diffusers 0.27.0__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 +50 -53
  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-0.27.0.dist-info → diffusers-0.32.2.dist-info}/METADATA +82 -91
  428. diffusers-0.32.2.dist-info/RECORD +550 -0
  429. {diffusers-0.27.0.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-0.27.0.dist-info/RECORD +0 -399
  443. {diffusers-0.27.0.dist-info → diffusers-0.32.2.dist-info}/LICENSE +0 -0
  444. {diffusers-0.27.0.dist-info → diffusers-0.32.2.dist-info}/entry_points.txt +0 -0
  445. {diffusers-0.27.0.dist-info → diffusers-0.32.2.dist-info}/top_level.txt +0 -0
@@ -38,8 +38,6 @@ from ...models.attention_processor import (
38
38
  Attention,
39
39
  AttnProcessor,
40
40
  AttnProcessor2_0,
41
- LoRAAttnProcessor2_0,
42
- LoRAXFormersAttnProcessor,
43
41
  XFormersAttnProcessor,
44
42
  )
45
43
  from ...models.lora import adjust_lora_scale_text_encoder
@@ -85,25 +83,23 @@ EXAMPLE_DOC_STRING = """
85
83
  ... )
86
84
  >>> pipe = pipe.to("cuda")
87
85
 
86
+
88
87
  >>> def download_image(url):
89
88
  ... response = requests.get(url)
90
89
  ... return PIL.Image.open(BytesIO(response.content)).convert("RGB")
91
90
 
91
+
92
92
  >>> img_url = "https://www.aiml.informatik.tu-darmstadt.de/people/mbrack/tennis.jpg"
93
93
  >>> image = download_image(img_url)
94
94
 
95
- >>> _ = pipe.invert(
96
- ... image = image,
97
- ... num_inversion_steps=50,
98
- ... skip=0.2
99
- ... )
95
+ >>> _ = pipe.invert(image=image, num_inversion_steps=50, skip=0.2)
100
96
 
101
97
  >>> edited_image = pipe(
102
- ... editing_prompt=["tennis ball","tomato"],
103
- ... reverse_editing_direction=[True,False],
104
- ... edit_guidance_scale=[5.0,10.0],
105
- ... edit_threshold=[0.9,0.85],
106
- ).images[0]
98
+ ... editing_prompt=["tennis ball", "tomato"],
99
+ ... reverse_editing_direction=[True, False],
100
+ ... edit_guidance_scale=[5.0, 10.0],
101
+ ... edit_threshold=[0.9, 0.85],
102
+ ... ).images[0]
107
103
  ```
108
104
  """
109
105
 
@@ -292,9 +288,9 @@ class LEditsPPPipelineStableDiffusionXL(
292
288
  """
293
289
  Pipeline for textual image editing using LEDits++ with Stable Diffusion XL.
294
290
 
295
- This model inherits from [`DiffusionPipeline`] and builds on the [`StableDiffusionXLPipeline`]. Check the superclass
296
- documentation for the generic methods implemented for all pipelines (downloading, saving, running on a particular
297
- device, etc.).
291
+ This model inherits from [`DiffusionPipeline`] and builds on the [`StableDiffusionXLPipeline`]. Check the
292
+ superclass documentation for the generic methods implemented for all pipelines (downloading, saving, running on a
293
+ particular device, etc.).
298
294
 
299
295
  In addition the pipeline inherits the following loading methods:
300
296
  - *LoRA*: [`LEditsPPPipelineStableDiffusionXL.load_lora_weights`]
@@ -325,8 +321,8 @@ class LEditsPPPipelineStableDiffusionXL(
325
321
  unet ([`UNet2DConditionModel`]): Conditional U-Net architecture to denoise the encoded image latents.
326
322
  scheduler ([`DPMSolverMultistepScheduler`] or [`DDIMScheduler`]):
327
323
  A scheduler to be used in combination with `unet` to denoise the encoded image latens. Can be one of
328
- [`DPMSolverMultistepScheduler`] or [`DDIMScheduler`]. If any other scheduler is passed it will automatically
329
- be set to [`DPMSolverMultistepScheduler`].
