Add Photon model and pipeline support#2

Open

DavidBert wants to merge 52 commits intomainfrom

Commits on Oct 21, 2025

Add Photon model and pipeline support

davidb
authored and
DavidBert
committed
just store the T5Gemma encoder

davidb
authored and
DavidBert
committed
enhance_vae_properties if vae is provided only

davidb
authored and
DavidBert
committed
remove autocast for text encoder forwad

davidb
authored and
DavidBert
committed
BF16 example

david-PHR
authored and
DavidBert
committed
conditioned CFG

davidb
authored and
DavidBert
committed
remove enhance vae and use vae.config directly when possible

davidb
authored and
DavidBert
committed
move PhotonAttnProcessor2_0 in transformer_photon

davidb
authored and
DavidBert
committed
remove einops dependency and now inherits from AttentionMixin

davidb
authored and
DavidBert
committed
unify the structure of the forward block

davidb
authored and
DavidBert
committed
update doc

davidb
authored and
DavidBert
committed
update doc

davidb
authored and
DavidBert
committed
fix T5Gemma loading from hub

davidb
authored and
DavidBert
committed
fix timestep shift

davidb
authored and
DavidBert
committed
remove lora support from doc

davidb
authored and
DavidBert
committed
Rename EmbedND for PhotoEmbedND
DavidBert
committed
remove modulation dataclass
DavidBert
committed
put _attn_forward and _ffn_forward logic in PhotonBlock's forward
DavidBert
committed
renam LastLayer for FinalLayer
DavidBert
committed
remove lora related code
DavidBert
committed
rename vae_spatial_compression_ratio for vae_scale_factor
DavidBert
committed
support prompt_embeds in call
DavidBert
committed
move xattention conditionning out computation out of the denoising loop
DavidBert
committed
add negative prompts
DavidBert
committed
Use _import_structure for lazy loading
DavidBert
committed
make quality + style
DavidBert
committed
add pipeline test + corresponding fixes
DavidBert
committed
utility function that determines the default resolution given the VAE
DavidBert
committed
Refactor PhotonAttention to match Flux pattern
DavidBert
committed
built-in RMSNorm
DavidBert
committed
Revert accidental .gitignore change
DavidBert
committed
parameter names match the standard diffusers conventions
DavidBert
committed
renaming and remove unecessary attributes setting
DavidBert
committed
Update docs/source/en/api/pipelines/photon.md

DavidBert
and
stevhliu
committed
quantization example
DavidBert
committed
added doc to toctree
DavidBert
committed
Update docs/source/en/api/pipelines/photon.md

DavidBert
and
stevhliu
committed
Update docs/source/en/api/pipelines/photon.md

DavidBert
and
stevhliu
committed
Update docs/source/en/api/pipelines/photon.md

DavidBert
and
stevhliu
committed
use dispatch_attention_fn for multiple attention backend support
DavidBert
committed
naming changes
DavidBert
committed
make fix copy
DavidBert
committed
Update docs/source/en/api/pipelines/photon.md

DavidBert
and
dg845
committed
Add PhotonTransformer2DModel to TYPE_CHECKING imports
DavidBert
committed
make fix-copies
DavidBert
committed
Use Tuple instead of tuple

DavidBert
and
dg845
committed
restrict the version of transformers

DavidBert
and
dg845
committed
Update tests/pipelines/photon/test_pipeline_photon.py

DavidBert
and
dg845
committed
Update tests/pipelines/photon/test_pipeline_photon.py

DavidBert
and
dg845
committed
change | for Optional
DavidBert
committed
fix nits.

sayakpaul
authored and
DavidBert
committed
use typing Dict
DavidBert
committed