diffusers 0.27.1__py3-none-any.whl → 0.32.2__py3-none-any.whl

Sign up to get free protection for your applications and to get access to all the features.
Files changed (445) hide show
  1. diffusers/__init__.py +233 -6
  2. diffusers/callbacks.py +209 -0
  3. diffusers/commands/env.py +102 -6
  4. diffusers/configuration_utils.py +45 -16
  5. diffusers/dependency_versions_table.py +4 -3
  6. diffusers/image_processor.py +434 -110
  7. diffusers/loaders/__init__.py +42 -9
  8. diffusers/loaders/ip_adapter.py +626 -36
  9. diffusers/loaders/lora_base.py +900 -0
  10. diffusers/loaders/lora_conversion_utils.py +991 -125
  11. diffusers/loaders/lora_pipeline.py +3812 -0
  12. diffusers/loaders/peft.py +571 -7
  13. diffusers/loaders/single_file.py +405 -173
  14. diffusers/loaders/single_file_model.py +385 -0
  15. diffusers/loaders/single_file_utils.py +1783 -713
  16. diffusers/loaders/textual_inversion.py +41 -23
  17. diffusers/loaders/transformer_flux.py +181 -0
  18. diffusers/loaders/transformer_sd3.py +89 -0
  19. diffusers/loaders/unet.py +464 -540
  20. diffusers/loaders/unet_loader_utils.py +163 -0
  21. diffusers/models/__init__.py +76 -7
  22. diffusers/models/activations.py +65 -10
  23. diffusers/models/adapter.py +53 -53
  24. diffusers/models/attention.py +605 -18
  25. diffusers/models/attention_flax.py +1 -1
  26. diffusers/models/attention_processor.py +4304 -687
  27. diffusers/models/autoencoders/__init__.py +8 -0
  28. diffusers/models/autoencoders/autoencoder_asym_kl.py +15 -17
  29. diffusers/models/autoencoders/autoencoder_dc.py +620 -0
  30. diffusers/models/autoencoders/autoencoder_kl.py +110 -28
  31. diffusers/models/autoencoders/autoencoder_kl_allegro.py +1149 -0
  32. diffusers/models/autoencoders/autoencoder_kl_cogvideox.py +1482 -0
  33. diffusers/models/autoencoders/autoencoder_kl_hunyuan_video.py +1176 -0
  34. diffusers/models/autoencoders/autoencoder_kl_ltx.py +1338 -0
  35. diffusers/models/autoencoders/autoencoder_kl_mochi.py +1166 -0
  36. diffusers/models/autoencoders/autoencoder_kl_temporal_decoder.py +19 -24
  37. diffusers/models/autoencoders/autoencoder_oobleck.py +464 -0
  38. diffusers/models/autoencoders/autoencoder_tiny.py +21 -18
  39. diffusers/models/autoencoders/consistency_decoder_vae.py +45 -20
  40. diffusers/models/autoencoders/vae.py +41 -29
  41. diffusers/models/autoencoders/vq_model.py +182 -0
  42. diffusers/models/controlnet.py +47 -800
  43. diffusers/models/controlnet_flux.py +70 -0
  44. diffusers/models/controlnet_sd3.py +68 -0
  45. diffusers/models/controlnet_sparsectrl.py +116 -0
  46. diffusers/models/controlnets/__init__.py +23 -0
  47. diffusers/models/controlnets/controlnet.py +872 -0
  48. diffusers/models/{controlnet_flax.py → controlnets/controlnet_flax.py} +9 -9
  49. diffusers/models/controlnets/controlnet_flux.py +536 -0
  50. diffusers/models/controlnets/controlnet_hunyuan.py +401 -0
  51. diffusers/models/controlnets/controlnet_sd3.py +489 -0
  52. diffusers/models/controlnets/controlnet_sparsectrl.py +788 -0
  53. diffusers/models/controlnets/controlnet_union.py +832 -0
  54. diffusers/models/controlnets/controlnet_xs.py +1946 -0
  55. diffusers/models/controlnets/multicontrolnet.py +183 -0
  56. diffusers/models/downsampling.py +85 -18
  57. diffusers/models/embeddings.py +1856 -158
  58. diffusers/models/embeddings_flax.py +23 -9
  59. diffusers/models/model_loading_utils.py +480 -0
  60. diffusers/models/modeling_flax_pytorch_utils.py +2 -1
  61. diffusers/models/modeling_flax_utils.py +2 -7
  62. diffusers/models/modeling_outputs.py +14 -0
  63. diffusers/models/modeling_pytorch_flax_utils.py +1 -1
  64. diffusers/models/modeling_utils.py +611 -146
  65. diffusers/models/normalization.py +361 -20
  66. diffusers/models/resnet.py +18 -23
  67. diffusers/models/transformers/__init__.py +16 -0
  68. diffusers/models/transformers/auraflow_transformer_2d.py +544 -0
  69. diffusers/models/transformers/cogvideox_transformer_3d.py +542 -0
  70. diffusers/models/transformers/dit_transformer_2d.py +240 -0
  71. diffusers/models/transformers/dual_transformer_2d.py +9 -8
  72. diffusers/models/transformers/hunyuan_transformer_2d.py +578 -0
  73. diffusers/models/transformers/latte_transformer_3d.py +327 -0
  74. diffusers/models/transformers/lumina_nextdit2d.py +340 -0
  75. diffusers/models/transformers/pixart_transformer_2d.py +445 -0
  76. diffusers/models/transformers/prior_transformer.py +13 -13
  77. diffusers/models/transformers/sana_transformer.py +488 -0
  78. diffusers/models/transformers/stable_audio_transformer.py +458 -0
  79. diffusers/models/transformers/t5_film_transformer.py +17 -19
  80. diffusers/models/transformers/transformer_2d.py +297 -187
  81. diffusers/models/transformers/transformer_allegro.py +422 -0
  82. diffusers/models/transformers/transformer_cogview3plus.py +386 -0
  83. diffusers/models/transformers/transformer_flux.py +593 -0
  84. diffusers/models/transformers/transformer_hunyuan_video.py +791 -0
  85. diffusers/models/transformers/transformer_ltx.py +469 -0
  86. diffusers/models/transformers/transformer_mochi.py +499 -0
  87. diffusers/models/transformers/transformer_sd3.py +461 -0
  88. diffusers/models/transformers/transformer_temporal.py +21 -19
  89. diffusers/models/unets/unet_1d.py +8 -8
  90. diffusers/models/unets/unet_1d_blocks.py +31 -31
  91. diffusers/models/unets/unet_2d.py +17 -10
  92. diffusers/models/unets/unet_2d_blocks.py +225 -149
  93. diffusers/models/unets/unet_2d_condition.py +41 -40
  94. diffusers/models/unets/unet_2d_condition_flax.py +6 -5
  95. diffusers/models/unets/unet_3d_blocks.py +192 -1057
  96. diffusers/models/unets/unet_3d_condition.py +22 -27
  97. diffusers/models/unets/unet_i2vgen_xl.py +22 -18
  98. diffusers/models/unets/unet_kandinsky3.py +2 -2
  99. diffusers/models/unets/unet_motion_model.py +1413 -89
  100. diffusers/models/unets/unet_spatio_temporal_condition.py +40 -16
  101. diffusers/models/unets/unet_stable_cascade.py +19 -18
  102. diffusers/models/unets/uvit_2d.py +2 -2
  103. diffusers/models/upsampling.py +95 -26
  104. diffusers/models/vq_model.py +12 -164
  105. diffusers/optimization.py +1 -1
  106. diffusers/pipelines/__init__.py +202 -3
  107. diffusers/pipelines/allegro/__init__.py +48 -0
  108. diffusers/pipelines/allegro/pipeline_allegro.py +938 -0
  109. diffusers/pipelines/allegro/pipeline_output.py +23 -0
  110. diffusers/pipelines/amused/pipeline_amused.py +12 -12
  111. diffusers/pipelines/amused/pipeline_amused_img2img.py +14 -12
  112. diffusers/pipelines/amused/pipeline_amused_inpaint.py +13 -11
  113. diffusers/pipelines/animatediff/__init__.py +8 -0
  114. diffusers/pipelines/animatediff/pipeline_animatediff.py +122 -109
  115. diffusers/pipelines/animatediff/pipeline_animatediff_controlnet.py +1106 -0
  116. diffusers/pipelines/animatediff/pipeline_animatediff_sdxl.py +1288 -0
  117. diffusers/pipelines/animatediff/pipeline_animatediff_sparsectrl.py +1010 -0
  118. diffusers/pipelines/animatediff/pipeline_animatediff_video2video.py +236 -180
  119. diffusers/pipelines/animatediff/pipeline_animatediff_video2video_controlnet.py +1341 -0
  120. diffusers/pipelines/animatediff/pipeline_output.py +3 -2
  121. diffusers/pipelines/audioldm/pipeline_audioldm.py +14 -14
  122. diffusers/pipelines/audioldm2/modeling_audioldm2.py +58 -39
  123. diffusers/pipelines/audioldm2/pipeline_audioldm2.py +121 -36
  124. diffusers/pipelines/aura_flow/__init__.py +48 -0
