3ded1a3a04
Refactor of sampler code to deal more easily with different model types.
1 year ago
5f57362613
Lower lora ram usage when in normal vram mode.
1 year ago
490771b7f4
Speed up lora loading a bit.
1 year ago
50b1180dde
Fix CLIPSetLastLayer not reverting when removed.
1 year ago
6fb084f39d
Reduce floating point rounding errors in loras.
1 year ago
91ed2815d5
Add a node to merge CLIP models.
1 year ago
6ad0a6d7e2
Don't patch weights when multiplier is zero.
1 year ago
a9a4ba7574
Fix merging not working when model2 of model merge node was a merge.
1 year ago
e7bee85df8
Add arguments to run the VAE in fp16 or bf16 for testing.
1 year ago
ddc6f12ad5
Disable autocast in unet for increased speed.
1 year ago
af7a49916b
Support loading unet files in diffusers format.
1 year ago
acf95191ff
Properly support SDXL diffusers loras for unet.
1 year ago
c3e96e637d
Pass device to CLIP model.
1 year ago
2c4e0b49b7
Switch to fp16 on some cards when the model is too big.
1 year ago
1c1b0e7299
--gpu-only now keeps the VAE on the device.
1 year ago
3b6fe51c1d
Leave text_encoder on the CPU when it can handle it.
1 year ago
b6a60fa696
Try to keep text encoders loaded and patched to increase speed.
...
load_model_gpu() is now used with the text encoder models instead of just
the unet.
1 year ago
97ee230682
Make highvram and normalvram shift the text encoders to vram and back.
...
This is faster on big text encoder models than running it on the CPU.
1 year ago
5a9ddf94eb
LoraLoader node now caches the lora file between executions.
1 year ago
62db11683b
Move unet to device right after loading on highvram mode.
1 year ago
2c7c14de56
Support for SDXL text encoder lora.
1 year ago
9b93b920be
Add CheckpointSave node to save checkpoints.
...
The created checkpoints contain workflow metadata that can be loaded by
dragging them on top of the UI or loading them with the "Load" button.
Checkpoints will be saved in fp16 or fp32 depending on the format ComfyUI
is using for inference on your hardware. To force fp32 use: --force-fp32
Anything that patches the model weights like merging or loras will be
saved.
The output directory is currently set to: output/checkpoints but that might
change in the future.
1 year ago
b72a7a835a
Support loras based on the stability unet implementation.
1 year ago
20f579d91d
Add DualClipLoader to load clip models for SDXL.
...
Update LoadClip to load clip models for SDXL refiner.
1 year ago
b7933960bb
Fix CLIPLoader node.
1 year ago
05676942b7
Add some more transformer hooks and move tomesd to comfy_extras.
...
Tomesd now uses q instead of x to decide which tokens to merge because
it seems to give better results.
1 year ago
8607c2d42d
Move latent scale factor from VAE to model.
1 year ago
30a3861946
Fix bug when yaml config has no clip params.
1 year ago
9e37f4c7d5
Fix error with ClipVision loader node.
1 year ago
9f83b098c9
Don't merge weights when shapes don't match and print a warning.
1 year ago
f87ec10a97
Support base SDXL and SDXL refiner models.
...
Large refactor of the model detection and loading code.
1 year ago
51581dbfa9
Fix last commits causing an issue with the text encoder lora.
1 year ago
8125b51a62
Keep a set of model_keys for faster add_patches.
1 year ago
45beebd33c
Add a type of model patch useful for model merging.
1 year ago
8883cb0f67
Add a way to set patches that modify the attn2 output.
...
Change the transformer patches function format to be more future proof.
1 year ago
fb4bf7f591
This is not needed anymore and causes issues with alphas_cumprod.
1 year ago
f7edcfd927
Add a --gpu-only argument to keep and run everything on the GPU.
...
Make the CLIP model work on the GPU.
1 year ago
6b774589a5
Set model to fp16 before loading the state dict to lower ram bump.
1 year ago
388567f20b
sampler_cfg_function now uses a dict for the argument.
...
This means arguments can be added without issues.
1 year ago
ff9b22d79e
Turn on safe load for a few models.
1 year ago
f0a2b81cd0
Cleanup: Remove a bunch of useless files.
1 year ago
f8c5931053
Split the batch in VAEEncode if there's not enough memory.
1 year ago
c069fc0730
Auto switch to tiled VAE encode if regular one runs out of memory.
1 year ago
de142eaad5
Simpler base model code.
1 year ago
0e425603fb
Small refactor.
1 year ago
700491d81a
Implement global average pooling for controlnet.
1 year ago
03da8a3426
This is useless for inference.
1 year ago
eb448dd8e1
Auto load model in lowvram if not enough memory.
1 year ago
a532888846
Support VAEs in diffusers format.
2 years ago
19c014f429
comment out annoying print statement
2 years ago
comfyanonymous
BlenderNeko