File size: 5,727 Bytes
a49cc2f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
import random
import unittest

import numpy as np
import torch
from transformers import AutoTokenizer, CLIPTextConfig, CLIPTextModel, CLIPTokenizer, T5EncoderModel

from diffusers import AutoencoderKL, FlowMatchEulerDiscreteScheduler, FluxInpaintPipeline, FluxTransformer2DModel
from diffusers.utils.testing_utils import (
    enable_full_determinism,
    floats_tensor,
    torch_device,
)

from ..test_pipelines_common import PipelineTesterMixin


enable_full_determinism()


class FluxInpaintPipelineFastTests(unittest.TestCase, PipelineTesterMixin):
    pipeline_class = FluxInpaintPipeline
    params = frozenset(["prompt", "height", "width", "guidance_scale", "prompt_embeds", "pooled_prompt_embeds"])
    batch_params = frozenset(["prompt"])
    test_xformers_attention = False

    def get_dummy_components(self):
        torch.manual_seed(0)
        transformer = FluxTransformer2DModel(
            patch_size=1,
            in_channels=8,
            num_layers=1,
            num_single_layers=1,
            attention_head_dim=16,
            num_attention_heads=2,
            joint_attention_dim=32,
            pooled_projection_dim=32,
            axes_dims_rope=[4, 4, 8],
        )
        clip_text_encoder_config = CLIPTextConfig(
            bos_token_id=0,
            eos_token_id=2,
            hidden_size=32,
            intermediate_size=37,
            layer_norm_eps=1e-05,
            num_attention_heads=4,
            num_hidden_layers=5,
            pad_token_id=1,
            vocab_size=1000,
            hidden_act="gelu",
            projection_dim=32,
        )

        torch.manual_seed(0)
        text_encoder = CLIPTextModel(clip_text_encoder_config)

        torch.manual_seed(0)
        text_encoder_2 = T5EncoderModel.from_pretrained("hf-internal-testing/tiny-random-t5")

        tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip")
        tokenizer_2 = AutoTokenizer.from_pretrained("hf-internal-testing/tiny-random-t5")

        torch.manual_seed(0)
        vae = AutoencoderKL(
            sample_size=32,
            in_channels=3,
            out_channels=3,
            block_out_channels=(4,),
            layers_per_block=1,
            latent_channels=2,
            norm_num_groups=1,
            use_quant_conv=False,
            use_post_quant_conv=False,
            shift_factor=0.0609,
            scaling_factor=1.5035,
        )

        scheduler = FlowMatchEulerDiscreteScheduler()

        return {
            "scheduler": scheduler,
            "text_encoder": text_encoder,
            "text_encoder_2": text_encoder_2,
            "tokenizer": tokenizer,
            "tokenizer_2": tokenizer_2,
            "transformer": transformer,
            "vae": vae,
        }

    def get_dummy_inputs(self, device, seed=0):
        image = floats_tensor((1, 3, 32, 32), rng=random.Random(seed)).to(device)
        mask_image = torch.ones((1, 1, 32, 32)).to(device)
        if str(device).startswith("mps"):
            generator = torch.manual_seed(seed)
        else:
            generator = torch.Generator(device="cpu").manual_seed(seed)

        inputs = {
            "prompt": "A painting of a squirrel eating a burger",
            "image": image,
            "mask_image": mask_image,
            "generator": generator,
            "num_inference_steps": 2,
            "guidance_scale": 5.0,
            "height": 32,
            "width": 32,
            "max_sequence_length": 48,
            "strength": 0.8,
            "output_type": "np",
        }
        return inputs

    def test_flux_inpaint_different_prompts(self):
        pipe = self.pipeline_class(**self.get_dummy_components()).to(torch_device)

        inputs = self.get_dummy_inputs(torch_device)
        output_same_prompt = pipe(**inputs).images[0]

        inputs = self.get_dummy_inputs(torch_device)
        inputs["prompt_2"] = "a different prompt"
        output_different_prompts = pipe(**inputs).images[0]

        max_diff = np.abs(output_same_prompt - output_different_prompts).max()

        # Outputs should be different here
        # For some reasons, they don't show large differences
        assert max_diff > 1e-6

    def test_flux_inpaint_prompt_embeds(self):
        pipe = self.pipeline_class(**self.get_dummy_components()).to(torch_device)
        inputs = self.get_dummy_inputs(torch_device)

        output_with_prompt = pipe(**inputs).images[0]

        inputs = self.get_dummy_inputs(torch_device)
        prompt = inputs.pop("prompt")

        (prompt_embeds, pooled_prompt_embeds, text_ids) = pipe.encode_prompt(
            prompt,
            prompt_2=None,
            device=torch_device,
            max_sequence_length=inputs["max_sequence_length"],
        )
        output_with_embeds = pipe(
            prompt_embeds=prompt_embeds,
            pooled_prompt_embeds=pooled_prompt_embeds,
            **inputs,
        ).images[0]

        max_diff = np.abs(output_with_prompt - output_with_embeds).max()
        assert max_diff < 1e-4

    def test_flux_image_output_shape(self):
        pipe = self.pipeline_class(**self.get_dummy_components()).to(torch_device)
        inputs = self.get_dummy_inputs(torch_device)

        height_width_pairs = [(32, 32), (72, 57)]
        for height, width in height_width_pairs:
            expected_height = height - height % (pipe.vae_scale_factor * 2)
            expected_width = width - width % (pipe.vae_scale_factor * 2)

            inputs.update({"height": height, "width": width})
            image = pipe(**inputs).images[0]
            output_height, output_width, _ = image.shape
            assert (output_height, output_width) == (expected_height, expected_width)