In the rapidly evolving world of AI image generation, a new player has emerged that's making waves across both professional and hobbyist communities. [HiDream-I1](https://hidream.art), released this month, represents a significant leap forward in what's possible with text-to-image AI technology. While there's no shortage of AI image generators today, HiDream-I1 stands out by addressing many of the frustrations users have experienced with previous tools, all while being completely free and open for anyone to use.

## What Makes HiDream-I1 Special?

### Images That Look Exactly How You Want Them

Anyone who's played with AI image generators knows the feeling: you type in a detailed description, hit generate, and get something... almost right. Maybe the blue car you asked for is actually red, or the "person sitting by a lake" has six fingers and a warped face.

HiDream-I1 excels at what experts call "prompt following" – the ability to actually create what you ask for. In testing, it correctly interpreted text prompts 92.1% of the time, compared to only 81.5% for one of the previous leading open-source models. In practical terms, this means less time spent rewriting prompts and regenerating images to get what you want.

### Quality That Rivals Professional Work

The images produced by HiDream-I1 aren't just accurate – they're beautiful. The technology evaluates images based on how appealing they are to actual humans, and HiDream-I1 scores higher on these metrics than any other freely available AI image generator.

What does this mean in practice? Sharper details, more realistic textures, better lighting and shadows, and fewer of those strange artifacts and distortions that have been the telltale signs of AI-generated art.

### Free and Open for Everyone

Perhaps most importantly, HiDream-I1 is released under what's called the MIT license – one of the most permissive software licenses available. This means:

- Individuals can use it for personal projects without restriction
- Researchers can examine and build upon it
- Businesses can incorporate the images into commercial products without complex legal concerns

This open approach stands in contrast to some popular image generation tools that come with usage restrictions or require subscription fees.

## How Does HiDream-I1 Work?

While the technical details get quite complex, the basic approach is fascinating even for non-engineers.

### A Team of AI Specialists Working Together

Rather than using a single approach to generate images, HiDream-I1 uses what's called a "Mixture of Experts" architecture. Think of it as a team of specialists rather than a single generalist.

Imagine commissioning artwork from a studio where one artist excels at composition, another at color theory, a third at texture details, and so on. By combining their expertise, the final result benefits from all their specialized skills. HiDream-I1 works similarly, with different parts of its system handling different aspects of image creation.

### Understanding Text Through Multiple Lenses

When you type a prompt like "a surrealist painting of a city at night with floating lanterns in the style of Salvador Dalí," HiDream-I1 processes your text through four different systems:

1. One system focuses on broad visual concepts (city, night, lanterns)
2. Another analyzes detailed semantic relationships (how the elements should relate to each other)
3. A third examines linguistic structure (understanding what "surrealist" and "in the style of" modify)
4. The fourth interprets contextual instructions (understanding what a Dalí-esque image should look like)

This multi-layered approach helps HiDream-I1 grasp nuanced requests that previous models often misinterpreted.

## What Can You Create With HiDream-I1?

The system excels at generating images across seven primary style categories:

### Photorealistic

Perfect for product visualizations, architectural renderings, or realistic scenes that could fool the eye. These images feature detailed textures and accurate lighting and shadows.

### Anime/Cartoon

Creates expressive characters with clean lines and vibrant colors, ideal for character concept art, illustrations, or stylized scenes.

### Painterly

Produces images with visible brushstrokes mimicking traditional media like oil or watercolor paintings – great for artistic projects or decorative imagery.

### Concept Art

Generates environmental designs and mood boards that could serve as starting points for games, films, or other creative projects.

### Surrealist

Creates dreamlike compositions with symbolic elements and unexpected juxtapositions, perfect for artistic expression or attention-grabbing marketing visuals.

### Low-Poly

Produces geometric simplifications with clean edges, useful for gaming assets, modern graphic design, or stylized illustrations.

### Pixel Art

Generates retro gaming aesthetics with limited color palettes, appealing to nostalgia or indie game development.

