מדריך מלא: איך לאמן מודל FLUX עם הפנים שלכם באתר ASTRIA + קבלו 10$ לחשבון שלכם מתנה!

שחר גולן

16 Sept 202417:30

Summary

TLDRThe video introduces Astrea, a platform for training AI models, specifically focusing on image generation using the advanced 'Flex' model. The presenter demonstrates how users can easily train custom models, such as faces, objects, or styles, using Flex's high-quality image generation and text embedding capabilities. Viewers are guided step-by-step through the process of uploading images, configuring model settings, and generating outputs. The presenter also shares a special offer of $10 worth of credits for first-time users, allowing them to experiment with the platform without needing to provide payment information.

Takeaways

🚀 A new model called 'Flax' was recently launched and has generated a lot of excitement in the field of AI and image creation.
🖼️ Flax allows for high-quality image generation with integrated text and impressive accuracy.
👨‍💻 You can train custom models on Flax, such as training it with the face of a person, an animal, or even a specific logo.
🎨 The website 'Astra' offers a simple, all-in-one platform to train models using Flax, avoiding complex steps required by other sites.
💡 The Israeli entrepreneur behind Astra, Alon Burg, provided credits for users to test the platform, and viewers of the video get $10 in free credits.
🖱️ To claim the free credits, users need to enter the promo code 'Shahar Golan' in their account settings.
🔧 Users can train a model by uploading images (8-16 recommended) and selecting the 'Flax' model for high-quality results.
💻 Training can be done using different speeds (faster or slower), which impacts both the time and quality of the output.
📧 After training the model, users can generate images with specific prompts and trigger the trained features using the designated code.
💰 Creating images costs a few cents, with the credit balance being preserved for future usage.

Q & A

What is the main topic of the video?
-The video discusses the use of a tool called Astria for training AI models using the Flex model, which is known for generating high-quality images and incorporating text.
What makes the Flex model special?
-The Flex model is notable for its ability to generate highly realistic images, even with the inclusion of text, offering a powerful tool for training models on various objects, faces, animals, or logos.
Who is behind the development of the Astria tool?
-Astria was developed by an Israeli entrepreneur named Alon Burg, and the tool allows users to train models more easily than using other platforms.
How can users get free credits to try Astria?
-Users can obtain $10 worth of credits by entering 'Shahar Golan' in the 'community' section of their account on Astria after registering.
What kind of images or objects can be trained using the Astria tool?
-Users can train the model on various objects such as faces, animals (e.g., dogs or cats), logos, or even specific styles, such as a particular drawing or fashion style.
How many images are recommended to train a model effectively on Astria?
-The recommendation is to use between 8 to 16 images when training a model, especially for portraits or objects.
What are 'steps' or 'sfs' mentioned in the video?
-'Steps' or 'sfs' represent the number of processing units used to train the model. For example, a slower but more precise training might require 100 sfs per image, while a faster but lower-quality training uses 27 sfs per image.
How much does it cost to generate images using Astria?
-Generating four images costs around 11 cents, or approximately 10 cents per image, making it affordable for users.
How can users customize the output style when generating images?
-Users can select from various styles, such as 'cinematic,' and adjust image parameters like height, width, and number of images before creating the final output.
What happens to the trained models after 30 days?
-Trained models are deleted after 30 days unless the user opts to renew them by adding a credit card and paying a small fee, such as 50 cents per month, to keep the model active.

Outlines

00:00

🚀 Introduction to the New AI Model Flex

In this paragraph, the speaker introduces a new AI model called 'Flex,' which has garnered significant attention in the field of artificial intelligence and image creation. The model excels at generating high-quality images with embedded text. The speaker explains that users can also train the model using their own images, such as human faces, animals (dogs, cats), or even logos, offering a wide range of creative possibilities. The speaker highlights a specific platform, Astrea, created by an Israeli entrepreneur named Alon Burg, which simplifies the process of training models, compared to other platforms like Papal, which have more complex steps.

05:00

💰 Getting Free Credits on Astrea

The speaker shows how viewers can claim $10 worth of free credits to experiment with the platform Astrea. To do this, users need to log in, access their account by clicking on their email address, and enter the code 'Shahar Golan' in the community section to receive the credits. These credits don't expire, and users can add more later if needed. The speaker then explains how to navigate the platform to begin training a model by clicking on 'Tunes' or following a featured ad for 'Flex Tuning.'

10:04

🖼️ How to Train Your Model on Flex

This paragraph details the step-by-step process for training a model on Flex. The speaker demonstrates by training the model with their own face. They explain how to name the model, select the type of subject (human, animal, object, or style), and choose images for training. The recommended number of images is between 8 and 16 for faces, with examples of different settings and looks provided. The speaker uploads 13 images and chooses the advanced settings, ensuring the Flex model is selected. They also describe the difference between the fast and slow modes of training, where slow mode yields better quality but costs more 'steps' (tokens).

15:06

⏳ Waiting for the Model to Train

The speaker continues by showing how they set up their training session, allocating 500 steps for the 13 images uploaded. After choosing a random keyword trigger ('SKS') to activate the trained model, they explain the system's estimation of 43 minutes for completion. The speaker then explains how to return to the page later once the training is done. They demonstrate inputting a prompt for generating images using the trigger keyword and show how users can choose additional style options for customization.

