0 - Latent Consistency Model Support.

[JohnClaw]

Could you add this upscaler (https://huggingface.co/ssube/stable-diffusion-x4-upscaler-onnx), please?

[saddam213]

This Fridays release will simplify model loading, and support model downloads via http and git clone, also one click load from local folder

@TheyCallMeHex is busy testing it now :)

[Amin456789]

i love it!

[saddam213]

Got 1,2 and 4 done, ran out of time to get batch and upscaler in, but 99% sure they will be done for next weeks release

v0.7.0 will be ready for download tomorrow :)

[saddam213]

Dark theme is a million time nicer on the eyes

[Amin456789]

thank u! yes it looks amazing! cant wait for it!

[Amin456789]

hey guys im back, great job Adam!
@TheyCallMeHex can u please make some fp16 models too? lyriel v1.6 and deliberate v2 or 3 will be great!
https://huggingface.co/stablediffusionapi/deliberate-v3/tree/main
https://huggingface.co/pagebrain/lyriel-v1-6
thanks

[TheyCallMeHex]

What's the lyriel one? It just has an empty readme?

https://huggingface.co/TheyCallMeHex/Deliberate-V3-ONNX

[Amin456789]

thanks buddy!
here a civitai one=
https://civitai.com/models/22922/lyriel
or
https://huggingface.co/nyxia/lyriel16/tree/main

[Amin456789]

btw it will be great if we can have fp16 models [they are half the size and use lower vram/ram and speed up things i guess]
i tried to convert deliberate onnx to fp16 with this guide but failed :D
https://onnxruntime.ai/docs/performance/model-optimizations/float16.html
got errors like = ImportError: cannot import name 'onnx_pb' at first but fixed it
then when it was to about convert unet model it gave me this = Message onnx.ModelProto exceeds maximum protobuf size of 2GB: 3439988952
after sometime i found out even onnx says they wont go over 2gb for quantization... but it seems optimum can do it or script of diffusers named convert_stable_diffusion_checkpoint_to_onnx.py we have to add --fp16 [but it wants cuda gpu i think], but it can be done with this too = https://github.com/Amblyopius/Stable-Diffusion-ONNX-FP16

@TheyCallMeHex can u please convert some of the models [lyriel and deliberate will be great] to fp16 and share if u r interested? it is getting too technical for me as i get errors left and right

Update: onnxstack works with quantizated int8 models too! even though quality is lower but it will be great if i can have some fp16 models to test

[Amin456789]

another LCM model released, didnt test it myself, i think it is anime related:
https://huggingface.co/furusu/LCM-Acertainty