  125. diffusers/pipelines/aura_flow/pipeline_aura_flow.py +584 -0
  126. diffusers/pipelines/auto_pipeline.py +196 -28
  127. diffusers/pipelines/blip_diffusion/blip_image_processing.py +1 -1
  128. diffusers/pipelines/blip_diffusion/modeling_blip2.py +6 -6
  129. diffusers/pipelines/blip_diffusion/modeling_ctx_clip.py +1 -1
  130. diffusers/pipelines/blip_diffusion/pipeline_blip_diffusion.py +2 -2
  131. diffusers/pipelines/cogvideo/__init__.py +54 -0
  132. diffusers/pipelines/cogvideo/pipeline_cogvideox.py +772 -0
  133. diffusers/pipelines/cogvideo/pipeline_cogvideox_fun_control.py +825 -0
  134. diffusers/pipelines/cogvideo/pipeline_cogvideox_image2video.py +885 -0
  135. diffusers/pipelines/cogvideo/pipeline_cogvideox_video2video.py +851 -0
  136. diffusers/pipelines/cogvideo/pipeline_output.py +20 -0
  137. diffusers/pipelines/cogview3/__init__.py +47 -0
  138. diffusers/pipelines/cogview3/pipeline_cogview3plus.py +674 -0
  139. diffusers/pipelines/cogview3/pipeline_output.py +21 -0
  140. diffusers/pipelines/consistency_models/pipeline_consistency_models.py +6 -6
  141. diffusers/pipelines/controlnet/__init__.py +86 -80
  142. diffusers/pipelines/controlnet/multicontrolnet.py +7 -182
  143. diffusers/pipelines/controlnet/pipeline_controlnet.py +134 -87
  144. diffusers/pipelines/controlnet/pipeline_controlnet_blip_diffusion.py +2 -2
  145. diffusers/pipelines/controlnet/pipeline_controlnet_img2img.py +93 -77
  146. diffusers/pipelines/controlnet/pipeline_controlnet_inpaint.py +88 -197
  147. diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py +136 -90
  148. diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl.py +176 -80
  149. diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl_img2img.py +125 -89
  150. diffusers/pipelines/controlnet/pipeline_controlnet_union_inpaint_sd_xl.py +1790 -0
  151. diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl.py +1501 -0
  152. diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl_img2img.py +1627 -0
  153. diffusers/pipelines/controlnet/pipeline_flax_controlnet.py +2 -2
  154. diffusers/pipelines/controlnet_hunyuandit/__init__.py +48 -0
  155. diffusers/pipelines/controlnet_hunyuandit/pipeline_hunyuandit_controlnet.py +1060 -0
  156. diffusers/pipelines/controlnet_sd3/__init__.py +57 -0
  157. diffusers/pipelines/controlnet_sd3/pipeline_stable_diffusion_3_controlnet.py +1133 -0
  158. diffusers/pipelines/controlnet_sd3/pipeline_stable_diffusion_3_controlnet_inpainting.py +1153 -0
  159. diffusers/pipelines/controlnet_xs/__init__.py +68 -0
  160. diffusers/pipelines/controlnet_xs/pipeline_controlnet_xs.py +916 -0
  161. diffusers/pipelines/controlnet_xs/pipeline_controlnet_xs_sd_xl.py +1111 -0
  162. diffusers/pipelines/ddpm/pipeline_ddpm.py +2 -2
  163. diffusers/pipelines/deepfloyd_if/pipeline_if.py +16 -30
  164. diffusers/pipelines/deepfloyd_if/pipeline_if_img2img.py +20 -35
  165. diffusers/pipelines/deepfloyd_if/pipeline_if_img2img_superresolution.py +23 -41
  166. diffusers/pipelines/deepfloyd_if/pipeline_if_inpainting.py +22 -38
  167. diffusers/pipelines/deepfloyd_if/pipeline_if_inpainting_superresolution.py +25 -41
  168. diffusers/pipelines/deepfloyd_if/pipeline_if_superresolution.py +19 -34
  169. diffusers/pipelines/deepfloyd_if/pipeline_output.py +6 -5
  170. diffusers/pipelines/deepfloyd_if/watermark.py +1 -1
  171. diffusers/pipelines/deprecated/alt_diffusion/modeling_roberta_series.py +11 -11
  172. diffusers/pipelines/deprecated/alt_diffusion/pipeline_alt_diffusion.py +70 -30
  173. diffusers/pipelines/deprecated/alt_diffusion/pipeline_alt_diffusion_img2img.py +48 -25
  174. diffusers/pipelines/deprecated/repaint/pipeline_repaint.py +2 -2
  175. diffusers/pipelines/deprecated/spectrogram_diffusion/pipeline_spectrogram_diffusion.py +7 -7
  176. diffusers/pipelines/deprecated/stable_diffusion_variants/pipeline_cycle_diffusion.py +21 -20
  177. diffusers/pipelines/deprecated/stable_diffusion_variants/pipeline_stable_diffusion_inpaint_legacy.py +27 -29
  178. diffusers/pipelines/deprecated/stable_diffusion_variants/pipeline_stable_diffusion_model_editing.py +33 -27
  179. diffusers/pipelines/deprecated/stable_diffusion_variants/pipeline_stable_diffusion_paradigms.py +33 -23
  180. diffusers/pipelines/deprecated/stable_diffusion_variants/pipeline_stable_diffusion_pix2pix_zero.py +36 -30
  181. diffusers/pipelines/deprecated/versatile_diffusion/modeling_text_unet.py +102 -69
  182. diffusers/pipelines/deprecated/versatile_diffusion/pipeline_versatile_diffusion.py +13 -13
  183. diffusers/pipelines/deprecated/versatile_diffusion/pipeline_versatile_diffusion_dual_guided.py +10 -5
  184. diffusers/pipelines/deprecated/versatile_diffusion/pipeline_versatile_diffusion_image_variation.py +11 -6
  185. diffusers/pipelines/deprecated/versatile_diffusion/pipeline_versatile_diffusion_text_to_image.py +10 -5
  186. diffusers/pipelines/deprecated/vq_diffusion/pipeline_vq_diffusion.py +5 -5
  187. diffusers/pipelines/dit/pipeline_dit.py +7 -4
  188. diffusers/pipelines/flux/__init__.py +69 -0
  189. diffusers/pipelines/flux/modeling_flux.py +47 -0
  190. diffusers/pipelines/flux/pipeline_flux.py +957 -0
  191. diffusers/pipelines/flux/pipeline_flux_control.py +889 -0
  192. diffusers/pipelines/flux/pipeline_flux_control_img2img.py +945 -0
  193. diffusers/pipelines/flux/pipeline_flux_control_inpaint.py +1141 -0
  194. diffusers/pipelines/flux/pipeline_flux_controlnet.py +1006 -0
  195. diffusers/pipelines/flux/pipeline_flux_controlnet_image_to_image.py +998 -0
  196. diffusers/pipelines/flux/pipeline_flux_controlnet_inpainting.py +1204 -0
  197. diffusers/pipelines/flux/pipeline_flux_fill.py +969 -0
  198. diffusers/pipelines/flux/pipeline_flux_img2img.py +856 -0
  199. diffusers/pipelines/flux/pipeline_flux_inpaint.py +1022 -0
  200. diffusers/pipelines/flux/pipeline_flux_prior_redux.py +492 -0
  201. diffusers/pipelines/flux/pipeline_output.py +37 -0
  202. diffusers/pipelines/free_init_utils.py +41 -38
  203. diffusers/pipelines/free_noise_utils.py +596 -0
  204. diffusers/pipelines/hunyuan_video/__init__.py +48 -0
  205. diffusers/pipelines/hunyuan_video/pipeline_hunyuan_video.py +687 -0
  206. diffusers/pipelines/hunyuan_video/pipeline_output.py +20 -0
  207. diffusers/pipelines/hunyuandit/__init__.py +48 -0
  208. diffusers/pipelines/hunyuandit/pipeline_hunyuandit.py +916 -0
  209. diffusers/pipelines/i2vgen_xl/pipeline_i2vgen_xl.py +33 -48
  210. diffusers/pipelines/kandinsky/pipeline_kandinsky.py +8 -8
  211. diffusers/pipelines/kandinsky/pipeline_kandinsky_combined.py +32 -29
  212. diffusers/pipelines/kandinsky/pipeline_kandinsky_img2img.py +11 -11
  213. diffusers/pipelines/kandinsky/pipeline_kandinsky_inpaint.py +12 -12
  214. diffusers/pipelines/kandinsky/pipeline_kandinsky_prior.py +10 -10
  215. diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2.py +6 -6
  216. diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_combined.py +34 -31
  217. diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_controlnet.py +10 -10
  218. diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_controlnet_img2img.py +10 -10
  219. diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_img2img.py +6 -6
  220. diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_inpainting.py +8 -8
  221. diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_prior.py +7 -7
  222. diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_prior_emb2emb.py +6 -6
  223. diffusers/pipelines/kandinsky3/convert_kandinsky3_unet.py +3 -3
