File size: 4,425 Bytes
4f48282
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# img_gen.py
#img_gen_modal.py
# img_gen.py
# img_gen_modal.py
import modal
import random
from datetime import datetime
import random
import io
from config.config import prompts, models  # Indirect import
import os
import torch
from huggingface_hub import login
from transformers import AutoTokenizer

# Define the Modal image
image = (
    #modal.Image.from_registry("nvidia/cuda:12.2.0-devel-ubuntu22.04", add_python="3.9")
    modal.Image.debian_slim(python_version="3.9")  # Base image

    .apt_install(
        "git",
    )
    .pip_install(
        "diffusers",
        "transformers",
        "torch",
        "accelerate",
        "gradio>=4.44.1",
        "safetensors",
        "pillow",
        "sentencepiece",
        "hf_transfer",
        "huggingface_hub[hf_transfer]",
        "aria2",  # aria2 for ultra-fast parallel downloads
        f"git+https://github.com/huggingface/transformers.git"
    )
    .env(
        {
            "HF_HUB_ENABLE_HF_TRANSFER": "1", "HF_HOME": "HF_HOME"
        }
    )
)

# Create a Modal app
app = modal.App("img-gen-modal", image=image)
with image.imports():
    import diffusers
    import os
    import gradio
    import torch
    import sentencepiece

#flux_model_vol = modal.Volume.from_name("flux-model-vol", create_if_missing=True)  # Reference your volume

@app.function(
        secrets=[modal.Secret.from_name("huggingface-token")],
        #volumes={"/data": flux_model_vol},
        gpu="a100-80gb"
        )
def generate_image(prompt_alias, team_color, model_alias, custom_prompt, height=36, width=64, num_inference_steps=2, guidance_scale=2.0, seed=-1):
    # Find the selected prompt and model
    try:
        prompt = next(p for p in prompts if p["alias"] == prompt_alias)["text"]
        model_name = next(m for m in models if m["alias"] == model_alias)["name"]
    except StopIteration:
        return None, "ERROR: Invalid prompt or model selected."

    # Determine the enemy color
    enemy_color = "blue" if team_color.lower() == "red" else "red"

    # Print the original prompt and dynamic values for debugging
    print("Original Prompt:")
    print(prompt)
    print(f"Enemy Color: {enemy_color}")
    print(f"Team Color: {team_color.lower()}")

    prompt = prompt.format(team_color=team_color.lower(), enemy_color=enemy_color)

    # Print the formatted prompt for debugging
    print("\nFormatted Prompt:")
    print(prompt)

    # Append the custom prompt (if provided)
    if custom_prompt and len(custom_prompt.strip()) > 0:
        prompt += " " + custom_prompt.strip()

    # Randomize the seed if needed
    if seed == -1:
        seed = random.randint(0, 1000000)

    try:
        from diffusers import DiffusionPipeline
        print("Initializing HF TOKEN")
        hf_token = os.environ["HF_TOKEN"]
        print(hf_token)
        print("HF TOKEN:")
        login(token=hf_token)
        print("model_name:")
        print(model_name)
        
        print("Initializing PIPE")
        pipe = DiffusionPipeline.from_pretrained(model_name)
        pipe = pipe.to("cuda")
    except Exception as e:
        print(f"Detailed error: {str(e)}")
        return None, f"ERROR: Failed to initialize PIPE. Details: {e}"
    try:
        print("Sending img gen to pipe")
        image = pipe(
            prompt,
            guidance_scale=guidance_scale,
            num_inference_steps=num_inference_steps,
            width=width,
            height=height,
            # seed=seed
        ).images[0]
        image.save("image.png")

    except Exception as e:
        return None, f"ERROR: Failed to generate image. Details: {e}"

    # Save the image with a timestamped filename
    timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
    output_filename = f"{timestamp}_{model_alias.replace(' ', '_').lower()}_{prompt_alias.replace(' ', '_').lower()}_{team_color.lower()}.png"
    try:
        # The pipeline typically returns images in a specific format
        # Usually it's image.images[0] for the first generated image
        image_output = image.images[0]  # Get the actual PIL Image from the output
        image_output.save(output_filename)  # Save using PIL's save method
    except Exception as e:
        return None, f"ERROR: Failed to save image. Details: {e}"
    print(f"Image output type: {type(image)}")
    print(f"Image output attributes: {dir(image)}")

    return output_filename, "Image generated successfully!"