[core] DyPE + SEGA hook for training-free high-resolution generation (upstream candidate)

docs/source/en/_toctree.yml

@@ -174,6 +174,8 @@
     title: bitsandbytes
   - local: quantization/gguf
     title: gguf
+  - local: quantization/nunchaku
+    title: Nunchaku Lite
   - local: quantization/torchao
     title: torchao
   - local: quantization/quanto
@@ -182,6 +184,8 @@
     title: NVIDIA ModelOpt
   - local: quantization/autoround
     title: AutoRound
+  - local: quantization/sdnq
+    title: SDNQ
   title: Quantization
 - isExpanded: false
   sections:
@@ -218,6 +222,8 @@
     title: Task recipes
   - local: using-diffusers/write_own_pipeline
     title: Understanding pipelines, models and schedulers
+  - local: using-diffusers/cli
+    title: Command line interface
   - local: community_projects
     title: Projects built with Diffusers
   - local: conceptual/philosophy
@@ -228,8 +234,6 @@
     title: How to contribute?
   - local: conceptual/ethical_guidelines
     title: Diffusers' Ethical Guidelines
-  - local: conceptual/evaluation
-    title: Evaluating Diffusion Models
   title: Resources
 - isExpanded: false
   sections:
@@ -353,8 +357,12 @@
         title: HunyuanVideoTransformer3DModel
       - local: api/models/ideogram4_transformer2d
         title: Ideogram4Transformer2DModel
+      - local: api/models/transformer_joyimage_edit_plus
+        title: JoyImageEditPlusTransformer3DModel
       - local: api/models/transformer_joyimage
         title: JoyImageEditTransformer3DModel
+      - local: api/models/krea2_transformer2d
+        title: Krea2Transformer2DModel
       - local: api/models/latte_transformer3d
         title: LatteTransformer3DModel
       - local: api/models/longcat_image_transformer2d
@@ -367,6 +375,10 @@
         title: Lumina2Transformer2DModel
       - local: api/models/lumina_nextdit2d
         title: LuminaNextDiT2DModel
+      - local: api/models/minimax_h3_transformer3d
+        title: MiniMaxH3Transformer3DModel
+      - local: api/models/minimax_music3_transformer
+        title: MiniMaxMusic3Transformer1DModel
       - local: api/models/mochi_transformer3d
         title: MochiTransformer3DModel
       - local: api/models/motif_video_transformer_3d
@@ -395,6 +407,8 @@
         title: Transformer2DModel
       - local: api/models/transformer_temporal
         title: TransformerTemporalModel
+      - local: api/models/wan_animate_2_transformer_3d
+        title: WanAnimate2Transformer3DModel
       - local: api/models/wan_animate_transformer_3d
         title: WanAnimateTransformer3DModel
       - local: api/models/wan_transformer_3d
@@ -451,6 +465,10 @@
         title: AutoencoderKLLTXVideo
       - local: api/models/autoencoderkl_magvit
         title: AutoencoderKLMagvit
+      - local: api/models/autoencoderkl_minimax_h3
+        title: AutoencoderKLMiniMaxH3
+      - local: api/models/autoencoderkl_minimax_h3_audio
+        title: AutoencoderKLMiniMaxH3Audio
       - local: api/models/autoencoderkl_mochi
         title: AutoencoderKLMochi
       - local: api/models/autoencoderkl_qwenimage
@@ -461,6 +479,8 @@
         title: AutoencoderRAE
       - local: api/models/consistency_decoder_vae
         title: ConsistencyDecoderVAE
+      - local: api/models/ltx2_diffusion_decoder
+        title: LTX2VideoDiffusionDecoderModel
       - local: api/models/autoencoder_oobleck
         title: Oobleck AutoEncoder
       - local: api/models/autoencoder_tiny
@@ -553,6 +573,8 @@
         title: InstructPix2Pix
       - local: api/pipelines/joyimage_edit
         title: JoyImage Edit
+      - local: api/pipelines/joyimage_edit_plus
+        title: JoyImage Edit Plus
       - local: api/pipelines/kandinsky
         title: Kandinsky 2.1
       - local: api/pipelines/kandinsky_v22
@@ -563,6 +585,8 @@
         title: Kandinsky 5.0 Image
       - local: api/pipelines/kolors
         title: Kolors
+      - local: api/pipelines/krea2
+        title: Krea 2
       - local: api/pipelines/latent_consistency_models
         title: Latent Consistency Models
       - local: api/pipelines/latent_diffusion
@@ -591,6 +615,8 @@
         title: PixArt-Σ
       - local: api/pipelines/prx
         title: PRX
+      - local: api/pipelines/prx_pixel
+        title: PRX Pixel
       - local: api/pipelines/qwenimage
         title: QwenImage
       - local: api/pipelines/sana
@@ -641,6 +667,8 @@
         title: Z-Image
       title: Image
     - sections:
+      - local: api/pipelines/diffusion_gemma
+        title: DiffusionGemma
       - local: api/pipelines/llada2
         title: LLaDA2
       title: Text
@@ -675,6 +703,10 @@
         title: LTX-2
       - local: api/pipelines/ltx_video
         title: LTXVideo
+      - local: api/pipelines/minimax_music3
+        title: MiniMax Music 3
+      - local: api/pipelines/minimax_h3
+        title: MiniMax-H3
       - local: api/pipelines/mochi
         title: Mochi
       - local: api/pipelines/motif_video
@@ -685,6 +717,8 @@
         title: Stable Video Diffusion
       - local: api/pipelines/wan
         title: Wan
+      - local: api/pipelines/wan_animate_2
+        title: Wan-Animate-2
       title: Video
     title: Pipelines
   - sections:
@@ -710,6 +744,8 @@
       title: DDPMScheduler
     - local: api/schedulers/deis
       title: DEISMultistepScheduler
+    - local: api/schedulers/discrete_ddim
+      title: DiscreteDDIMScheduler
     - local: api/schedulers/multistep_dpm_solver_inverse
       title: DPMSolverMultistepInverse
     - local: api/schedulers/multistep_dpm_solver
@@ -722,6 +758,8 @@
       title: EDMDPMSolverMultistepScheduler
     - local: api/schedulers/edm_euler
       title: EDMEulerScheduler
+    - local: api/schedulers/entropy_bound
+      title: EntropyBoundScheduler
     - local: api/schedulers/euler_ancestral
       title: EulerAncestralDiscreteScheduler
     - local: api/schedulers/euler
@@ -750,6 +788,8 @@
       title: LCMScheduler
     - local: api/schedulers/lms_discrete
       title: LMSDiscreteScheduler
+    - local: api/schedulers/minimax_h3
+      title: MiniMaxH3Scheduler
     - local: api/schedulers/pndm
       title: PNDMScheduler
     - local: api/schedulers/repaint
@@ -774,6 +814,8 @@
       title: Custom activation functions
     - local: api/cache
       title: Caching methods
+    - local: api/dype
+      title: Resolution extrapolation
     - local: api/normalization
       title: Custom normalization layers
     - local: api/utilities