  224. diffusers/pipelines/kandinsky3/pipeline_kandinsky3.py +22 -35
  225. diffusers/pipelines/kandinsky3/pipeline_kandinsky3_img2img.py +26 -37
  226. diffusers/pipelines/kolors/__init__.py +54 -0
  227. diffusers/pipelines/kolors/pipeline_kolors.py +1070 -0
  228. diffusers/pipelines/kolors/pipeline_kolors_img2img.py +1250 -0
  229. diffusers/pipelines/kolors/pipeline_output.py +21 -0
  230. diffusers/pipelines/kolors/text_encoder.py +889 -0
  231. diffusers/pipelines/kolors/tokenizer.py +338 -0
  232. diffusers/pipelines/latent_consistency_models/pipeline_latent_consistency_img2img.py +82 -62
  233. diffusers/pipelines/latent_consistency_models/pipeline_latent_consistency_text2img.py +77 -60
  234. diffusers/pipelines/latent_diffusion/pipeline_latent_diffusion.py +12 -12
  235. diffusers/pipelines/latte/__init__.py +48 -0
  236. diffusers/pipelines/latte/pipeline_latte.py +881 -0
  237. diffusers/pipelines/ledits_pp/pipeline_leditspp_stable_diffusion.py +80 -74
  238. diffusers/pipelines/ledits_pp/pipeline_leditspp_stable_diffusion_xl.py +85 -76
  239. diffusers/pipelines/ledits_pp/pipeline_output.py +2 -2
  240. diffusers/pipelines/ltx/__init__.py +50 -0
  241. diffusers/pipelines/ltx/pipeline_ltx.py +789 -0
  242. diffusers/pipelines/ltx/pipeline_ltx_image2video.py +885 -0
  243. diffusers/pipelines/ltx/pipeline_output.py +20 -0
  244. diffusers/pipelines/lumina/__init__.py +48 -0
  245. diffusers/pipelines/lumina/pipeline_lumina.py +890 -0
  246. diffusers/pipelines/marigold/__init__.py +50 -0
  247. diffusers/pipelines/marigold/marigold_image_processing.py +576 -0
  248. diffusers/pipelines/marigold/pipeline_marigold_depth.py +813 -0
  249. diffusers/pipelines/marigold/pipeline_marigold_normals.py +690 -0
  250. diffusers/pipelines/mochi/__init__.py +48 -0
  251. diffusers/pipelines/mochi/pipeline_mochi.py +748 -0
  252. diffusers/pipelines/mochi/pipeline_output.py +20 -0
  253. diffusers/pipelines/musicldm/pipeline_musicldm.py +14 -14
  254. diffusers/pipelines/pag/__init__.py +80 -0
  255. diffusers/pipelines/pag/pag_utils.py +243 -0
  256. diffusers/pipelines/pag/pipeline_pag_controlnet_sd.py +1328 -0
  257. diffusers/pipelines/pag/pipeline_pag_controlnet_sd_inpaint.py +1543 -0
  258. diffusers/pipelines/pag/pipeline_pag_controlnet_sd_xl.py +1610 -0
  259. diffusers/pipelines/pag/pipeline_pag_controlnet_sd_xl_img2img.py +1683 -0
  260. diffusers/pipelines/pag/pipeline_pag_hunyuandit.py +969 -0
  261. diffusers/pipelines/pag/pipeline_pag_kolors.py +1136 -0
  262. diffusers/pipelines/pag/pipeline_pag_pixart_sigma.py +865 -0
  263. diffusers/pipelines/pag/pipeline_pag_sana.py +886 -0
  264. diffusers/pipelines/pag/pipeline_pag_sd.py +1062 -0
  265. diffusers/pipelines/pag/pipeline_pag_sd_3.py +994 -0
  266. diffusers/pipelines/pag/pipeline_pag_sd_3_img2img.py +1058 -0
  267. diffusers/pipelines/pag/pipeline_pag_sd_animatediff.py +866 -0
  268. diffusers/pipelines/pag/pipeline_pag_sd_img2img.py +1094 -0
  269. diffusers/pipelines/pag/pipeline_pag_sd_inpaint.py +1356 -0
  270. diffusers/pipelines/pag/pipeline_pag_sd_xl.py +1345 -0
  271. diffusers/pipelines/pag/pipeline_pag_sd_xl_img2img.py +1544 -0
  272. diffusers/pipelines/pag/pipeline_pag_sd_xl_inpaint.py +1776 -0
  273. diffusers/pipelines/paint_by_example/pipeline_paint_by_example.py +17 -12
  274. diffusers/pipelines/pia/pipeline_pia.py +74 -164
  275. diffusers/pipelines/pipeline_flax_utils.py +5 -10
  276. diffusers/pipelines/pipeline_loading_utils.py +515 -53
  277. diffusers/pipelines/pipeline_utils.py +411 -222
  278. diffusers/pipelines/pixart_alpha/__init__.py +8 -1
  279. diffusers/pipelines/pixart_alpha/pipeline_pixart_alpha.py +76 -93
  280. diffusers/pipelines/pixart_alpha/pipeline_pixart_sigma.py +873 -0
  281. diffusers/pipelines/sana/__init__.py +47 -0
  282. diffusers/pipelines/sana/pipeline_output.py +21 -0
  283. diffusers/pipelines/sana/pipeline_sana.py +884 -0
  284. diffusers/pipelines/semantic_stable_diffusion/pipeline_semantic_stable_diffusion.py +27 -23
  285. diffusers/pipelines/shap_e/pipeline_shap_e.py +3 -3
  286. diffusers/pipelines/shap_e/pipeline_shap_e_img2img.py +14 -14
  287. diffusers/pipelines/shap_e/renderer.py +1 -1
  288. diffusers/pipelines/stable_audio/__init__.py +50 -0
  289. diffusers/pipelines/stable_audio/modeling_stable_audio.py +158 -0
  290. diffusers/pipelines/stable_audio/pipeline_stable_audio.py +756 -0
  291. diffusers/pipelines/stable_cascade/pipeline_stable_cascade.py +71 -25
  292. diffusers/pipelines/stable_cascade/pipeline_stable_cascade_combined.py +23 -19
  293. diffusers/pipelines/stable_cascade/pipeline_stable_cascade_prior.py +35 -34
  294. diffusers/pipelines/stable_diffusion/__init__.py +0 -1
  295. diffusers/pipelines/stable_diffusion/convert_from_ckpt.py +20 -11
  296. diffusers/pipelines/stable_diffusion/pipeline_flax_stable_diffusion.py +1 -1
  297. diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion.py +2 -2
  298. diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion_upscale.py +6 -6
  299. diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py +145 -79
  300. diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_depth2img.py +43 -28
  301. diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_image_variation.py +13 -8
  302. diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_img2img.py +100 -68
  303. diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint.py +109 -201
  304. diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_instruct_pix2pix.py +131 -32
  305. diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_latent_upscale.py +247 -87
  306. diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_upscale.py +30 -29
  307. diffusers/pipelines/stable_diffusion/pipeline_stable_unclip.py +35 -27
  308. diffusers/pipelines/stable_diffusion/pipeline_stable_unclip_img2img.py +49 -42
  309. diffusers/pipelines/stable_diffusion/safety_checker.py +2 -1
  310. diffusers/pipelines/stable_diffusion_3/__init__.py +54 -0
  311. diffusers/pipelines/stable_diffusion_3/pipeline_output.py +21 -0
  312. diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3.py +1140 -0
  313. diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3_img2img.py +1036 -0
  314. diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3_inpaint.py +1250 -0
  315. diffusers/pipelines/stable_diffusion_attend_and_excite/pipeline_stable_diffusion_attend_and_excite.py +29 -20
  316. diffusers/pipelines/stable_diffusion_diffedit/pipeline_stable_diffusion_diffedit.py +59 -58
  317. diffusers/pipelines/stable_diffusion_gligen/pipeline_stable_diffusion_gligen.py +31 -25
  318. diffusers/pipelines/stable_diffusion_gligen/pipeline_stable_diffusion_gligen_text_image.py +38 -22
  319. diffusers/pipelines/stable_diffusion_k_diffusion/pipeline_stable_diffusion_k_diffusion.py +30 -24
  320. diffusers/pipelines/stable_diffusion_k_diffusion/pipeline_stable_diffusion_xl_k_diffusion.py +24 -23
  321. diffusers/pipelines/stable_diffusion_ldm3d/pipeline_stable_diffusion_ldm3d.py +107 -67
  322. diffusers/pipelines/stable_diffusion_panorama/pipeline_stable_diffusion_panorama.py +316 -69
  323. diffusers/pipelines/stable_diffusion_safe/pipeline_stable_diffusion_safe.py +10 -5
  324. diffusers/pipelines/stable_diffusion_safe/safety_checker.py +1 -1
  325. diffusers/pipelines/stable_diffusion_sag/pipeline_stable_diffusion_sag.py +98 -30
  326. diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py +121 -83