324
+ [`DPMSolverMultistepScheduler`] or [`DDIMScheduler`]. If any other scheduler is passed it will
325
+ automatically be set to [`DPMSolverMultistepScheduler`].
330
326
  force_zeros_for_empty_prompt (`bool`, *optional*, defaults to `"True"`):
331
327
  Whether the negative prompt embeddings shall be forced to always be set to 0. Also see the config of
332
328
  `stabilityai/stable-diffusion-xl-base-1-0`.
@@ -411,14 +407,14 @@ class LEditsPPPipelineStableDiffusionXL(
411
407
  num_images_per_prompt: int = 1,
412
408
  negative_prompt: Optional[str] = None,
413
409
  negative_prompt_2: Optional[str] = None,
414
- negative_prompt_embeds: Optional[torch.FloatTensor] = None,
415
- negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
410
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
411
+ negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
416
412
  lora_scale: Optional[float] = None,
417
413
  clip_skip: Optional[int] = None,
418
414
  enable_edit_guidance: bool = True,
419
415
  editing_prompt: Optional[str] = None,
420
- editing_prompt_embeds: Optional[torch.FloatTensor] = None,
421
- editing_pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
416
+ editing_prompt_embeds: Optional[torch.Tensor] = None,
417
+ editing_pooled_prompt_embeds: Optional[torch.Tensor] = None,
422
418
  ) -> object:
423
419
  r"""
424
420
  Encodes the prompt into text encoder hidden states.
@@ -434,11 +430,11 @@ class LEditsPPPipelineStableDiffusionXL(
434
430
  negative_prompt_2 (`str` or `List[str]`, *optional*):
435
431
  The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
436
432
  `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
437
- negative_prompt_embeds (`torch.FloatTensor`, *optional*):
433
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
438
434
  Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
439
435
  weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
440
436
  argument.
441
- negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
437
+ negative_pooled_prompt_embeds (`torch.Tensor`, *optional*):
442
438
  Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
443
439
  weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
444
440
  input argument.
@@ -452,11 +448,11 @@ class LEditsPPPipelineStableDiffusionXL(
452
448
  editing_prompt (`str` or `List[str]`, *optional*):
453
449
  Editing prompt(s) to be encoded. If not defined and 'enable_edit_guidance' is True, one has to pass
454
450
  `editing_prompt_embeds` instead.
455
- editing_prompt_embeds (`torch.FloatTensor`, *optional*):
456
- Pre-generated edit text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
457
- weighting. If not provided and 'enable_edit_guidance' is True, editing_prompt_embeds will be generated from `editing_prompt` input
458
- argument.
459
- editing_pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
451
+ editing_prompt_embeds (`torch.Tensor`, *optional*):
452
+ Pre-generated edit text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
453
+ If not provided and 'enable_edit_guidance' is True, editing_prompt_embeds will be generated from
454
+ `editing_prompt` input argument.
455
+ editing_pooled_prompt_embeds (`torch.Tensor`, *optional*):
460
456
  Pre-generated edit pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
461
457
  weighting. If not provided, pooled editing_pooled_prompt_embeds will be generated from `editing_prompt`
462
458
  input argument.
@@ -701,8 +697,6 @@ class LEditsPPPipelineStableDiffusionXL(
701
697
  (
702
698
  AttnProcessor2_0,
703
699
  XFormersAttnProcessor,
704
- LoRAXFormersAttnProcessor,
705
- LoRAAttnProcessor2_0,
706
700
  ),
707
701
  )
708
702
  # if xformers or torch_2_0 is used attention block does not need
@@ -713,20 +707,22 @@ class LEditsPPPipelineStableDiffusionXL(
713
707
  self.vae.decoder.mid_block.to(dtype)
714
708
 
715
709
  # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
716
- def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32):
710
+ def get_guidance_scale_embedding(
711
+ self, w: torch.Tensor, embedding_dim: int = 512, dtype: torch.dtype = torch.float32
712
+ ) -> torch.Tensor:
717
713
  """
718
714
  See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
719
715
 
720
716
  Args:
721
- timesteps (`torch.Tensor`):
722
- generate embedding vectors at these timesteps
717
+ w (`torch.Tensor`):
718
+ Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings.
