A step-by-step guide for application developers to get started with Imagen 3 on Vertex AI
Early adopters tested Imagen 3 on Vertex AI over the course of the last few months and provided with insightful feedback. Users obviously want an AI model that powers your practical creative applications and produces visually attractive content. Based on their input, Google Cloud has determined three recurring themes:
- Demand for unmatched excellence in a variety of artistic mediums and styles
- Strong swift adherence and quick image creation are desired.
- Advanced safety filters and SynthID watermarking are controls that safeguard and foster confidence.
We’ll go over each of these ideas in detail throughout this essay. To help you get the most of Imagen 3, we will also include some code examples and best prompt practices.
Unwavering versatility and quality
A new benchmark for quality and control over your created photographs is set with Imagen 3. This text-to-image model creates remarkably composed, crisp, color-accurate, and high-resolution photorealistic images. You can experiment with a greater range of artistic formats and styles with Imagen 3. With the model’s wider variety of styles and forms, you may create anything from amusing claymation scenarios to photorealistic masterpieces, giving you the freedom to express your own artistic vision.
Let’s go through an example of making image mockups for a new cookbook cover to show off these photorealistic capabilities. The resulting image, which was created with the prompt below, has amazing detail, composition, and photorealism.
import vertexai
from vertexai.preview.vision_models import ImageGenerationModel
TODO(developer): Update and un-comment below lines
project_id = “PROJECT_ID”
vertexai.init(project=PROJECT_ID, location=”us-central1″)
generation_model = ImageGenerationModel.from_pretrained(“imagen-3.0-generate-001”)
prompt = “””
A photorealistic image of a cookbook laying on a wooden kitchen table, the cover facing forward featuring a smiling family sitting at a similar table, soft overhead lighting illuminating the scene, the cookbook is the main focus of the image.
“””
image = generation_model.generate_images(
prompt=prompt,
number_of_images=1,
aspect_ratio=”1:1″,
safety_filter_level=”block_some”,
person_generation=”allow_all”,
)
OPTIONAL: View the generated image in a notebook
image[0].show()
Text Rendering
Regarding text rendering inside images, Imagen 3 also opens up new possibilities. Creating pictures of greeting cards, posters, and social media posts with captions in different fonts and colors is a great way to experiment with this tool. To use this function, simply write a brief written description of what you would like to see in the prompt.
Nearer to your purpose
No matter how complex your natural language descriptions are, Imagen 3’s rapid comprehension converts them into precisely matched pictures. In your description, you can detail everything from particular camera angles to different kinds of lenses to image compositions. Imagen 3 closely follows the cue, assisting in bridging the mental image and the final image. Simple subject-action-setting instructions or complex, multi-layered descriptions can be given to the model; it will adjust to your creative process to support a wide variety of styles.
Given that Imagen 3 performs well with complex prompts, giving strong details typically results in outcomes that are higher quality and more accurate. Some choices to think about when creating your prompts are listed below:
- Arrangement: Set the tone for the scene by indicating the locations of the subjects.
- Lighting: Adjust the lighting’s focus and direction to create a mood. Use strong or soft lighting.
- Camera angles and lens selections can be used to create depth and perspective.
- Styles: Create digital art, cinematic, vintage, minimalist, and more go beyond photorealism.
Decreased latency
Imagen 3 Fast
Google Cloud provide Imagen 3 Fast, which is geared for generation speed, in addition to Imagen 3, which is their highest quality model to date. For producing photographs with greater contrast and brightness, Imagen 3 Fast is appropriate. You can observe a 40% reduction in latency when compared to Imagen 2. You can use the same prompt to create two photos that illustrate these two models.
Take care of your creations and preserve them
With built-in security features, Imagen 3 allows you to maintain control while concentrating on your creative ideas. Imagen 3 uses a technique called SynthID, which is partnered with Google DeepMind and embeds an undetectable watermark at the pixel level. All photos created by Imagen 3 have a digital watermark attached to them by default, but you may specifically enable this functionality by using the add_watermark argument. The API can also be used to confirm if an image was created with Imagen. By confirming the legitimacy of your AI-generated photos, this promotes transparency and helps protect your creations from exploitation.
You have more control over the kinds of images that are created to ensure that they align with your brand’s values or ideals by using Imagen 3’s sophisticated safety filters. Change the safety_filter_level to set safety filter thresholds for generated photos. You can select “block_most,” “block_some,” or “block_few” as the safety level. Person_generation can be modified to “allow_all,” “allow_adult,” or “dont_allow” in order to alter the safety setting that determines who gets generated.
Imagen 3 image generation
image = generation_model.generate_images(
prompt=prompt,
number_of_images=1,
aspect_ratio=”1:1″,
safety_filter_level=”block_some”,
person_generation=”allow_all”,
add_watermark=True,
)
What comes next?
Imagen 3 with an allow list is now publicly accessible. Currently, developers at companies with clear use cases get priority access to Imagen 3 on Vertex AI.