Unlike some earlier models that often blended styles unintentionally, HiDream-I1 maintains distinct characteristics for each style category, giving users more precise control over their creations.

## How Does It Compare to Other AI Image Tools?

### Versus Premium Services (like DALL-E 3)

While proprietary systems like DALL-E 3 still hold some advantages in generating extremely complex scenes, HiDream-I1 closes the gap significantly:

- Human figures look more anatomically correct (89% accuracy versus DALL-E 3's 92%)
- Text within images is more legible (76% success rate versus 83%)
- Perspectives and viewpoints appear more consistent (91% accuracy versus 94%)

Considering that HiDream-I1 is free and open-source while DALL-E 3 requires paid credits, these comparisons are impressive. HiDream-I1 actually outperforms DALL-E 3 in depicting diverse cultural elements, like non-Western architectural styles and traditional clothing from around the world.

### Versus Other Open-Source Options

Compared to previous open-source leader SDXL, HiDream-I1 offers several improvements:

- Better adherence to prompts (reducing the problem where requested objects blend together or have incorrect attributes)
- Built-in high-resolution capability without quality loss
- More efficient resource usage in some configurations

However, SDXL still benefits from its established community, with thousands more custom adaptations available and better support for video generation.

## Current Limitations

No technology is perfect, and understanding HiDream-I1's limitations is important:

### Hardware Requirements

The full version of HiDream-I1 requires substantial computing power – specifically, a graphics card with at least 24GB of video memory. This puts it out of reach for casual users with average consumer computers. However, optimized variants are available that can run on more modest hardware.

### Still Learning Some Skills

HiDream-I1 has room for improvement in several areas:

- Complex training: Its sophisticated architecture makes it harder for users to customize or train for specific domains
- Cultural balance: It still shows some bias toward Western imagery in default outputs
- Video limitations: Unlike some competitors, it doesn't yet have built-in tools for creating consistent animations

## What's Next for HiDream?

The developers have released three variants to suit different needs:

1. **HiDream-I1-Full**: The complete model for maximum quality (but requiring powerful hardware)
2. **HiDream-I1-Dev**: A balanced version for developers and serious users
3. **HiDream-I1-Fast**: A streamlined version prioritizing speed for interactive use

Looking ahead, HiDream plans to launch an API service later this year, making the technology accessible through cloud services for those without powerful local hardware. Early adopters have already created specialized versions for medical imaging and product design, showing the model's flexibility.

## Why HiDream-I1 Matters

The release of HiDream-I1 represents an important step in democratizing high-quality AI image generation. By combining advanced capabilities with an open-source approach, it puts powerful creative tools in the hands of more people – from individual artists to businesses to researchers.

While it has some limitations and won't replace human artists (nor should it), HiDream-I1 offers an impressive new option in the growing ecosystem of AI creativity tools. Whether you're looking to generate concept art, visualize product ideas, or simply experiment with a cutting-edge technology, HiDream-I1 deserves attention.

The model can be downloaded from [HiDream's website](https://hidream.art) or through popular AI toolkits like ComfyUI and Automatic1111's web interface. As the community continues to build around this technology, we can expect to see even more creative applications emerge.

Meet HiDream-I1: The Open-Source AI That's Changing Digital Art Creation

Great news for all FLUX enthusiasts! BlackForestLabs has just released an amazing new suite of tools that will give you even more control over your AI image creation process. Let's dive into what's new and what it means for your creative projects!

## What's New?

BlackForestLabs has introduced four powerful new tools that work alongside the FLUX.1 image generator you already know and love. Think of these as special add-ons that give you more precise control over your creations. Here's what each tool can do:

### 1. FLUX.1 Fill: The Magic Eraser and Canvas Expander
Ever wished you could change just part of an image or extend it beyond its borders? That's exactly what FLUX.1 Fill does! It comes with two amazing features:
#### Inpainting
Think of it as a super-smart eraser tool. You can select any part of an image and replace it with something new while keeping everything else perfectly intact.

