jinx/ComfyUI
1
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2025-08-03 07:26:31 +08:00
Code Issues Activity

Add PixArt model support (#6055)

Browse Source 
* PixArt initial version

* PixArt Diffusers convert logic

* pos_emb and interpolation logic

* Reduce  duplicate code

* Formatting

* Use optimized attention

* Edit empty token logic

* Basic PixArt LoRA support

* Fix aspect ratio logic

* PixArtAlpha text encode with conds

* Use same detection key logic for PixArt diffusers
This commit is contained in:
City
2024-12-20 21:25:00 +01:00
committed by GitHub
parent 418eb7062d
commit bddb02660c
13 changed files with 1079 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
comfy/supported_models.py
View File

@@ -8,6 +8,7 @@ import comfy.text_encoders.sd2_clip
import comfy.text_encoders.sd3_clip
import comfy.text_encoders.sa_t5
import comfy.text_encoders.aura_t5
import comfy.text_encoders.pixart_t5
import comfy.text_encoders.hydit
import comfy.text_encoders.flux
import comfy.text_encoders.genmo
@@ -592,6 +593,37 @@ class AuraFlow(supported_models_base.BASE):
def clip_target(self, state_dict={}):
return supported_models_base.ClipTarget(comfy.text_encoders.aura_t5.AuraT5Tokenizer, comfy.text_encoders.aura_t5.AuraT5Model)
class PixArtAlpha(supported_models_base.BASE):
unet_config = {
"image_model": "pixart_alpha",
}
sampling_settings = {
"beta_schedule" : "sqrt_linear",
"linear_start" : 0.0001,
"linear_end" : 0.02,
"timesteps" : 1000,
}
unet_extra_config = {}
latent_format = latent_formats.SD15
vae_key_prefix = ["vae."]
text_encoder_key_prefix = ["text_encoders."]
def get_model(self, state_dict, prefix="", device=None):
out = model_base.PixArt(self, device=device)
return out.eval()
def clip_target(self, state_dict={}):
return supported_models_base.ClipTarget(comfy.text_encoders.pixart_t5.PixArtTokenizer, comfy.text_encoders.pixart_t5.PixArtT5XXL)
class PixArtSigma(PixArtAlpha):
unet_config = {
"image_model": "pixart_sigma",
}
latent_format = latent_formats.SDXL
class HunyuanDiT(supported_models_base.BASE):
unet_config = {
"image_model": "hydit",
@@ -787,6 +819,6 @@ class HunyuanVideo(supported_models_base.BASE):
hunyuan_detect = comfy.text_encoders.hunyuan_video.llama_detect(state_dict, "{}llama.transformer.".format(pref))
return supported_models_base.ClipTarget(comfy.text_encoders.hunyuan_video.HunyuanVideoTokenizer, comfy.text_encoders.hunyuan_video.hunyuan_video_clip(**hunyuan_detect))
models = [Stable_Zero123, SD15_instructpix2pix, SD15, SD20, SD21UnclipL, SD21UnclipH, SDXL_instructpix2pix, SDXLRefiner, SDXL, SSD1B, KOALA_700M, KOALA_1B, Segmind_Vega, SD_X4Upscaler, Stable_Cascade_C, Stable_Cascade_B, SV3D_u, SV3D_p, SD3, StableAudio, AuraFlow, HunyuanDiT, HunyuanDiT1, FluxInpaint, Flux, FluxSchnell, GenmoMochi, LTXV, HunyuanVideo]
models = [Stable_Zero123, SD15_instructpix2pix, SD15, SD20, SD21UnclipL, SD21UnclipH, SDXL_instructpix2pix, SDXLRefiner, SDXL, SSD1B, KOALA_700M, KOALA_1B, Segmind_Vega, SD_X4Upscaler, Stable_Cascade_C, Stable_Cascade_B, SV3D_u, SV3D_p, SD3, StableAudio, AuraFlow, PixArtAlpha, PixArtSigma, HunyuanDiT, HunyuanDiT1, FluxInpaint, Flux, FluxSchnell, GenmoMochi, LTXV, HunyuanVideo]
models += [SVD_img2vid]