723
719
  embedding_dim (`int`, *optional*, defaults to 512):
724
- dimension of the embeddings to generate
725
- dtype:
726
- data type of the generated embeddings
720
+ Dimension of the embeddings to generate.
721
+ dtype (`torch.dtype`, *optional*, defaults to `torch.float32`):
722
+ Data type of the generated embeddings.
727
723
 
728
724
  Returns:
729
- `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
725
+ `torch.Tensor`: Embedding vectors with shape `(len(w), embedding_dim)`.
730
726
  """
731
727
  assert len(w.shape) == 1
732
728
  w = w * 1000.0
@@ -804,8 +800,8 @@ class LEditsPPPipelineStableDiffusionXL(
804
800
  denoising_end: Optional[float] = None,
805
801
  negative_prompt: Optional[Union[str, List[str]]] = None,
806
802
  negative_prompt_2: Optional[Union[str, List[str]]] = None,
807
- negative_prompt_embeds: Optional[torch.FloatTensor] = None,
808
- negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
803
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
804
+ negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
809
805
  ip_adapter_image: Optional[PipelineImageInput] = None,
810
806
  output_type: Optional[str] = "pil",
811
807
  return_dict: bool = True,
@@ -824,7 +820,7 @@ class LEditsPPPipelineStableDiffusionXL(
824
820
  sem_guidance: Optional[List[torch.Tensor]] = None,
825
821
  use_cross_attn_mask: bool = False,
826
822
  use_intersect_mask: bool = False,
827
- user_mask: Optional[torch.FloatTensor] = None,
823
+ user_mask: Optional[torch.Tensor] = None,
828
824
  attn_store_steps: Optional[List[int]] = [],
829
825
  store_averaged_over_steps: bool = True,
830
826
  clip_skip: Optional[int] = None,
@@ -833,8 +829,9 @@ class LEditsPPPipelineStableDiffusionXL(
833
829
  **kwargs,
834
830
  ):
835
831
  r"""
836
- The call function to the pipeline for editing. The [`~pipelines.ledits_pp.LEditsPPPipelineStableDiffusionXL.invert`]
837
- method has to be called beforehand. Edits will always be performed for the last inverted image(s).
832
+ The call function to the pipeline for editing. The
833
+ [`~pipelines.ledits_pp.LEditsPPPipelineStableDiffusionXL.invert`] method has to be called beforehand. Edits
834
+ will always be performed for the last inverted image(s).
838
835
 
839
836
  Args:
840
837
  denoising_end (`float`, *optional*):
@@ -850,11 +847,11 @@ class LEditsPPPipelineStableDiffusionXL(
850
847
  negative_prompt_2 (`str` or `List[str]`, *optional*):
851
848
  The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
852
849
  `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
853
- negative_prompt_embeds (`torch.FloatTensor`, *optional*):
850
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
854
851
  Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
855
852
  weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
856
853
  argument.
857
- negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
854
+ negative_pooled_prompt_embeds (`torch.Tensor`, *optional*):
858
855
  Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
859
856
  weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
860
857
  input argument.
@@ -868,7 +865,7 @@ class LEditsPPPipelineStableDiffusionXL(
868
865
  of a plain tuple.
869
866
  callback (`Callable`, *optional*):
870
867
  A function that will be called every `callback_steps` steps during inference. The function will be
871
- called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
868
+ called with the following arguments: `callback(step: int, timestep: int, latents: torch.Tensor)`.
872
869
  callback_steps (`int`, *optional*, defaults to 1):
873
870
  The frequency at which the `callback` function will be called. If not specified, the callback will be
874
871
  called at every step.
@@ -892,11 +889,11 @@ class LEditsPPPipelineStableDiffusionXL(
892
889
  section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
893
890
  editing_prompt (`str` or `List[str]`, *optional*):
894
891
  The prompt or prompts to guide the image generation. The image is reconstructed by setting
895
- `editing_prompt = None`. Guidance direction of prompt should be specified via `reverse_editing_direction`.
892
+ `editing_prompt = None`. Guidance direction of prompt should be specified via
893
+ `reverse_editing_direction`.