![img](https://lh7-rt.googleusercontent.com/docsz/AD_4nXfM84Nir6HIU1xcJeNOolz7ZH7SIYTBudA0jeM2EQJb0LHoJMOZR17PAwf43peoFydja_z_j83LAYVMRL5gry5QF-q-Cm-SQG_iwhJcH7m8Sg1yXrQTi2JYjJQEQfLm5F30_fCnYQ?key=Ult89JQdBB7CX3l5ClfyS9B_)

#### Outpainting
Want your image to be bigger? This tool lets you expand your image beyond its original boundaries, seamlessly adding new content that matches your original style.

![img](https://lh7-rt.googleusercontent.com/docsz/AD_4nXdN_mwhw6xp3EmgDgHqo5zchmEY0TaRznhrwyaHn8Qxz5kvdA5q7OnKsoeeyPWMWUTXD1DfhJ-6X1hDGMDqxR1nspxBNB8pCjb4IKLEMkdUKBgn-kaTOXx6rgWfZk-HgkjaFvjBsw?key=Ult89JQdBB7CX3l5ClfyS9B_)


### 2. FLUX.1 Depth: Master of Perspective
This tool understands the 3D structure of your images. It's like having an art assistant who perfectly understands depth and perspective, ensuring that when you make changes to an image, everything looks naturally three-dimensional and properly positioned.

![img](https://lh7-rt.googleusercontent.com/docsz/AD_4nXe065BZ3q1xRcnpFUxUQ7YTkjOrery_E2Mgo94buBKFKlV8w8p0CCYyB89Ya1A_ePiH1i9KxjzE6bLnoFf1nasWzFH-dIhChPv5jdig5alqBdeUPeS1RBRMyPFEx2qCwA7AVqHepg?key=Ult89JQdBB7CX3l5ClfyS9B_)

### 3. FLUX.1 Canny: The Detail Keeper
When you want to change an image while keeping its basic structure intact, FLUX.1 Canny is your friend. It's particularly great when you want to change the style or appearance of something while maintaining its original shape and form.

![img](https://lh7-rt.googleusercontent.com/docsz/AD_4nXd9F-aIYKft6az72jM9X1DZl7WNkxNxLepuUe9rGhNAZCPVGIWN8RR10oq-CgsYGkN3tcuIUnGAGubHj6aPsSgb8547iOXHB_9s4J_o-xITJ0P90f2Z4Hlu0L6hkAJ4jK2TUZk9fg?key=Ult89JQdBB7CX3l5ClfyS9B_)

### 4. FLUX.1 Redux: The Style Mixer
This is your tool for creating variations of existing images or mixing different styles. Want to take an image and see different versions of it? Or maybe combine the style of one image with the content of another? FLUX.1 Redux makes this possible!

![img](https://blackforestlabs.ai/wp-content/uploads/2024/11/image298-3-7-5-1024x393.png)

## Why This Matters for You

These new tools represent a huge leap forward in what you can do with AI image generation. Whether you're:
- Editing existing photos
- Creating new artwork
- Expanding your compositions
- Experimenting with different styles

You now have more precise control over your creative process than ever before.

## Real-World Applications

These tools excel in various practical scenarios:
1. **Photo Editing**:
   - Removing unwanted elements with Fill
   - Extending photo compositions with outpainting
   - Maintaining structural integrity during edits

2. **Digital Art**:
   - Style transfer while preserving depth and structure
   - Creating variations of existing artwork
   - Expanding canvas size with consistent style

3. **Design Work**:
   - Precise control over image modifications
   - Maintaining perspective in architectural visualizations
   - Creating multiple variations of design concepts


## Where to Try These Tools

We are working hard to bring these new tools to FluxPro.art but in the meantime, you can access these new features through several platforms:
- Directly through the BFL API (for pro users)
- Through partner platforms like fal.ai, Replicate, Together.ai, Freepik, and krea.ai
- For developers and researchers, open-access versions are available on Hugging Face

## The Bottom Line

This update makes FLUX more powerful and versatile than ever before. The new tools give you unprecedented control over your AI-generated images, whether you're touching up existing work or creating something entirely new.