  327. diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py +161 -105
  328. diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py +142 -218
  329. diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py +45 -29
  330. diffusers/pipelines/stable_diffusion_xl/watermark.py +9 -3
  331. diffusers/pipelines/stable_video_diffusion/pipeline_stable_video_diffusion.py +110 -57
  332. diffusers/pipelines/t2i_adapter/pipeline_stable_diffusion_adapter.py +69 -39
  333. diffusers/pipelines/t2i_adapter/pipeline_stable_diffusion_xl_adapter.py +105 -74
  334. diffusers/pipelines/text_to_video_synthesis/pipeline_output.py +3 -2
  335. diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_synth.py +29 -49
  336. diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_synth_img2img.py +32 -93
  337. diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_zero.py +37 -25
  338. diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_zero_sdxl.py +54 -40
  339. diffusers/pipelines/unclip/pipeline_unclip.py +6 -6
  340. diffusers/pipelines/unclip/pipeline_unclip_image_variation.py +6 -6
  341. diffusers/pipelines/unidiffuser/modeling_text_decoder.py +1 -1
  342. diffusers/pipelines/unidiffuser/modeling_uvit.py +12 -12
  343. diffusers/pipelines/unidiffuser/pipeline_unidiffuser.py +29 -28
  344. diffusers/pipelines/wuerstchen/modeling_paella_vq_model.py +5 -5
  345. diffusers/pipelines/wuerstchen/modeling_wuerstchen_common.py +5 -10
  346. diffusers/pipelines/wuerstchen/modeling_wuerstchen_prior.py +6 -8
  347. diffusers/pipelines/wuerstchen/pipeline_wuerstchen.py +4 -4
  348. diffusers/pipelines/wuerstchen/pipeline_wuerstchen_combined.py +12 -12
  349. diffusers/pipelines/wuerstchen/pipeline_wuerstchen_prior.py +15 -14
  350. diffusers/{models/dual_transformer_2d.py → quantizers/__init__.py} +2 -6
  351. diffusers/quantizers/auto.py +139 -0
  352. diffusers/quantizers/base.py +233 -0
  353. diffusers/quantizers/bitsandbytes/__init__.py +2 -0
  354. diffusers/quantizers/bitsandbytes/bnb_quantizer.py +561 -0
  355. diffusers/quantizers/bitsandbytes/utils.py +306 -0
  356. diffusers/quantizers/gguf/__init__.py +1 -0
  357. diffusers/quantizers/gguf/gguf_quantizer.py +159 -0
  358. diffusers/quantizers/gguf/utils.py +456 -0
  359. diffusers/quantizers/quantization_config.py +669 -0
  360. diffusers/quantizers/torchao/__init__.py +15 -0
  361. diffusers/quantizers/torchao/torchao_quantizer.py +292 -0
  362. diffusers/schedulers/__init__.py +12 -2
  363. diffusers/schedulers/deprecated/__init__.py +1 -1
  364. diffusers/schedulers/deprecated/scheduling_karras_ve.py +25 -25
  365. diffusers/schedulers/scheduling_amused.py +5 -5
  366. diffusers/schedulers/scheduling_consistency_decoder.py +11 -11
  367. diffusers/schedulers/scheduling_consistency_models.py +23 -25
  368. diffusers/schedulers/scheduling_cosine_dpmsolver_multistep.py +572 -0
  369. diffusers/schedulers/scheduling_ddim.py +27 -26
  370. diffusers/schedulers/scheduling_ddim_cogvideox.py +452 -0
  371. diffusers/schedulers/scheduling_ddim_flax.py +2 -1
  372. diffusers/schedulers/scheduling_ddim_inverse.py +16 -16
  373. diffusers/schedulers/scheduling_ddim_parallel.py +32 -31
  374. diffusers/schedulers/scheduling_ddpm.py +27 -30
  375. diffusers/schedulers/scheduling_ddpm_flax.py +7 -3
  376. diffusers/schedulers/scheduling_ddpm_parallel.py +33 -36
  377. diffusers/schedulers/scheduling_ddpm_wuerstchen.py +14 -14
  378. diffusers/schedulers/scheduling_deis_multistep.py +150 -50
  379. diffusers/schedulers/scheduling_dpm_cogvideox.py +489 -0
  380. diffusers/schedulers/scheduling_dpmsolver_multistep.py +221 -84
  381. diffusers/schedulers/scheduling_dpmsolver_multistep_flax.py +2 -2
  382. diffusers/schedulers/scheduling_dpmsolver_multistep_inverse.py +158 -52
  383. diffusers/schedulers/scheduling_dpmsolver_sde.py +153 -34
  384. diffusers/schedulers/scheduling_dpmsolver_singlestep.py +275 -86
  385. diffusers/schedulers/scheduling_edm_dpmsolver_multistep.py +81 -57
  386. diffusers/schedulers/scheduling_edm_euler.py +62 -39
  387. diffusers/schedulers/scheduling_euler_ancestral_discrete.py +30 -29
  388. diffusers/schedulers/scheduling_euler_discrete.py +255 -74
  389. diffusers/schedulers/scheduling_flow_match_euler_discrete.py +458 -0
  390. diffusers/schedulers/scheduling_flow_match_heun_discrete.py +320 -0
  391. diffusers/schedulers/scheduling_heun_discrete.py +174 -46
  392. diffusers/schedulers/scheduling_ipndm.py +9 -9
  393. diffusers/schedulers/scheduling_k_dpm_2_ancestral_discrete.py +138 -29
  394. diffusers/schedulers/scheduling_k_dpm_2_discrete.py +132 -26
  395. diffusers/schedulers/scheduling_karras_ve_flax.py +6 -6
  396. diffusers/schedulers/scheduling_lcm.py +23 -29
  397. diffusers/schedulers/scheduling_lms_discrete.py +105 -28
  398. diffusers/schedulers/scheduling_pndm.py +20 -20
  399. diffusers/schedulers/scheduling_repaint.py +21 -21
  400. diffusers/schedulers/scheduling_sasolver.py +157 -60
  401. diffusers/schedulers/scheduling_sde_ve.py +19 -19
  402. diffusers/schedulers/scheduling_tcd.py +41 -36
  403. diffusers/schedulers/scheduling_unclip.py +19 -16
  404. diffusers/schedulers/scheduling_unipc_multistep.py +243 -47
  405. diffusers/schedulers/scheduling_utils.py +12 -5
  406. diffusers/schedulers/scheduling_utils_flax.py +1 -3
  407. diffusers/schedulers/scheduling_vq_diffusion.py +10 -10
  408. diffusers/training_utils.py +214 -30
  409. diffusers/utils/__init__.py +17 -1
  410. diffusers/utils/constants.py +3 -0
  411. diffusers/utils/doc_utils.py +1 -0
  412. diffusers/utils/dummy_pt_objects.py +592 -7
  413. diffusers/utils/dummy_torch_and_torchsde_objects.py +15 -0
  414. diffusers/utils/dummy_torch_and_transformers_and_sentencepiece_objects.py +47 -0
  415. diffusers/utils/dummy_torch_and_transformers_objects.py +1001 -71
  416. diffusers/utils/dynamic_modules_utils.py +34 -29
  417. diffusers/utils/export_utils.py +50 -6
  418. diffusers/utils/hub_utils.py +131 -17
  419. diffusers/utils/import_utils.py +210 -8
  420. diffusers/utils/loading_utils.py +118 -5
  421. diffusers/utils/logging.py +4 -2
  422. diffusers/utils/peft_utils.py +37 -7
  423. diffusers/utils/state_dict_utils.py +13 -2
  424. diffusers/utils/testing_utils.py +193 -11
  425. diffusers/utils/torch_utils.py +4 -0
  426. diffusers/video_processor.py +113 -0
  427. {diffusers-0.27.1.dist-info → diffusers-0.32.2.dist-info}/METADATA +82 -91
  428. diffusers-0.32.2.dist-info/RECORD +550 -0
  429. {diffusers-0.27.1.dist-info → diffusers-0.32.2.dist-info}/WHEEL +1 -1
  430. diffusers/loaders/autoencoder.py +0 -146
  431. diffusers/loaders/controlnet.py +0 -136
  432. diffusers/loaders/lora.py +0 -1349
  433. diffusers/models/prior_transformer.py +0 -12
  434. diffusers/models/t5_film_transformer.py +0 -70
  435. diffusers/models/transformer_2d.py +0 -25
  436. diffusers/models/transformer_temporal.py +0 -34
  437. diffusers/models/unet_1d.py +0 -26
  438. diffusers/models/unet_1d_blocks.py +0 -203
  439. diffusers/models/unet_2d.py +0 -27
  440. diffusers/models/unet_2d_blocks.py +0 -375
  441. diffusers/models/unet_2d_condition.py +0 -25
  442. diffusers-0.27.1.dist-info/RECORD +0 -399
  443. {diffusers-0.27.1.dist-info → diffusers-0.32.2.dist-info}/LICENSE +0 -0
  444. {diffusers-0.27.1.dist-info → diffusers-0.32.2.dist-info}/entry_points.txt +0 -0
  445. {diffusers-0.27.1.dist-info → diffusers-0.32.2.dist-info}/top_level.txt +0 -0
@@ -18,10 +18,10 @@ from typing import Any, Callable, Dict, List, Optional, Union
18
18
 