896
894
  editing_prompt_embeddings (`torch.Tensor`, *optional*):
897
- Pre-generated edit text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
898
- weighting. If not provided, editing_prompt_embeddings will be generated from `editing_prompt` input
899
- argument.
895
+ Pre-generated edit text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
896
+ If not provided, editing_prompt_embeddings will be generated from `editing_prompt` input argument.
900
897
  editing_pooled_prompt_embeddings (`torch.Tensor`, *optional*):
901
898
  Pre-generated pooled edit text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
902
899
  weighting. If not provided, editing_prompt_embeddings will be generated from `editing_prompt` input
@@ -904,35 +901,36 @@ class LEditsPPPipelineStableDiffusionXL(
904
901
  reverse_editing_direction (`bool` or `List[bool]`, *optional*, defaults to `False`):
905
902
  Whether the corresponding prompt in `editing_prompt` should be increased or decreased.
906
903
  edit_guidance_scale (`float` or `List[float]`, *optional*, defaults to 5):
907
- Guidance scale for guiding the image generation. If provided as list values should correspond to `editing_prompt`.
908
- `edit_guidance_scale` is defined as `s_e` of equation 12 of
909
- [LEDITS++ Paper](https://arxiv.org/abs/2301.12247).
904
+ Guidance scale for guiding the image generation. If provided as list values should correspond to
905
+ `editing_prompt`. `edit_guidance_scale` is defined as `s_e` of equation 12 of [LEDITS++
906
+ Paper](https://arxiv.org/abs/2301.12247).
910
907
  edit_warmup_steps (`float` or `List[float]`, *optional*, defaults to 10):
911
908
  Number of diffusion steps (for each prompt) for which guidance is not applied.
912
909
  edit_cooldown_steps (`float` or `List[float]`, *optional*, defaults to `None`):
913
910
  Number of diffusion steps (for each prompt) after which guidance is no longer applied.
914
911
  edit_threshold (`float` or `List[float]`, *optional*, defaults to 0.9):
915
912
  Masking threshold of guidance. Threshold should be proportional to the image region that is modified.
916
- 'edit_threshold' is defined as 'λ' of equation 12 of [LEDITS++ Paper](https://arxiv.org/abs/2301.12247).
913
+ 'edit_threshold' is defined as 'λ' of equation 12 of [LEDITS++
914
+ Paper](https://arxiv.org/abs/2301.12247).
917
915
  sem_guidance (`List[torch.Tensor]`, *optional*):
918
916
  List of pre-generated guidance vectors to be applied at generation. Length of the list has to
919
917
  correspond to `num_inference_steps`.
920
918
  use_cross_attn_mask:
921
919
  Whether cross-attention masks are used. Cross-attention masks are always used when use_intersect_mask
922
- is set to true. Cross-attention masks are defined as 'M^1' of equation 12 of
923
- [LEDITS++ paper](https://arxiv.org/pdf/2311.16711.pdf).
920
+ is set to true. Cross-attention masks are defined as 'M^1' of equation 12 of [LEDITS++
921
+ paper](https://arxiv.org/pdf/2311.16711.pdf).
924
922
  use_intersect_mask:
925
- Whether the masking term is calculated as intersection of cross-attention masks and masks derived
926
- from the noise estimate. Cross-attention mask are defined as 'M^1' and masks derived from the noise
927
- estimate are defined as 'M^2' of equation 12 of [LEDITS++ paper](https://arxiv.org/pdf/2311.16711.pdf).
923
+ Whether the masking term is calculated as intersection of cross-attention masks and masks derived from
924
+ the noise estimate. Cross-attention mask are defined as 'M^1' and masks derived from the noise estimate
925
+ are defined as 'M^2' of equation 12 of [LEDITS++ paper](https://arxiv.org/pdf/2311.16711.pdf).
928
926
  user_mask:
929
- User-provided mask for even better control over the editing process. This is helpful when LEDITS++'s implicit
930
- masks do not meet user preferences.
927
+ User-provided mask for even better control over the editing process. This is helpful when LEDITS++'s
928
+ implicit masks do not meet user preferences.
931
929
  attn_store_steps:
932
930
  Steps for which the attention maps are stored in the AttentionStore. Just for visualization purposes.
