a094b45c93
Load clipvision model to GPU for faster performance.
1 year ago
1300a1bb4c
Text encoder should initially load on the offload_device not the regular.
1 year ago
f92074b84f
Move ModelPatcher to model_patcher.py
1 year ago
4798cf5a62
Implement loras with norm keys.
1 year ago
b8c7c770d3
Enable bf16-vae by default on ampere and up.
1 year ago
1c794a2161
Fallback to slice attention if xformers doesn't support the operation.
1 year ago
d935ba50c4
Make --bf16-vae work on torch 2.0
1 year ago
a57b0c797b
Fix lowvram model merging.
1 year ago
f72780a7e3
The new smart memory management makes this unnecessary.
1 year ago
c77f02e1c6
Move controlnet code to comfy/controlnet.py
1 year ago
15a7716fa6
Move lora code to comfy/lora.py
1 year ago
ec96f6d03a
Move text_projection to base clip model.
1 year ago
30eb92c3cb
Code cleanups.
1 year ago
51dde87e97
Try to free enough vram for control lora inference.
1 year ago
e3d0a9a490
Fix potential issue with text projection matrix multiplication.
1 year ago
cc44ade79e
Always shift text encoder to GPU when the device supports fp16.
1 year ago
a6ef08a46a
Even with forced fp16 the cpu device should never use it.
1 year ago
00c0b2c507
Initialize text encoder to target dtype.
1 year ago
f081017c1a
Save memory by storing text encoder weights in fp16 in most situations.
...
Do inference in fp32 to make sure quality stays the exact same.
1 year ago
afcb9cb1df
All resolutions now work with t2i adapter for SDXL.
1 year ago
85fde89d7f
T2I adapter SDXL.
1 year ago
cf5ae46928
Controlnet/t2iadapter cleanup.
1 year ago
763b0cf024
Fix control lora not working in fp32.
1 year ago
199d73364a
Fix ControlLora on lowvram.
1 year ago
d08e53de2e
Remove autocast from controlnet code.
1 year ago
0d7b0a4dc7
Small cleanups.
1 year ago
9225465975
Further tuning and fix mem_free_total.
1 year ago
2c096e4260
Add ipex optimize and other enhancements for Intel GPUs based on recent memory changes.
1 year ago
e9469e732d
--disable-smart-memory now disables loading model directly to vram.
1 year ago
c9b562aed1
Free more memory before VAE encode/decode.
1 year ago
b80c3276dc
Fix issue with gligen.
1 year ago
d6e4b342e6
Support for Control Loras.
...
Control loras are controlnets where some of the weights are stored in
"lora" format: an up and a down low rank matrice that when multiplied
together and added to the unet weight give the controlnet weight.
This allows a much smaller memory footprint depending on the rank of the
matrices.
These controlnets are used just like regular ones.
1 year ago
39ac856a33
ReVision support: unclip nodes can now be used with SDXL.
1 year ago
76d53c4622
Add support for clip g vision model to CLIPVisionLoader.
1 year ago
e59fe0537a
Fix referenced before assignment
...
For https://github.com/BlenderNeko/ComfyUI_TiledKSampler/issues/13
1 year ago
be9c5e25bc
Fix issue with not freeing enough memory when sampling.
1 year ago
ac0758a1a4
Fix bug with lowvram and controlnet advanced node.
1 year ago
c28db1f315
Fix potential issues with patching models when saving checkpoints.
1 year ago
3aee33b54e
Add --disable-smart-memory for those that want the old behaviour.
1 year ago
2be2742711
Fix issue with regular torch version.
1 year ago
89a0767abf
Smarter memory management.
...
Try to keep models on the vram when possible.
Better lowvram mode for controlnets.
1 year ago
2c97c30256
Support small diffusers controlnet so both types are now supported.
1 year ago
53f326a3d8
Support diffusers mini controlnets.
1 year ago
58f0c616ed
Fix clip vision issue with old transformers versions.
1 year ago
ae270f79bc
Fix potential issue with batch size and clip vision.
1 year ago
a2ce9655ca
Refactor unclip code.
1 year ago
9cc12c833d
CLIPVisionEncode can now encode multiple images.
1 year ago
0cb6dac943
Remove 3m from PR #1213 because of some small issues.
1 year ago
e244b2df83
Add sgm_uniform scheduler that acts like the default one in sgm.
1 year ago
58c7da3665
Gpu variant of dpmpp_3m_sde. Note: use 3m with exponential or karras.
1 year ago
comfyanonymous
Simon Lui
Alexopus