Add support for Ovis-Image

[DoctorKey]

What does this PR do?

This PR introduces Ovis-Image into the diffusers library. Ovis-Image integrates a diffusion-based visual decoder with the Ovis 2.5 multimodal backbone, leveraging a text-centric training pipeline that combines large-scale pre-training with carefully tailored post-training refinements. Despite its compact architecture, Ovis-Image achieves text rendering performance on par with significantly larger open models such as Qwen-Image and approaches closed-source systems like Seedream and GPT4o.

[DoctorKey]

Ovis-Image has been released:

• GitHub: https://github.com/AIDC-AI/Ovis-Image
• Hugging Face: https://huggingface.co/AIDC-AI/Ovis-Image-7B

[yiyixuxu]

@bot /style

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[yiyixuxu]

Thanks so much for the PR! I left a few feedbacks, and I think we can merge this very soon

Congrats on the release!! Sorry, we overlooked the PR (it was the thanksgiving holiday in US)
We will reach out to set up a collaboration channel for your future release.

[yiyixuxu]

Suggested change

	def enable_vae_slicing(self):
	r"""
	Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
	compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
	"""
	depr_message = f"Calling `enable_vae_slicing()` on a `{self.__class__.__name__}` is deprecated and this method will be removed in a future version. Please use `pipe.vae.enable_slicing()`."
	deprecate(
	"enable_vae_slicing",
	"0.40.0",
	depr_message,
	)
	self.vae.enable_slicing()
	def disable_vae_slicing(self):
	r"""
	Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
	computing decoding in one step.
	"""
	depr_message = f"Calling `disable_vae_slicing()` on a `{self.__class__.__name__}` is deprecated and this method will be removed in a future version. Please use `pipe.vae.disable_slicing()`."
	deprecate(
	"disable_vae_slicing",
	"0.40.0",
	depr_message,
	)
	self.vae.disable_slicing()
	def enable_vae_tiling(self):
	r"""
	Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to
	compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
	processing larger images.
	"""
	depr_message = f"Calling `enable_vae_tiling()` on a `{self.__class__.__name__}` is deprecated and this method will be removed in a future version. Please use `pipe.vae.enable_tiling()`."
	deprecate(
	"enable_vae_tiling",
	"0.40.0",
	depr_message,
	)
	self.vae.enable_tiling()
	def disable_vae_tiling(self):
	r"""
	Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
	computing decoding in one step.
	"""
	depr_message = f"Calling `disable_vae_tiling()` on a `{self.__class__.__name__}` is deprecated and this method will be removed in a future version. Please use `pipe.vae.disable_tiling()`."
	deprecate(
	"disable_vae_tiling",
	"0.40.0",
	depr_message,
	)
	self.vae.disable_tiling()

[yiyixuxu]

Suggested change

	true_cfg_scale: float = 5.0,
	guidance_scale: float = 5.0,

can we use guidance_scale if it is not a distilled checkpoint? since model is already out with this PR, we can add a deprecation message if you prefer

[yiyixuxu]

Suggested change

	if image_embeds is not None:
	self._joint_attention_kwargs["ip_adapter_image_embeds"] = image_embeds

let's remove the IP-adapter related logics if we don't support it yet

[yiyixuxu]

so flux/qwen pipelines were written this way to support both distilled guidance and regular CFG - the user experience was pretty bad, and we regret the design choise so very much.
If Ovis only supports regular CFG let's not follow their path :)

[yiyixuxu]

for standard CFG, one pattern you can do is

prompt_embeds, text_ids = self.encode_prompt(...)
if do_classifier_free_guidance:
    negative_prompt_embeds, negative_text_ids = self.encode_prompt(..)

[yiyixuxu]

Suggested change

	image_embeds = None
	negative_image_embeds = None