19
19
  import PIL.Image
20
20
  import torch
21
- from transformers import CLIPFeatureExtractor, CLIPTextModel, CLIPTokenizer
21
+ from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer
22
22
 
23
23
  from ...image_processor import VaeImageProcessor
24
- from ...loaders import LoraLoaderMixin, TextualInversionLoaderMixin
24
+ from ...loaders import StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
25
25
  from ...models import AutoencoderKL, UNet2DConditionModel
26
26
  from ...models.attention import GatedSelfAttentionDense
27
27
  from ...models.lora import adjust_lora_scale_text_encoder
@@ -138,7 +138,7 @@ class StableDiffusionGLIGENPipeline(DiffusionPipeline, StableDiffusionMixin):
138
138
  unet: UNet2DConditionModel,
139
139
  scheduler: KarrasDiffusionSchedulers,
140
140
  safety_checker: StableDiffusionSafetyChecker,
141
- feature_extractor: CLIPFeatureExtractor,
141
+ feature_extractor: CLIPImageProcessor,
142
142
  requires_safety_checker: bool = True,
143
143
  ):
144
144
  super().__init__()
@@ -180,8 +180,8 @@ class StableDiffusionGLIGENPipeline(DiffusionPipeline, StableDiffusionMixin):
180
180
  num_images_per_prompt,
181
181
  do_classifier_free_guidance,
182
182
  negative_prompt=None,
183
- prompt_embeds: Optional[torch.FloatTensor] = None,
184
- negative_prompt_embeds: Optional[torch.FloatTensor] = None,
183
+ prompt_embeds: Optional[torch.Tensor] = None,
184
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
185
185
  lora_scale: Optional[float] = None,
186
186
  **kwargs,
187
187
  ):
@@ -213,8 +213,8 @@ class StableDiffusionGLIGENPipeline(DiffusionPipeline, StableDiffusionMixin):
213
213
  num_images_per_prompt,
214
214
  do_classifier_free_guidance,
215
215
  negative_prompt=None,
216
- prompt_embeds: Optional[torch.FloatTensor] = None,
217
- negative_prompt_embeds: Optional[torch.FloatTensor] = None,
216
+ prompt_embeds: Optional[torch.Tensor] = None,
217
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
218
218
  lora_scale: Optional[float] = None,
219
219
  clip_skip: Optional[int] = None,
220
220
  ):
@@ -234,10 +234,10 @@ class StableDiffusionGLIGENPipeline(DiffusionPipeline, StableDiffusionMixin):
234
234
  The prompt or prompts not to guide the image generation. If not defined, one has to pass
235
235
  `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
236
236
  less than `1`).
237
- prompt_embeds (`torch.FloatTensor`, *optional*):
237
+ prompt_embeds (`torch.Tensor`, *optional*):
238
238
  Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
239
239
  provided, text embeddings will be generated from `prompt` input argument.
240
- negative_prompt_embeds (`torch.FloatTensor`, *optional*):
240
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
241
241
  Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
242
242
  weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
243
243
  argument.
@@ -249,7 +249,7 @@ class StableDiffusionGLIGENPipeline(DiffusionPipeline, StableDiffusionMixin):
249
249
  """
250
250
  # set lora scale so that monkey patched LoRA
251
251
  # function of text encoder can correctly access it
252
- if lora_scale is not None and isinstance(self, LoraLoaderMixin):
252
+ if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
253
253
  self._lora_scale = lora_scale
254
254
 
255
255
  # dynamically adjust the LoRA scale
@@ -381,9 +381,10 @@ class StableDiffusionGLIGENPipeline(DiffusionPipeline, StableDiffusionMixin):
381
381
  negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
382
382
  negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
383
383
 
384
- if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
385
- # Retrieve the original scale by scaling back the LoRA layers
386
- unscale_lora_layers(self.text_encoder, lora_scale)
384
+ if self.text_encoder is not None:
385
+ if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
386
+ # Retrieve the original scale by scaling back the LoRA layers
387
+ unscale_lora_layers(self.text_encoder, lora_scale)
387
388
 
388
389
  return prompt_embeds, negative_prompt_embeds
389
390
 
@@ -469,14 +470,19 @@ class StableDiffusionGLIGENPipeline(DiffusionPipeline, StableDiffusionMixin):
469
470
  )
470
471
 
471
472
  if len(gligen_phrases) != len(gligen_boxes):
472
- ValueError(
473
+ raise ValueError(
473
474
  "length of `gligen_phrases` and `gligen_boxes` has to be same, but"
474
475
  f" got: `gligen_phrases` {len(gligen_phrases)} != `gligen_boxes` {len(gligen_boxes)}"
475
476
  )
476
477
 
