[Core] is_cosxl_edit arg in SDXL ip2p.

[sayakpaul]

What does this PR do?

As found out by @yiyixuxu here:
#7621 (comment)

Related to #7621.

Cc: @apolinario for awareness.

[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.

[sayakpaul]

@DN6 your comments have been addressed.

[sayakpaul]

Failing test is quite unrelated.

[sayakpaul]

Here's testing code:

import torch
from diffusers import StableDiffusionXLInstructPix2PixPipeline, EDMEulerScheduler
from diffusers.utils import load_image
import numpy as np
import math

def set_timesteps_patched(self, num_inference_steps: int, device = None):
    self.num_inference_steps = num_inference_steps
    
    ramp = np.linspace(0, 1, self.num_inference_steps)
    sigmas = torch.linspace(math.log(self.config.sigma_min), math.log(self.config.sigma_max), len(ramp)).exp().flip(0)
    
    sigmas = (sigmas).to(dtype=torch.float32, device=device)
    self.timesteps = self.precondition_noise(sigmas)
    
    self.sigmas = torch.cat([sigmas, torch.zeros(1, device=sigmas.device)])
    self._step_index = None
    self._begin_index = None
    self.sigmas = self.sigmas.to("cpu")  # to avoid too much CPU/GPU communication

inst_file = "https://huggingface.co/stabilityai/cosxl/blob/main/cosxl_edit.safetensors"

pipe = StableDiffusionXLInstructPix2PixPipeline.from_single_file(
    inst_file, num_in_channels=8, is_cosxl_edit=True
).to("cuda")

# This won't be needed after https://github.com/huggingface/diffusers/pull/7649
EDMEulerScheduler.set_timesteps = set_timesteps_patched
pipe.scheduler = EDMEulerScheduler(sigma_min=0.002, sigma_max=120.0, sigma_data=1.0, prediction_type="v_prediction")

resolution = 1024
image = load_image(
    "https://hf.co/datasets/diffusers/diffusers-images-docs/resolve/main/mountain.png"
).resize((resolution, resolution))

edit_instruction = "Turn sky into a cloudy one"
edited_image = pipe(
    prompt=edit_instruction,
    image=image,
    height=resolution,
    width=resolution,
    num_inference_steps=20,
).images[0]
edited_image.save("edited_image.png")

^ matches what Yiyi got here: #7621 (comment).

[sayakpaul]

Created an issue for the installation failure here: astral-sh/uv#3031.

I will submit a quick PR replacing the uv installation step that's causing the failure to unblock this PR.

[DN6]

LGTM 👍🏽. The failing test looks unrelated.