We can't wait to see what amazing creations our FluxPro.art community will make with these new tools! Share your experiences and creations with fellow community members, and let's explore these new possibilities together.

Stay creative! 🎨✨

*Note: This post covers the major features of the new release. For technical details and specific implementation information, you can refer to the official BlackForestLabs documentation.*

Flux Tools: More Creative Control

Great news, the latest Flux 1.1 Pro Ultra model from BlackForestLabs  is now available on FluxPro.art 

## What's New?

The standout feature is the ability to create images at 4x higher resolution than before. This means your AI-generated artwork can now reach up to 4 megapixels while maintaining incredible detail and quality. And the best part? These high-res masterpieces are generated in just 10 seconds!

## Ultra Model
Want to create larger, more detailed images? Ultra mode delivers exactly that, without compromising on prompt accuracy or generation speed. In fact, it's 2.5x faster than similar high-resolution options out there.

![img](https://pbs.twimg.com/media/Gbul1WvXEBUdeEk?format=jpg&name=4096x4096)

### Raw Mode
For those seeking a more authentic feel, the new raw mode creates images that look more natural and less artificial. It's particularly great for:
- Candid-style photography
- Diverse human subjects
- Natural landscape photography


![img](https://pbs.twimg.com/media/GbupU5wWEAgkWhr?format=jpg&name=4096x4096)

## How to Access

These new features are available to all users on paid plans. Each generation using the new model costs 8 credits. Simply select Flux 1.1 Pro Ultra from the model options when creating your next piece!

## Ready to Try It?

Jump in and start creating higher-resolution artwork today! We can't wait to see what you'll make with these new capabilities. As always, feel free to share your creations with the community and let us know what you think of the new model.

Happy creating!

*The FluxPro.art Team*

Introducing Flux [pro] 1.1 Ultra

Dear valued users,

For the past two months, we've had the privilege of offering our FluxPro.art to you free of charge. This period has been an incredible journey of growth and learning. As we evolve, we're announcing significant changes to our service structure to ensure sustainability and continued improvement.

## What's Changing?

1. **Access to All Flux Models**: Users can now generate images using all models of Flux: Pro, Dev, and Schnell.

2. **New Queue System**: We're implementing a queue system to manage traffic more efficiently. Requests will be prioritized based on user tiers.

3. **New Credit System**: We're moving to a monthly credit system. Each model will consume a different number of credits per generation.

4. **Paid Plans**: Subscribers will receive a set number of credits each month and enjoy higher priority in the queue.

5. **Free Tier**: Free accounts will receive 10 credits that refresh every day.



## Why the change?
1. **Sustainability**: To continue providing and improving our service, we need to cover operational costs and invest in future developments.
2. **Enhanced Features**: With a sustainable revenue stream, we can dedicate more resources to developing new features and refining existing ones.
3. **Fairness**: The new queue system ensures that paying users receive reliable, high-priority service.
4. **Continued Free Access**: We're committed to maintaining a free tier, allowing users to continue exploring FluxPro.art.

## What this means for you

### For All Users:
- You'll have access to all Flux models: Pro, Dev, and Schnell.
- Each model consumes a different number of credits:
  - **Flux Pro 1.0**: 6 credits per generation (highest quality, most detailed outputs)
  - **Flux Pro 1.1**: 5 credits per generation (highest quality, and speed)
  - **Flux Dev**: 3 credits per generation (balanced between quality and speed)
  - **Flux Schnell**: 1 credits per generation (fastest generation, suitable for quick iterations)

### For Paid Subscribers:
- You'll receive a set number of credits each month based on your subscription tier.
- Your requests will be prioritized in the queue, ensuring faster and more reliable generation.