477
478
  # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents
478
479
  def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):
479
- shape = (batch_size, num_channels_latents, height // self.vae_scale_factor, width // self.vae_scale_factor)
480
+ shape = (
481
+ batch_size,
482
+ num_channels_latents,
483
+ int(height) // self.vae_scale_factor,
484
+ int(width) // self.vae_scale_factor,
485
+ )
480
486
  if isinstance(generator, list) and len(generator) != batch_size:
481
487
  raise ValueError(
482
488
  f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
@@ -536,12 +542,12 @@ class StableDiffusionGLIGENPipeline(DiffusionPipeline, StableDiffusionMixin):
536
542
  num_images_per_prompt: Optional[int] = 1,
537
543
  eta: float = 0.0,
538
544
  generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
539
- latents: Optional[torch.FloatTensor] = None,
540
- prompt_embeds: Optional[torch.FloatTensor] = None,
541
- negative_prompt_embeds: Optional[torch.FloatTensor] = None,
545
+ latents: Optional[torch.Tensor] = None,
546
+ prompt_embeds: Optional[torch.Tensor] = None,
547
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
542
548
  output_type: Optional[str] = "pil",
543
549
  return_dict: bool = True,
544
- callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
550
+ callback: Optional[Callable[[int, int, torch.Tensor], None]] = None,
545
551
  callback_steps: int = 1,
546
552
  cross_attention_kwargs: Optional[Dict[str, Any]] = None,
547
553
  clip_skip: Optional[int] = None,
@@ -587,14 +593,14 @@ class StableDiffusionGLIGENPipeline(DiffusionPipeline, StableDiffusionMixin):
587
593
  generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
588
594
  A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
589
595
  generation deterministic.
590
- latents (`torch.FloatTensor`, *optional*):
596
+ latents (`torch.Tensor`, *optional*):
591
597
  Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
592
598
  generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
593
599
  tensor is generated by sampling using the supplied random `generator`.
594
- prompt_embeds (`torch.FloatTensor`, *optional*):
600
+ prompt_embeds (`torch.Tensor`, *optional*):
595
601
  Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
596
602
  provided, text embeddings are generated from the `prompt` input argument.
597
- negative_prompt_embeds (`torch.FloatTensor`, *optional*):
603
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
598
604
  Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
599
605
  not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
600
606
  output_type (`str`, *optional*, defaults to `"pil"`):
@@ -604,7 +610,7 @@ class StableDiffusionGLIGENPipeline(DiffusionPipeline, StableDiffusionMixin):
604
610
  plain tuple.
605
611
  callback (`Callable`, *optional*):
606
612
  A function that calls every `callback_steps` steps during inference. The function is called with the
607
- following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
613
+ following arguments: `callback(step: int, timestep: int, latents: torch.Tensor)`.
608
614
  callback_steps (`int`, *optional*, defaults to 1):
609
615
  The frequency at which the `callback` function is called. If not specified, the callback is called at
610
616
  every step.
@@ -680,7 +686,7 @@ class StableDiffusionGLIGENPipeline(DiffusionPipeline, StableDiffusionMixin):
680
686
  timesteps = self.scheduler.timesteps
681
687
 
682
688
  # 5. Prepare latent variables
683
- num_channels_latents = self.unet.in_channels
689
+ num_channels_latents = self.unet.config.in_channels
684
690
  latents = self.prepare_latents(
685
691
  batch_size * num_images_per_prompt,
686
692
  num_channels_latents,
@@ -713,7 +719,7 @@ class StableDiffusionGLIGENPipeline(DiffusionPipeline, StableDiffusionMixin):
713
719
  boxes = torch.zeros(max_objs, 4, device=device, dtype=self.text_encoder.dtype)
714
720
  boxes[:n_objs] = torch.tensor(gligen_boxes)
715
721
  text_embeddings = torch.zeros(
716
- max_objs, self.unet.cross_attention_dim, device=device, dtype=self.text_encoder.dtype
722
+ max_objs, self.unet.config.cross_attention_dim, device=device, dtype=self.text_encoder.dtype
717
723
  )
718
724
  text_embeddings[:n_objs] = _text_embeddings
719
725
  # Generate a mask for each object that is entity described by phrases
@@ -19,7 +19,7 @@ from typing import Any, Callable, Dict, List, Optional, Union
19
19
  import PIL.Image
20
20
  import torch
21
21
  from transformers import (
22
- CLIPFeatureExtractor,
22
+ CLIPImageProcessor,
23
23
  CLIPProcessor,
24
24
  CLIPTextModel,
25
25
  CLIPTokenizer,
@@ -27,7 +27,7 @@ from transformers import (
27
27
  )
28
28
 
29
29
  from ...image_processor import VaeImageProcessor
30
- from ...loaders import LoraLoaderMixin, TextualInversionLoaderMixin
30
+ from ...loaders import StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
31
31
  from ...models import AutoencoderKL, UNet2DConditionModel
32
32
  from ...models.attention import GatedSelfAttentionDense
33
33
  from ...models.lora import adjust_lora_scale_text_encoder
@@ -193,7 +193,7 @@ class StableDiffusionGLIGENTextImagePipeline(DiffusionPipeline, StableDiffusionM
193
193
  unet: UNet2DConditionModel,
194
194
  scheduler: KarrasDiffusionSchedulers,
195
195
  safety_checker: StableDiffusionSafetyChecker,
196
- feature_extractor: CLIPFeatureExtractor,
196
+ feature_extractor: CLIPImageProcessor,
197
197
  requires_safety_checker: bool = True,
198
198
  ):
199
199
  super().__init__()
@@ -238,8 +238,8 @@ class StableDiffusionGLIGENTextImagePipeline(DiffusionPipeline, StableDiffusionM
238
238
  num_images_per_prompt,
239
239
  do_classifier_free_guidance,
240
240
  negative_prompt=None,
241
- prompt_embeds: Optional[torch.FloatTensor] = None,
242
- negative_prompt_embeds: Optional[torch.FloatTensor] = None,
241
+ prompt_embeds: Optional[torch.Tensor] = None,
242
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
243
243
  lora_scale: Optional[float] = None,
244
244
  clip_skip: Optional[int] = None,
245
245
  ):
@@ -259,10 +259,10 @@ class StableDiffusionGLIGENTextImagePipeline(DiffusionPipeline, StableDiffusionM
259
259
  The prompt or prompts not to guide the image generation. If not defined, one has to pass
260
260
  `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
261
261
  less than `1`).
262
- prompt_embeds (`torch.FloatTensor`, *optional*):
262
+ prompt_embeds (`torch.Tensor`, *optional*):
263
263
  Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
264
264
  provided, text embeddings will be generated from `prompt` input argument.
265
- negative_prompt_embeds (`torch.FloatTensor`, *optional*):
265
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
266
266
  Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
267
267
  weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
268
268
  argument.
@@ -274,7 +274,7 @@ class StableDiffusionGLIGENTextImagePipeline(DiffusionPipeline, StableDiffusionM
274
274
  """
275
275
  # set lora scale so that monkey patched LoRA
276
276
  # function of text encoder can correctly access it
277
- if lora_scale is not None and isinstance(self, LoraLoaderMixin):
277
+ if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
278
278
  self._lora_scale = lora_scale
279
279
 
280
280
  # dynamically adjust the LoRA scale
@@ -406,9 +406,10 @@ class StableDiffusionGLIGENTextImagePipeline(DiffusionPipeline, StableDiffusionM
406
406
  negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
407
407
  negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
408
408
 
409
- if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
410
- # Retrieve the original scale by scaling back the LoRA layers
411
- unscale_lora_layers(self.text_encoder, lora_scale)
409
+ if self.text_encoder is not None:
410
+ if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
411
+ # Retrieve the original scale by scaling back the LoRA layers
412
+ unscale_lora_layers(self.text_encoder, lora_scale)
412
413
 
413
414
  return prompt_embeds, negative_prompt_embeds
414
415
 
@@ -445,13 +446,14 @@ class StableDiffusionGLIGENTextImagePipeline(DiffusionPipeline, StableDiffusionM
445
446
  extra_step_kwargs["generator"] = generator
446
447
  return extra_step_kwargs
447
448
 
448
- # Copied from diffusers.pipelines.stable_diffusion_k_diffusion.pipeline_stable_diffusion_k_diffusion.StableDiffusionKDiffusionPipeline.check_inputs
449
449
  def check_inputs(
450
450
  self,
451
451
  prompt,
452
452
  height,
453
453
  width,
454
454
  callback_steps,
455
+ gligen_images,
456
+ gligen_phrases,
455
457
  negative_prompt=None,
456
458
  prompt_embeds=None,
457
459
  negative_prompt_embeds=None,
@@ -498,9 +500,21 @@ class StableDiffusionGLIGENTextImagePipeline(DiffusionPipeline, StableDiffusionM
498
500
  f" {negative_prompt_embeds.shape}."
499
501
  )
500
502
 