933
931
  store_averaged_over_steps:
934
- Whether the attention maps for the 'attn_store_steps' are stored averaged over the diffusion steps.
935
- If False, attention maps for each step are stores separately. Just for visualization purposes.
932
+ Whether the attention maps for the 'attn_store_steps' are stored averaged over the diffusion steps. If
933
+ False, attention maps for each step are stores separately. Just for visualization purposes.
936
934
  clip_skip (`int`, *optional*):
937
935
  Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
938
936
  the output of the pre-final layer will be used for computing the prompt embeddings.
@@ -950,8 +948,8 @@ class LEditsPPPipelineStableDiffusionXL(
950
948
 
951
949
  Returns:
952
950
  [`~pipelines.ledits_pp.LEditsPPDiffusionPipelineOutput`] or `tuple`:
953
- [`~pipelines.ledits_pp.LEditsPPDiffusionPipelineOutput`] if `return_dict` is True,
954
- otherwise a `tuple. When returning a tuple, the first element is a list with the generated images.
951
+ [`~pipelines.ledits_pp.LEditsPPDiffusionPipelineOutput`] if `return_dict` is True, otherwise a `tuple. When
952
+ returning a tuple, the first element is a list with the generated images.
955
953
  """
956
954
  if self.inversion_steps is None:
957
955
  raise ValueError(
@@ -1417,7 +1415,6 @@ class LEditsPPPipelineStableDiffusionXL(
1417
1415
  if needs_upcasting:
1418
1416
  image = image.float()
1419
1417
  self.upcast_vae()
1420
- image = image.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)
1421
1418
 
1422
1419
  x0 = self.vae.encode(image).latent_dist.mode()
1423
1420
  x0 = x0.to(dtype)
@@ -1444,11 +1441,11 @@ class LEditsPPPipelineStableDiffusionXL(
1444
1441
  cross_attention_kwargs: Optional[Dict[str, Any]] = None,
1445
1442
  ):
1446
1443
  r"""
1447
- The function to the pipeline for image inversion as described by the [LEDITS++ Paper](https://arxiv.org/abs/2301.12247).
1448
- If the scheduler is set to [`~schedulers.DDIMScheduler`] the inversion proposed by [edit-friendly DPDM](https://arxiv.org/abs/2304.06140)
1449
- will be performed instead.
1444
+ The function to the pipeline for image inversion as described by the [LEDITS++
1445
+ Paper](https://arxiv.org/abs/2301.12247). If the scheduler is set to [`~schedulers.DDIMScheduler`] the
1446
+ inversion proposed by [edit-friendly DPDM](https://arxiv.org/abs/2304.06140) will be performed instead.
1450
1447
 
1451
- Args:
1448
+ Args:
1452
1449
  image (`PipelineImageInput`):
1453
1450
  Input for the image(s) that are to be edited. Multiple input images have to default to the same aspect
1454
1451
  ratio.
@@ -1470,8 +1467,8 @@ class LEditsPPPipelineStableDiffusionXL(
1470
1467
  Portion of initial steps that will be ignored for inversion and subsequent generation. Lower values
1471
1468
  will lead to stronger changes to the input image. `skip` has to be between `0` and `1`.
1472
1469
  generator (`torch.Generator`, *optional*):
1473
- A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
1474
- inversion deterministic.
1470
+ A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make inversion
1471
+ deterministic.
1475
1472
  crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)):
1476
1473
  `crops_coords_top_left` can be used to generate an image that appears to be "cropped" from the position
1477
1474
  `crops_coords_top_left` downwards. Favorable, well-centered images are usually achieved by setting
@@ -1486,8 +1483,8 @@ class LEditsPPPipelineStableDiffusionXL(
1486
1483
  [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
1487
1484
 
1488
1485
  Returns:
1489
- [`~pipelines.ledits_pp.LEditsPPInversionPipelineOutput`]:
1490
- Output will contain the resized input image(s) and respective VAE reconstruction(s).
1486
+ [`~pipelines.ledits_pp.LEditsPPInversionPipelineOutput`]: Output will contain the resized input image(s)
1487
+ and respective VAE reconstruction(s).