### For Free Users:
 - You'll receive 10 credits every day at no cost.
 - These credits can be used across all Flux models, but with some limitations:
   - Free tier generations are processed at a lower priority in the queue.
   - Due to the lower priority, free tier generations may fail more frequently during high-traffic periods.
   - Failed generations will not consume credits.
- You can always choose to upgrade to a paid plan for higher priority and more reliable generations.

## Important Note for Free Tier Users
While we're excited to continue offering free access to our service, please be aware that the free tier is subject to resource availability. During peak times, you may experience longer wait times or more frequent generation failures. We encourage users who require consistent, high-priority access to consider our paid plans.

We want to express our heartfelt gratitude for your support. Your feedback and creations have been invaluable in shaping our service. We're committed to providing a top-notch AI image generation platform and we're excited about the future innovations these changes will enable.


Thank you for being part of our community. As we enter this next chapter, and we can't wait to see the incredible things you will create on FluxPro.art

Flux Models, New Credit System and Paid Plans

# Training Your Own Flux LoRA Model for Realistic AI Image Generation

Flux is one of the most advanced AI image generation models available today, capable of producing highly realistic results on par with Midjourney. A powerful new feature allows you to train your own likeness into the Flux model using a technique called LoRA (Low-Rank Adaptation). This enables generating creative AI images of yourself in any scenario, such as appearing as a wizard, astronaut, or superhero.

There are currently three main ways to train a custom Flux LoRA model:

1. Training on FluxPro.art
2. Using the OstrisAI toolkit on Google Colab
3. Through the Fal.ai platform
4. Via the Replicate.com

Let's dive into the details of each method, along with some tips and tricks for optimal results.

## 1. FluxPro.art Custom LoRa training

Compared to other options below, FluxPro.art makes it extremely easy to train custom models without technical expertise.

1. Go to [https://fluxpro.art/models](https://fluxpro.art/models) and click "Train new model"
2. Provide a name for your model and select the subject type you wish to train (options include Man, Woman, Product, Style).
3. Upload 10-20 High-quality example images
4. Our system will train a custom model in just 20-40 minutes, saving you time and effort.
5. Once training is complete, you can click the "Generate image" button on the model page to create stunning, personalized images that capture your unique essence and style.

## 2. Training Flux LoRA with OstrisAI Toolkit on Google Colab

The OstrisAI toolkit provides a script for training and inference of Flux LoRA models that is used by both Fal.ai and Replicate under the hood. You can run it directly from Google Colab using this notebook: [https://colab.research.google.com/drive/1r09aImgL1YhQsJgsLWnb67-bjTV88-W0?usp=sharing](https://colab.research.google.com/drive/1r09aImgL1YhQsJgsLWnb67-bjTV88-W0?usp=sharing)

### Requirements

To train a Flux LoRA model, you need a GPU with at least 24GB of VRAM. If the GPU is also controlling your monitors, set `low_vram: true` in the config file under `model:` to quantize the model on CPU. The toolkit has been tested on Linux, with some reports of bugs on Windows.

### Model License

Currently, training only works with the FLUX.1-dev model, which has a non-commercial license. You must accept the license on Hugging Face before use. Here are the steps:

1. Sign into Hugging Face and accept the model access for [black-forest-labs/FLUX.1-dev](https://huggingface.co/black-forest-labs/FLUX.1-dev)
2. Get a READ API key from Hugging Face
3. Place the key in a file named `.env` in the root folder like: `HF_TOKEN=your_key_here`  

### Training Steps

1. Copy the example config file `config/examples/train_lora_flux_24gb.yaml` to the `config` folder and rename it
2. Edit the config file following the comments
3. Run the training with `python3 run.py config/your_config_file.yml`

A folder will be created with checkpoints and generated images. You can stop and resume training anytime.

### Dataset Preparation 

Your dataset should be a folder of images (jpg, jpeg, png) and matching `.txt` caption files, e.g. `image1.jpg` and `image1.txt`. The word `[trigger]` in captions will be replaced by the `trigger_word` from your config. Images will be automatically resized and cropped.

