Best AI Photorealism yet? NEW Model!

In this video, the presenter explores the latest advancements in photorealistic AI, specifically focusing on a new model that has been trained to generate highly realistic images. The video demonstrates the use of this model in conjunction with additional enhancements such as 'detail eyes' and skin texture improvements to create photorealistic portraits. The presenter also discusses common issues with generative AI, like unrealistic skin textures, and how the new model has made significant strides in achieving more natural and authentic results. The video showcases live renders of various scenes, including a portrait of a woman, a man astronaut, and a Viking woman warrior, highlighting the model's ability to produce images that are not only detailed but also closely resemble real-life photographs. The presenter emphasizes the model's potential for professional use and its progress towards achieving a level of realism that was previously challenging for AI.

Takeaways

• 🎨 The video discusses the pursuit of achieving the best photorealistic images using a new AI model with stable diffusion.
• 🚀 The presenter introduces a new model trained specifically on realism to improve photorealism in AI-generated images.
• 👀 The video demonstrates how to address common issues with AI-generated images, such as unrealistic skin texture and eyes.
• 🖼️ The use of 'detail eyes' and 'skin blemishes' prompts are highlighted to enhance the realism of the generated images.
• 🌅 Techniques to achieve a more natural look, including dry skin and visible skin hair, are discussed.
• 📸 The video shows live renders of various scenes, emphasizing the model's ability to produce non-cherry-picked, consistently good images.
• 💎 The presenter shares a preferred model called 'realistic stock photos' for close-up photos of people, which is praised for its plain and realistic look.
• 📂 Instructions are given on how to download and implement the new model and additional lores (details) into the user's AI system.
• 🎭 The video explores different styles, such as 'cinematic' and 'analog film', and their impact on the final image's realism.
• 🛠️ The presenter mentions the use of 'Juggernaut cinematic lora' for a more cinematic vibe and discusses the importance of mixing and matching different models and lores for desired effects.
• ⏱️ The video acknowledges the rapid progress of AI in photorealism, comparing the current model's performance to that of stable Fusion 1.5, which required significant manual in-painting.

Q & A

What is the main focus of the video?

The main focus of the video is to demonstrate how to create photorealistic images using a new generative AI model known as stable diffusion.

What is the name of the new model being discussed in the video?

The new model being discussed is called 'realistic stock photos' and it has been trained specifically on realism.

What are some of the techniques used to enhance the realism of the generated images?

Techniques used to enhance realism include adding details such as dry skin, skin fast, visible skin hair, and skin blemishes to the generated images.

What is the role of the 'detail eyes' model in the process?

The 'detail eyes' model is used to improve the quality and detail of the eyes in the generated images, which is a common area where AI models often struggle.

How does the 'realistic stock photos' model differ from previous models?

The 'realistic stock photos' model is designed to produce images that are more plain and less hyper-realistic, making them appear more like regular stock photos or selfies seen on social media.

What does the video suggest about the progress of AI in photorealism?

The video suggests that AI has made significant strides in photorealism, with the new model producing consistently good images and requiring less post-processing than previous models.

What are some of the challenges faced by AI models in generating realistic skin textures?

AI models often struggle to accurately represent skin texture, with the skin in generated images appearing oily, plastic, or otherwise unrealistic.

How does the video demonstrate the use of different styles and prompts to achieve different looks?

The video shows the process of changing the style and adding prompts such as 'dry skin' and 'skin blemishes' to achieve a more authentic and less perfect look in the generated images.

What is the significance of the 'Juggernaut cinematic' model mentioned in the video?

The 'Juggernaut cinematic' model is used to add a more cinematic vibe to the images, with a color grade that enhances the realism and creates a different aesthetic.

How does the video address the issue of imperfections in AI-generated images?

The video acknowledges that life isn't perfect and neither are images or humans, suggesting that adding imperfections like birthmarks and skin blemishes can make AI-generated images appear more natural and authentic.

What advice does the video give for users looking to work with photorealism in AI?

The video advises users to start with the 'realistic stock photos' model when working with photorealism in AI, as it performs better in producing plain and realistic images.

Outlines

00:00

🎨 Exploring Photorealism with Stable Diffusion

The video script introduces viewers to a new model for generating photorealistic images using stable diffusion. The host shares their enthusiasm for the progress made in photorealism and generative AI, and invites the audience to join them in exploring this technology. The focus is on using a model trained specifically on realism and incorporating additional features like 'detail eyes' to enhance the realism of the generated images. The host also discusses common issues with skin texture in AI-generated images and how the new model addresses these concerns. They showcase live renders of various portraits, highlighting improvements in skin texture and eyes, and provide guidance on how to achieve a more natural and imperfect look to mimic real-life imperfections.