503
+ if gligen_images is not None and gligen_phrases is not None:
504
+ if len(gligen_images) != len(gligen_phrases):
505
+ raise ValueError(
506
+ "`gligen_images` and `gligen_phrases` must have the same length when both are provided, but"
507
+ f" got: `gligen_images` with length {len(gligen_images)} != `gligen_phrases` with length {len(gligen_phrases)}."
508
+ )
509
+
501
510
  # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents
502
511
  def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):
503
- shape = (batch_size, num_channels_latents, height // self.vae_scale_factor, width // self.vae_scale_factor)
512
+ shape = (
513
+ batch_size,
514
+ num_channels_latents,
515
+ int(height) // self.vae_scale_factor,
516
+ int(width) // self.vae_scale_factor,
517
+ )
504
518
  if isinstance(generator, list) and len(generator) != batch_size:
505
519
  raise ValueError(
506
520
  f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
@@ -700,12 +714,12 @@ class StableDiffusionGLIGENTextImagePipeline(DiffusionPipeline, StableDiffusionM
700
714
  num_images_per_prompt: Optional[int] = 1,
701
715
  eta: float = 0.0,
702
716
  generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
703
- latents: Optional[torch.FloatTensor] = None,
704
- prompt_embeds: Optional[torch.FloatTensor] = None,
705
- negative_prompt_embeds: Optional[torch.FloatTensor] = None,
717
+ latents: Optional[torch.Tensor] = None,
718
+ prompt_embeds: Optional[torch.Tensor] = None,
719
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
706
720
  output_type: Optional[str] = "pil",
707
721
  return_dict: bool = True,
708
- callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
722
+ callback: Optional[Callable[[int, int, torch.Tensor], None]] = None,
709
723
  callback_steps: int = 1,
710
724
  cross_attention_kwargs: Optional[Dict[str, Any]] = None,
711
725
  gligen_normalize_constant: float = 28.7,
@@ -759,14 +773,14 @@ class StableDiffusionGLIGENTextImagePipeline(DiffusionPipeline, StableDiffusionM
759
773
  generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
760
774
  A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
761
775
  generation deterministic.
762
- latents (`torch.FloatTensor`, *optional*):
776
+ latents (`torch.Tensor`, *optional*):
763
777
  Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
764
778
  generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
765
779
  tensor is generated by sampling using the supplied random `generator`.
766
- prompt_embeds (`torch.FloatTensor`, *optional*):
780
+ prompt_embeds (`torch.Tensor`, *optional*):
767
781
  Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
768
782
  provided, text embeddings are generated from the `prompt` input argument.
769
- negative_prompt_embeds (`torch.FloatTensor`, *optional*):
783
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
770
784
  Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
771
785
  not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
772
786
  output_type (`str`, *optional*, defaults to `"pil"`):
@@ -776,7 +790,7 @@ class StableDiffusionGLIGENTextImagePipeline(DiffusionPipeline, StableDiffusionM
776
790
  plain tuple.
777
791
  callback (`Callable`, *optional*):
778
792
  A function that calls every `callback_steps` steps during inference. The function is called with the
779
- following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
793
+ following arguments: `callback(step: int, timestep: int, latents: torch.Tensor)`.
780
794
  callback_steps (`int`, *optional*, defaults to 1):
781
795
  The frequency at which the `callback` function is called. If not specified, the callback is called at
782
796
  every step.
@@ -808,6 +822,8 @@ class StableDiffusionGLIGENTextImagePipeline(DiffusionPipeline, StableDiffusionM
808
822
  height,
809
823
  width,
810
824
  callback_steps,
825
+ gligen_images,
826
+ gligen_phrases,
811
827
  negative_prompt,
812
828
  prompt_embeds,
813
829
  negative_prompt_embeds,
@@ -847,7 +863,7 @@ class StableDiffusionGLIGENTextImagePipeline(DiffusionPipeline, StableDiffusionM
847
863
  timesteps = self.scheduler.timesteps
848
864
 
849
865
  # 5. Prepare latent variables
850
- num_channels_latents = self.unet.in_channels
866
+ num_channels_latents = self.unet.config.in_channels
851
867
  latents = self.prepare_latents(
852
868
  batch_size * num_images_per_prompt,
853
869
  num_channels_latents,
@@ -21,7 +21,7 @@ from k_diffusion.external import CompVisDenoiser, CompVisVDenoiser
21
21
  from k_diffusion.sampling import BrownianTreeNoiseSampler, get_sigmas_karras
22
22
 
23
23
  from ...image_processor import VaeImageProcessor
24
- from ...loaders import LoraLoaderMixin, TextualInversionLoaderMixin
24
+ from ...loaders import StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
25
25
  from ...models.lora import adjust_lora_scale_text_encoder
26
26
  from ...schedulers import LMSDiscreteScheduler
27
27
  from ...utils import USE_PEFT_BACKEND, deprecate, logging, scale_lora_layers, unscale_lora_layers
@@ -48,7 +48,7 @@ class ModelWrapper:
48
48
 
49
49
 
50
50
  class StableDiffusionKDiffusionPipeline(
51
- DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, LoraLoaderMixin
51
+ DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, StableDiffusionLoraLoaderMixin
52
52
  ):
53
53
  r"""
54
54
  Pipeline for text-to-image generation using Stable Diffusion.
@@ -58,8 +58,8 @@ class StableDiffusionKDiffusionPipeline(
58
58
 
59
59
  The pipeline also inherits the following loading methods:
60
60
  - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings
61
- - [`~loaders.LoraLoaderMixin.load_lora_weights`] for loading LoRA weights
62
- - [`~loaders.LoraLoaderMixin.save_lora_weights`] for saving LoRA weights
61
+ - [`~loaders.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
62
+ - [`~loaders.StableDiffusionLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
63
63
 
64
64
  <Tip warning={true}>
65
65
 
@@ -154,8 +154,8 @@ class StableDiffusionKDiffusionPipeline(
154
154
  num_images_per_prompt,
155
155
  do_classifier_free_guidance,
156
156
  negative_prompt=None,
157
- prompt_embeds: Optional[torch.FloatTensor] = None,
158
- negative_prompt_embeds: Optional[torch.FloatTensor] = None,
157
+ prompt_embeds: Optional[torch.Tensor] = None,
158
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
159
159
  lora_scale: Optional[float] = None,
160
160
  **kwargs,
161
161
  ):
@@ -187,8 +187,8 @@ class StableDiffusionKDiffusionPipeline(
187
187
  num_images_per_prompt,
188
188
  do_classifier_free_guidance,
189
189
  negative_prompt=None,
190
- prompt_embeds: Optional[torch.FloatTensor] = None,
191
- negative_prompt_embeds: Optional[torch.FloatTensor] = None,
190
+ prompt_embeds: Optional[torch.Tensor] = None,
191
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
192
192
  lora_scale: Optional[float] = None,
193
193
  clip_skip: Optional[int] = None,
194
194
  ):
@@ -208,10 +208,10 @@ class StableDiffusionKDiffusionPipeline(
208
208
  The prompt or prompts not to guide the image generation. If not defined, one has to pass
209
209
  `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
210
210
  less than `1`).
211
- prompt_embeds (`torch.FloatTensor`, *optional*):
211
+ prompt_embeds (`torch.Tensor`, *optional*):
212
212
  Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
213
213
  provided, text embeddings will be generated from `prompt` input argument.
214
- negative_prompt_embeds (`torch.FloatTensor`, *optional*):
214
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
215
215
  Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
216
216
  weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
217
217
  argument.
@@ -223,7 +223,7 @@ class StableDiffusionKDiffusionPipeline(
223
223
  """
224
224
  # set lora scale so that monkey patched LoRA
225
225
  # function of text encoder can correctly access it
226
- if lora_scale is not None and isinstance(self, LoraLoaderMixin):
226
+ if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
227
227
  self._lora_scale = lora_scale
228
228
 
229
229
  # dynamically adjust the LoRA scale
@@ -355,9 +355,10 @@ class StableDiffusionKDiffusionPipeline(
355
355
  negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
356
356
  negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
357
357
 
358
- if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
359
- # Retrieve the original scale by scaling back the LoRA layers
360
- unscale_lora_layers(self.text_encoder, lora_scale)
358
+ if self.text_encoder is not None:
359
+ if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
360
+ # Retrieve the original scale by scaling back the LoRA layers
361
+ unscale_lora_layers(self.text_encoder, lora_scale)
361
362
 
362
363
  return prompt_embeds, negative_prompt_embeds
363
364
 
@@ -441,7 +442,12 @@ class StableDiffusionKDiffusionPipeline(
441
442
  )
442
443
 