## 3. Training Flux LoRA on Fal.ai

Fal.ai provides a simple interface for training Flux LoRA models at [https://fal.ai/models/fal-ai/flux-lora-general-training](https://fal.ai/models/fal-ai/flux-lora-general-training). The process is:

1. Upload 12-15 images of yourself
2. Set training steps to 1,000 and specify a trigger word 
3. Start training (takes 25-30 minutes, costs ~$5)
4. When complete, copy the URL of the generated `.safetensors` file

To generate images with your trained LoRA:

1. Go to [https://fal.ai/models/fal-ai/flux-general](https://fal.ai/models/fal-ai/flux-general)
2. Paste the `.safetensors` URL in the "LoRAs -> Path" box 
3. Use your trigger word in the prompt

Each image generation costs $0.075. You'll need a Fal.ai account with credits.

## 4. Training Flux LoRA on Replicate

The Replicate by Luca Taco provides a simple interface for training Flux LoRA at [https://replicate.com/lucataco/ai-toolkit/train](https://replicate.com/lucataco/ai-toolkit/train). Here are the steps:

1. Gather 12-20 high-quality photos of your face
2. Rename them as "a photo of {your name} 1.jpg", "a photo of {your name} 2.jpg", etc.
3. Zip the photos into a single file
4. Create an account on Replicate.com
5. Go to the "Luca Taco AI Toolkit" model page
6. Under "Train", create a new model and give it a name 
7. Upload your zip file of photos
8. Set training steps to 1000
9. Obtain a Hugging Face token and paste it in
10. Click "Create Training" (takes ~25 minutes, costs ~$2.10)

Once training completes, use the "Flux Dev LoRA" model on Replicate to generate images:

1. Go to [https://replicate.com/lucataco/flux-dev-lora](https://replicate.com/lucataco/flux-dev-lora)
2. Provide a prompt like "{your name} as a wizard"
3. Set the LoRA scale to 1

The creator of the Replicate has provided a one-time $10 coupon code to cover initial training and generation costs.

## Tips and Tricks for Optimal Results

Through experimentation, the following tips have been found to improve Flux LoRA training results:

- **Avoid mixing data types:** Combining realistic photos with cartoon/anime style training images tends to produce plastic-looking, uncanny results. Stick to one consistent image style in your dataset.

- **Use natural language captions:** Comma-separated tags (like those used on Danbooru) are less effective than full, narrative-style captions written in natural language. Describe the images as you would explain them to another person.

- **Keep datasets small:** Surprisingly, smaller datasets of 20-30 images provide more flexible prompting compared to larger sets. Adding just 15 more images drastically reduced the prompting flexibility in one test.

- **Aim for 20 images at 1000 steps:** Excellent likeness capture has been achieved using the Kohya trainer with a dataset of 20 well-chosen images trained for 1000 steps and 1 epoch. This seems to be a sweet spot.

- **Train at 512x512 resolution:** While higher resolutions like 1024x1024 are often used, training at 512x512 produces great results in much less time and at lower cost. The model is able to handle different resolutions regardless.

- **Put trigger/name first in prompt:** When generating images, place your trigger word or name token at the very start of the prompt for best results.

- **Use AI prompt generators:** Tools like Claude can help optimize your prompts to get more aesthetic, creative results from your model.

- **Animate with Runway Gen-3:** For the ultimate output, consider animating the generated images using Runway's Gen-3 model to bring them to life.



With just a dozen photos and 30 minutes of setup using any of these three methods, you can harness the power of Flux to realistically insert yourself into any imagined scenario. By following the tips above, you can achieve impressive results rivaling the best AI image generation available today. The LoRA technique opens up a world of creative possibilities for generating high-fidelity, personalized AI images.


Blog

Meet HiDream-I1: The Open-Source AI That's Changing Digital Art Creation

Flux Tools: More Creative Control

Introducing Flux [pro] 1.1 Ultra

Flux Models, New Credit System and Paid Plans

Training Your Own Flux LoRA Model

What is FLUX.1 [schnell]?

What is FLUX.1 [dev]?