1491
1488
  """
1492
1489
 
1493
1490
  # Reset attn processor, we do not want to store attn maps during inversion
@@ -1646,9 +1643,21 @@ class LEditsPPPipelineStableDiffusionXL(
1646
1643
 
1647
1644
  # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.rescale_noise_cfg
1648
1645
  def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
1649
- """
1650
- Rescale `noise_cfg` according to `guidance_rescale`. Based on findings of [Common Diffusion Noise Schedules and
1651
- Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf). See Section 3.4
1646
+ r"""
1647
+ Rescales `noise_cfg` tensor based on `guidance_rescale` to improve image quality and fix overexposure. Based on
1648
+ Section 3.4 from [Common Diffusion Noise Schedules and Sample Steps are
1649
+ Flawed](https://arxiv.org/pdf/2305.08891.pdf).
1650
+
1651
+ Args:
1652
+ noise_cfg (`torch.Tensor`):
1653
+ The predicted noise tensor for the guided diffusion process.
1654
+ noise_pred_text (`torch.Tensor`):
1655
+ The predicted noise tensor for the text-guided diffusion process.
1656
+ guidance_rescale (`float`, *optional*, defaults to 0.0):
1657
+ A rescale factor applied to the noise predictions.
1658
+
1659
+ Returns:
1660
+ noise_cfg (`torch.Tensor`): The rescaled noise prediction tensor.
1652
1661
  """
1653
1662
  std_text = noise_pred_text.std(dim=list(range(1, noise_pred_text.ndim)), keepdim=True)
1654
1663
  std_cfg = noise_cfg.std(dim=list(range(1, noise_cfg.ndim)), keepdim=True)
@@ -35,8 +35,8 @@ class LEditsPPInversionPipelineOutput(BaseOutput):
35
35
  List of the cropped and resized input images as PIL images of length `batch_size` or NumPy array of shape `
36
36
  (batch_size, height, width, num_channels)`.
37
37
  vae_reconstruction_images (`List[PIL.Image.Image]` or `np.ndarray`)
38
- List of VAE reconstruction of all input images as PIL images of length `batch_size` or NumPy array of shape `
39
- (batch_size, height, width, num_channels)`.
38
+ List of VAE reconstruction of all input images as PIL images of length `batch_size` or NumPy array of shape
39
+ ` (batch_size, height, width, num_channels)`.
40
40
  """
41
41
 
42
42
  images: Union[List[PIL.Image.Image], np.ndarray]
@@ -0,0 +1,50 @@
1
+ from typing import TYPE_CHECKING
2
+
3
+ from ...utils import (
4
+ DIFFUSERS_SLOW_IMPORT,
5
+ OptionalDependencyNotAvailable,
6
+ _LazyModule,
7
+ get_objects_from_module,
8
+ is_torch_available,
9
+ is_transformers_available,
10
+ )
11
+
12
+
13
+ _dummy_objects = {}
14
+ _import_structure = {}
15
+
16
+
17
+ try:
18
+ if not (is_transformers_available() and is_torch_available()):
19
+ raise OptionalDependencyNotAvailable()
20
+ except OptionalDependencyNotAvailable:
21
+ from ...utils import dummy_torch_and_transformers_objects # noqa F403
22
+
23
+ _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
24
+ else:
25
+ _import_structure["pipeline_ltx"] = ["LTXPipeline"]
26
+ _import_structure["pipeline_ltx_image2video"] = ["LTXImageToVideoPipeline"]
27
+
28
+ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
29
+ try:
30
+ if not (is_transformers_available() and is_torch_available()):
31
+ raise OptionalDependencyNotAvailable()
32
+
33
+ except OptionalDependencyNotAvailable:
34
+ from ...utils.dummy_torch_and_transformers_objects import *
35
+ else:
36
+ from .pipeline_ltx import LTXPipeline
37
+ from .pipeline_ltx_image2video import LTXImageToVideoPipeline
38
+
39
+ else:
40
+ import sys
41
+
42
+ sys.modules[__name__] = _LazyModule(
43
+ __name__,
44
+ globals()["__file__"],
45
+ _import_structure,
46
+ module_spec=__spec__,
47
+ )
48
+
49
+ for name, value in _dummy_objects.items():
50
+ setattr(sys.modules[__name__], name, value)