443
444
  def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):
444
- shape = (batch_size, num_channels_latents, height // self.vae_scale_factor, width // self.vae_scale_factor)
445
+ shape = (
446
+ batch_size,
447
+ num_channels_latents,
448
+ int(height) // self.vae_scale_factor,
449
+ int(width) // self.vae_scale_factor,
450
+ )
445
451
  if latents is None:
446
452
  latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
447
453
  else:
@@ -464,12 +470,12 @@ class StableDiffusionKDiffusionPipeline(
464
470
  num_images_per_prompt: Optional[int] = 1,
465
471
  eta: float = 0.0,
466
472
  generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
467
- latents: Optional[torch.FloatTensor] = None,
468
- prompt_embeds: Optional[torch.FloatTensor] = None,
469
- negative_prompt_embeds: Optional[torch.FloatTensor] = None,
473
+ latents: Optional[torch.Tensor] = None,
474
+ prompt_embeds: Optional[torch.Tensor] = None,
475
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
470
476
  output_type: Optional[str] = "pil",
471
477
  return_dict: bool = True,
472
- callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
478
+ callback: Optional[Callable[[int, int, torch.Tensor], None]] = None,
473
479
  callback_steps: int = 1,
474
480
  use_karras_sigmas: Optional[bool] = False,
475
481
  noise_sampler_seed: Optional[int] = None,
@@ -507,14 +513,14 @@ class StableDiffusionKDiffusionPipeline(
507
513
  generator (`torch.Generator`, *optional*):
508
514
  One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
509
515
  to make generation deterministic.
510
- latents (`torch.FloatTensor`, *optional*):
516
+ latents (`torch.Tensor`, *optional*):
511
517
  Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
512
518
  generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
513
519
  tensor will ge generated by sampling using the supplied random `generator`.
514
- prompt_embeds (`torch.FloatTensor`, *optional*):
520
+ prompt_embeds (`torch.Tensor`, *optional*):
515
521
  Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
516
522
  provided, text embeddings will be generated from `prompt` input argument.
517
- negative_prompt_embeds (`torch.FloatTensor`, *optional*):
523
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
518
524
  Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
519
525
  weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
520
526
  argument.
@@ -526,7 +532,7 @@ class StableDiffusionKDiffusionPipeline(
526
532
  plain tuple.
527
533
  callback (`Callable`, *optional*):
528
534
  A function that will be called every `callback_steps` steps during inference. The function will be
529
- called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
535
+ called with the following arguments: `callback(step: int, timestep: int, latents: torch.Tensor)`.
530
536
  callback_steps (`int`, *optional*, defaults to 1):
531
537
  The frequency at which the `callback` function will be called. If not specified, the callback will be
532
538
  called at every step.
@@ -596,9 +602,9 @@ class StableDiffusionKDiffusionPipeline(
596
602
  sigma_min: float = self.k_diffusion_model.sigmas[0].item()
597
603
  sigma_max: float = self.k_diffusion_model.sigmas[-1].item()
598
604
  sigmas = get_sigmas_karras(n=num_inference_steps, sigma_min=sigma_min, sigma_max=sigma_max)
599
- sigmas = sigmas.to(device)
600
605
  else:
601
606
  sigmas = self.scheduler.sigmas
607
+ sigmas = sigmas.to(device)
602
608
  sigmas = sigmas.to(prompt_embeds.dtype)
603
609
 
604
610
  # 6. Prepare latent variables
@@ -36,8 +36,6 @@ from ...models import AutoencoderKL, UNet2DConditionModel
36
36
  from ...models.attention_processor import (
37
37
  AttnProcessor2_0,
38
38
  FusedAttnProcessor2_0,
39
- LoRAAttnProcessor2_0,
40
- LoRAXFormersAttnProcessor,
41
39
  XFormersAttnProcessor,
42
40
  )
43
41
  from ...models.lora import adjust_lora_scale_text_encoder
@@ -207,10 +205,10 @@ class StableDiffusionXLKDiffusionPipeline(
207
205
  do_classifier_free_guidance: bool = True,
208
206
  negative_prompt: Optional[str] = None,
209
207
  negative_prompt_2: Optional[str] = None,
210
- prompt_embeds: Optional[torch.FloatTensor] = None,
211
- negative_prompt_embeds: Optional[torch.FloatTensor] = None,
212
- pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
213
- negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
208
+ prompt_embeds: Optional[torch.Tensor] = None,
209
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
210
+ pooled_prompt_embeds: Optional[torch.Tensor] = None,
211
+ negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
214
212
  lora_scale: Optional[float] = None,
215
213
  clip_skip: Optional[int] = None,
216
214
  ):
@@ -236,17 +234,17 @@ class StableDiffusionXLKDiffusionPipeline(
236
234
  negative_prompt_2 (`str` or `List[str]`, *optional*):
237
235
  The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
238
236
  `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
239
- prompt_embeds (`torch.FloatTensor`, *optional*):
237
+ prompt_embeds (`torch.Tensor`, *optional*):
240
238
  Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
241
239
  provided, text embeddings will be generated from `prompt` input argument.
242
- negative_prompt_embeds (`torch.FloatTensor`, *optional*):
240
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
243
241
  Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
244
242
  weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
245
243
  argument.
246
- pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
244
+ pooled_prompt_embeds (`torch.Tensor`, *optional*):
247
245
  Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
248
246
  If not provided, pooled text embeddings will be generated from `prompt` input argument.
249
- negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
247
+ negative_pooled_prompt_embeds (`torch.Tensor`, *optional*):
250
248
  Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
251
249
  weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
252
250
  input argument.
@@ -497,7 +495,12 @@ class StableDiffusionXLKDiffusionPipeline(
497
495
  )
498
496
 
499
497
  def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):
500
- shape = (batch_size, num_channels_latents, height // self.vae_scale_factor, width // self.vae_scale_factor)
498
+ shape = (
499
+ batch_size,
500
+ num_channels_latents,
501
+ int(height) // self.vae_scale_factor,
502
+ int(width) // self.vae_scale_factor,
503
+ )
501
504
  if isinstance(generator, list) and len(generator) != batch_size:
502
505
  raise ValueError(
503
506
  f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
@@ -538,8 +541,6 @@ class StableDiffusionXLKDiffusionPipeline(
538
541
  (
539
542
  AttnProcessor2_0,
540
543
  XFormersAttnProcessor,
541
- LoRAXFormersAttnProcessor,
542
- LoRAAttnProcessor2_0,
543
544
  FusedAttnProcessor2_0,
544
545
  ),
545
546
  )
@@ -579,11 +580,11 @@ class StableDiffusionXLKDiffusionPipeline(
579
580
  negative_prompt_2: Optional[Union[str, List[str]]] = None,
580
581
  num_images_per_prompt: Optional[int] = 1,
581
582
  generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
582
- latents: Optional[torch.FloatTensor] = None,
583
- prompt_embeds: Optional[torch.FloatTensor] = None,
584
- negative_prompt_embeds: Optional[torch.FloatTensor] = None,
585
- pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
586
- negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
583
+ latents: Optional[torch.Tensor] = None,
584
+ prompt_embeds: Optional[torch.Tensor] = None,
585
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
586
+ pooled_prompt_embeds: Optional[torch.Tensor] = None,
587
+ negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
587
588
  output_type: Optional[str] = "pil",
588
589
  return_dict: bool = True,
589
590
  original_size: Optional[Tuple[int, int]] = None,
@@ -637,21 +638,21 @@ class StableDiffusionXLKDiffusionPipeline(
637
638
  generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
638
639
  One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
639
640
  to make generation deterministic.
640
- latents (`torch.FloatTensor`, *optional*):
641
+ latents (`torch.Tensor`, *optional*):
641
642
  Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
642
643
  generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
643
644
  tensor will ge generated by sampling using the supplied random `generator`.
644
- prompt_embeds (`torch.FloatTensor`, *optional*):
645
+ prompt_embeds (`torch.Tensor`, *optional*):
645
646
  Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
646
647
  provided, text embeddings will be generated from `prompt` input argument.
647
- negative_prompt_embeds (`torch.FloatTensor`, *optional*):
648
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
648
649
  Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
649
650
  weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
650
651
  argument.
651
- pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
652
+ pooled_prompt_embeds (`torch.Tensor`, *optional*):
652
653
  Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
653
654
  If not provided, pooled text embeddings will be generated from `prompt` input argument.
654
- negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
655
+ negative_pooled_prompt_embeds (`torch.Tensor`, *optional*):
655
656
  Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
656
657
  weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
657
658
  input argument.