docs/source/en/api/dype.md

@@ -0,0 +1,46 @@
+<!-- Copyright 2025 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License. -->
+
+# Resolution extrapolation
+
+Training-free methods that let RoPE-based diffusion transformers such as [`FluxTransformer2DModel`] generate above their trained resolution (for example 4096x4096 from a model trained at 1024x1024), with no fine-tuning and no extra sampling cost.
+
+[DyPE](https://huggingface.co/papers/2510.20766) (Dynamic Position Extrapolation) swaps the transformer's rotary positional embedding for a timestep-aware YaRN / NTK-by-parts schedule that only engages above the trained resolution. [SEGA](https://huggingface.co/papers/2605.22668) (Spectral-Energy Guided Attention) additionally applies a per-frequency, content-aware attention temperature derived from the latent's spectrum, which suppresses the high-frequency speckle a scalar temperature can leave in flat regions at very high resolutions. Both are enabled through [`apply_dype`].
+
+```python
+import torch
+from diffusers import FluxPipeline, apply_dype
+
+pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-Krea-dev", torch_dtype=torch.bfloat16)
+pipe.enable_model_cpu_offload()
+
+# method="yarn" (default) is plain DyPE; method="spectral" adds SEGA spectral attention.
+apply_dype(pipe.transformer, method="spectral")
+
+# Above the trained resolution, also flatten the flow-matching shift schedule so the sampler
+# does not stall near pure noise (the default shift `mu` grows with the image sequence length).
+pipe.scheduler.register_to_config(base_shift=1.15, max_shift=1.15)
+
+image = pipe(
+    "a sunlit alpine meadow, snow-capped peaks, clear blue sky",
+    height=4096,
+    width=4096,
+    guidance_scale=4.5,
+    num_inference_steps=28,
+).images[0]
+```
+
+> [!TIP]
+> `apply_dype` is a no-op at or below the trained resolution (1024x1024 for Flux), so the hook can stay applied for standard-resolution generation.
+
+## apply_dype
+
+[[autodoc]] apply_dype

src/diffusers/__init__.py

@@ -175,6 +175,7 @@
     )
     _import_structure["hooks"].extend(
         [
+            "DyPEHook",
             "FasterCacheConfig",
             "FirstBlockCacheConfig",
             "HookRegistry",
@@ -184,6 +185,7 @@
             "SmoothedEnergyGuidanceConfig",
             "TaylorSeerCacheConfig",
             "TextKVCacheConfig",
+            "apply_dype",
             "apply_faster_cache",
             "apply_first_block_cache",
             "apply_layer_skip",
@@ -1036,6 +1038,7 @@
             TangentialClassifierFreeGuidance,
         )
         from .hooks import (
+            DyPEHook,
             FasterCacheConfig,
             FirstBlockCacheConfig,
             HookRegistry,
@@ -1045,6 +1048,7 @@
             SmoothedEnergyGuidanceConfig,
             TaylorSeerCacheConfig,
             TextKVCacheConfig,
+            apply_dype,
             apply_faster_cache,
             apply_first_block_cache,
             apply_layer_skip,

src/diffusers/hooks/__init__.py

@@ -17,6 +17,7 @@
 
 if is_torch_available():
     from .context_parallel import apply_context_parallel
+    from .dype import DyPEHook, apply_dype
     from .faster_cache import FasterCacheConfig, apply_faster_cache
     from .first_block_cache import FirstBlockCacheConfig, apply_first_block_cache
     from .group_offloading import apply_group_offloading