05:01

🖼️ Enhancing Realism with Skin Details and Styles

The second paragraph delves into the specifics of enhancing photorealistic images through the addition of skin details and the use of different styles. The host demonstrates how to transform a portrait into a fashion model on a runway, changing the style from cinematic to analog film for a vintage vibe. They also experiment with rendering a portrait of a Viking woman warrior in a coffee shop, showcasing the flexibility of the model to adapt to various themes and styles. The host emphasizes the importance of imperfections in achieving realism, such as skin blemishes and asymmetry, which are not typically found in AI-generated images. They also discuss the use of different lora models to achieve a cinematic feel or a more plain, realistic look, and share their excitement about the advancements in AI, particularly in comparison to earlier versions of stable diffusion.

Keywords

Photorealism

Photorealism refers to the depiction of images or objects with a high degree of resemblance to real-life subjects. In the context of the video, photorealism is the primary goal when using generative AI and stable diffusion models to create images. The video discusses the progress made in achieving photorealistic results with the new model, emphasizing the importance of lifelike details in the generated images.

Generative AI

Generative AI is a branch of artificial intelligence that focuses on creating new content, such as images, music, or text, that did not exist before. In the video, generative AI is used to generate photorealistic images. The host discusses a new model's capabilities in creating images that closely mimic real-life visuals.

Stable Diffusion

Stable Diffusion is a term used in the video to refer to a specific type of generative AI model that is capable of producing high-quality images. The video explores the advancements in this technology, particularly focusing on a new model that has been trained to enhance photorealism.

Model Training

Model training in the context of AI involves the process of teaching the AI system to perform a specific task, such as image generation, by providing it with a large amount of data to learn from. The video mentions that the new model has been trained specifically on realism, which is crucial for achieving photorealistic images.

Skin Texture

Skin texture is an important aspect of photorealistic image generation, as it contributes to the realism of the human subjects in the images. The video discusses the challenges of achieving realistic skin texture and how the new model addresses this issue, making the skin look less 'oily and plastic' and more natural.

Detail Eyes

Detail Eyes is a term used in the video to describe a specific feature or model component designed to enhance the detail and realism of the eyes in generated images. The video highlights the importance of detailed and realistic eyes in achieving photorealism and how the Detail Eyes model improves upon this aspect.

Stock Photos

Stock photos are pre-taken photographs that can be used for various purposes, such as in advertising, websites, or print media. In the video, the host mentions using a model trained on realistic stock photos, which is intended to produce images that closely resemble professional, close-up photographs of people.

Imperfections

Imperfections refer to the natural irregularities and blemishes found in real-life subjects, such as skin blemishes, visible skin hair, and pores. The video discusses the inclusion of these imperfections in the generated images to enhance their realism, as they contribute to a more authentic and less 'perfect' appearance.

Cinematic

Cinematic, in the context of the video, refers to a style of image rendering that mimics the visual aesthetics of films, often characterized by dramatic lighting and color grading. The host experiments with a cinematic style and compares it with a more vintage or analog look to achieve different effects in the generated images.

Vintage

Vintage, as used in the video, describes a style that emulates the look of older photographs, often associated with a specific era such as the 1970s. The video explores the use of a vintage style to give the generated images a retro or old-fashioned feel, which is contrasted with a more modern cinematic style.

Progression

Progression in the video refers to the ongoing development and improvement of the AI model's capabilities in generating photorealistic images. The host expresses satisfaction with the advancements made, noting that earlier versions of the model struggled to produce realistic images, whereas the new model shows significant improvement.

Highlights

The journey to find the best photorealistic images with stable diffusion is progressing with a new model.

The new model is trained specifically on realism and aims to create photorealistic images.

A dad joke becomes a dad joke when it's fully grown, showcasing the humorous side of AI development.

A portrait of a man astronaut is created, resembling the aesthetic of 'Space Odyssey 2001'.

Live renders are being done to achieve a photorealistic style, focusing on detailed eyes and skin texture.

The model addresses the common issue of skin looking oily and plastic in SDXL images.

Results are promising, with images showing less imperfection and a closer resemblance to realism compared to Stable Fusion 1.5.

The live renders include four non-cherry picked images, demonstrating the model's consistency.

Tips are given on how to achieve more realistic skin by adding details like dry skin, skin fast, and visible skin hair.

A 17th-century portrait of a woman is rendered with added prompts for dry skin and blemishes, enhancing realism.

The model used is 'Realistic Stock Photos', trained with stock photos for close-up photos of people.

The 'Detail Eyes' model is introduced to improve the quality and detail of the eyes in the images.

Images produced are described as 'plain', resembling regular selfies or stock photos, which is a sought-after quality in professional use.

Instructions are provided on how to download and implement the new models and lora into the user's existing setup.

The 'Juggernaut Cinematic' lora is used to add a cinematic vibe to the images, although it's not the primary focus for photorealism.

Experimentation with different styles, such as 'Analog Film', is shown to create a vintage vibe in the images.

The progression of SDXL is praised for its rapid advancement in producing realistic images, with fewer images needing post-processing.

A comparison is made to the base model of Excel, which struggled with producing realistic images, highlighting the model's improvement.