🔄 Generating Images with the Trained Model

The speaker demonstrates the final steps of generating images after the model has finished training. They show how to input prompts, select image dimensions, and generate multiple images. The speaker shares examples of images generated by their trained model, noting how the AI successfully replicated facial features like a birthmark under their eye. They also mention the cost of generating images, which is about 11 cents for four images, and how credits are deducted for each generation. The paragraph concludes by encouraging viewers to try the process themselves using the free credits offered.

📣 Final Thoughts and Invitation to Join Communities

In this final paragraph, the speaker wraps up the tutorial, reminding viewers that they can claim $10 in free credits and start training their own models without needing to enter credit card details. They also invite viewers to join a WhatsApp group for learning, updates, and discussions about AI and the Astrea platform. Additionally, they mention other groups available on Telegram and provide links to join these communities. The speaker ends by wishing the viewers success and encouraging them to explore the platform further.

Mindmap

Keywords

💡Flax Model

The 'Flax Model' is a powerful AI engine mentioned in the video, designed to generate high-quality images with integrated text. It is highlighted as a key advancement in the field of AI-driven image creation, allowing for realistic image generation from prompts. The video emphasizes its superiority in generating detailed and complex images, like human faces or objects, based on user input.

💡Astrea

Astrea is a platform introduced in the video for training AI models using the Flax engine. It offers an all-in-one solution for users to upload images, train models, and generate new images. The video contrasts it with other platforms like 'Pepel,' suggesting that Astrea simplifies the process of model training with fewer complications and more user-friendly features.

💡Model Training

Model training refers to the process of feeding an AI engine with data (such as images) to create a customized model capable of generating specific outputs. In the video, the presenter explains how to train a model on Astrea, whether for faces, animals, or styles. For example, users can train the model on their own face or style preferences by uploading a set number of images and then generating new images based on those trained models.

💡Lora

Lora is a specific type of model used within the Flax engine that specializes in generating realistic images. The video explains that Lora must be selected during model training on Astrea, as it is the recommended and default option. Lora ensures the generated images maintain high quality and accuracy based on the user’s input data.

💡Steps (SFPs)

Steps, or SFPs, refer to the computational steps needed to generate an image in the Flax model. The video explains that the number of steps impacts the quality and time taken for image generation. For example, faster models use fewer steps (e.g., 27 SFPs per image) but may produce lower quality results, while slower models (100 SFPs) yield higher quality images.

💡Advanced Settings

Advanced settings in Astrea allow users to fine-tune how their AI models are trained. The presenter demonstrates how to access these settings to adjust the type of model (such as Lora), the quality of the images, and even toggle between faster or slower training options. The video emphasizes the importance of these settings for achieving the best results.

💡Trigger Word

The trigger word is a specific keyword or set of letters the user inputs during the model training phase. This word acts as a reference to activate the trained model when generating images. In the video, the presenter uses the trigger word 'sks' to ensure the Flax model applies the trained data when generating images, such as in the example of a person wearing a red hat in the desert.

💡Image Creation

Image creation is the process of generating a new image using the trained model in Astrea. The presenter demonstrates this by using a prompt that describes the desired image (e.g., 'man with red hat in the desert'), and shows how the AI generates images based on this input. The video emphasizes how the platform allows users to create multiple images quickly, each costing a small number of credits.

💡Community Credits

Community credits are a reward system in Astrea, allowing users to earn credits to train models and create images without paying upfront. The video explains that viewers can use the code 'shahar golan' to receive $10 worth of credits, encouraging them to try out the platform for free. These credits are necessary to conduct model training and image generation.

💡Style Training

Style training allows users to train a model on a specific artistic style, like a drawing or painting technique. In the video, the presenter shares an example where they trained the Flax model to replicate a style from a particular series, allowing them to generate images in that same artistic manner. This feature opens up possibilities for more creative and customized outputs.

Highlights

A new model called Flux was launched, creating a big buzz in the AI and image generation field.

Flux offers high-quality image generation with built-in text capabilities.

Users can train models on Flux to generate specific images, such as faces, pets, or logos.

The presenter introduces a website, Astrea, where users can train models more easily compared to other platforms like Pepel.

Astrea is founded by Israeli entrepreneur Alon Borg, who offers free credits to new users.

Steps to receive $10 worth of credits on Astrea are demonstrated in the video.

The presenter shows how to train a model on Flux by uploading images, including 8-16 pictures for faces or objects.

The advanced settings allow users to choose between a faster or slower training model, impacting the cost and quality of the generated images.

Users can set trigger words for their model to activate the specific trained style or image.

The presenter demonstrates the model training with his own face images, showing the process of uploading and labeling.

Once trained, the model can create images based on textual prompts that reference the trained data.

After about 43 minutes, the trained model is ready for use, generating high-quality, realistic images.

Images generated using Flux are realistic, with details like facial features accurately represented.

Each image costs a few cents to generate, and users can create various images based on their trained models.

Users are encouraged to sign up for Astrea, where they can retain models for a small monthly fee to avoid deletion after 30 days.