comfyanonymous/ComfyUI
1
0
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2026-04-30 03:11:33 +00:00
Code Issues Packages Projects Releases Wiki Activity

Feat: z-image pixel space (model still training atm) (#12709)

Browse Source 
* draft zeta (z-image pixel space)

* revert gitignore

* model loaded and able to run however vector direction still wrong tho

* flip the vector direction to original again this time

* Move wrongly positioned Z image pixel space class

* inherit Radiance LatentFormat class

* Fix parameters in classes for Zeta x0 dino

* remove arbitrary nn.init instances

* Remove unused import of lru_cache

---------

Co-authored-by: silveroxides <ishimarukaito@gmail.com>
This commit is contained in:
Lodestone
2026-03-03 07:43:47 +07:00
committed by GitHub
parent 57dd6c1aad
commit 9ebee0a217
5 changed files with 315 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
comfy/supported_models.py
View File

@@ -1118,6 +1118,20 @@ class ZImage(Lumina2):
hunyuan_detect = comfy.text_encoders.hunyuan_video.llama_detect(state_dict, "{}qwen3_4b.transformer.".format(pref))
return supported_models_base.ClipTarget(comfy.text_encoders.z_image.ZImageTokenizer, comfy.text_encoders.z_image.te(**hunyuan_detect))
class ZImagePixelSpace(ZImage):
unet_config = {
"image_model": "zimage_pixel",
}
# Pixel-space model: no spatial compression, operates on raw RGB patches.
latent_format = latent_formats.ZImagePixelSpace
# Much lower memory than latent-space models (no VAE, small patches).
memory_usage_factor = 0.05 # TODO: figure out the optimal value for this.
def get_model(self, state_dict, prefix="", device=None):
return model_base.ZImagePixelSpace(self, device=device)
class WAN21_T2V(supported_models_base.BASE):
unet_config = {
"image_model": "wan2.1",
@@ -1720,6 +1734,6 @@ class LongCatImage(supported_models_base.BASE):
hunyuan_detect = comfy.text_encoders.hunyuan_video.llama_detect(state_dict, "{}qwen25_7b.transformer.".format(pref))
return supported_models_base.ClipTarget(comfy.text_encoders.longcat_image.LongCatImageTokenizer, comfy.text_encoders.longcat_image.te(**hunyuan_detect))
models = [LotusD, 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, LongCatImage, FluxSchnell, GenmoMochi, LTXV, LTXAV, HunyuanVideo15_SR_Distilled, HunyuanVideo15, HunyuanImage21Refiner, HunyuanImage21, HunyuanVideoSkyreelsI2V, HunyuanVideoI2V, HunyuanVideo, CosmosT2V, CosmosI2V, CosmosT2IPredict2, CosmosI2VPredict2, ZImage, Lumina2, WAN22_T2V, WAN21_T2V, WAN21_I2V, WAN21_FunControl2V, WAN21_Vace, WAN21_Camera, WAN22_Camera, WAN22_S2V, WAN21_HuMo, WAN22_Animate, WAN21_FlowRVS, WAN21_SCAIL, Hunyuan3Dv2mini, Hunyuan3Dv2, Hunyuan3Dv2_1, HiDream, Chroma, ChromaRadiance, ACEStep, ACEStep15, Omnigen2, QwenImage, Flux2, Kandinsky5Image, Kandinsky5, Anima]
models = [LotusD, 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, LongCatImage, FluxSchnell, GenmoMochi, LTXV, LTXAV, HunyuanVideo15_SR_Distilled, HunyuanVideo15, HunyuanImage21Refiner, HunyuanImage21, HunyuanVideoSkyreelsI2V, HunyuanVideoI2V, HunyuanVideo, CosmosT2V, CosmosI2V, CosmosT2IPredict2, CosmosI2VPredict2, ZImagePixelSpace, ZImage, Lumina2, WAN22_T2V, WAN21_T2V, WAN21_I2V, WAN21_FunControl2V, WAN21_Vace, WAN21_Camera, WAN22_Camera, WAN22_S2V, WAN21_HuMo, WAN22_Animate, WAN21_FlowRVS, WAN21_SCAIL, Hunyuan3Dv2mini, Hunyuan3Dv2, Hunyuan3Dv2_1, HiDream, Chroma, ChromaRadiance, ACEStep, ACEStep15, Omnigen2, QwenImage, Flux2, Kandinsky5Image, Kandinsky5, Anima]
models += [SVD_img2vid]