TL;DR

We’ve added several new image generation models—including Imagen 4 Ultra, GPT-Image-1, and Recraft—while improving the overall reliability and flexibility of our image creation system. Upscaling quality also gets a big upgrade thanks to Topaz. Lastly, we’ve simplified a few things: pricing, naming, and error handling are all more straightforward now.

More Models, More Possibilities

A lot of recent work has focused on expanding the kinds of images you can generate and how those images are processed. We've added support for a handful of new models, each with its own personality and strengths:

• GPT-Image-1: Especially good at sticking closely to complex text prompts.

• Recraft: Offers a more design-focused aesthetic. Good for layout-aware content.

• Photon: Clean, high-resolution results with a natural tone.

• Imagen 4 & Imagen 4 Ultra: These bring top-tier realism and detail, especially helpful for professional use cases or clean visuals.

Each model is now wired into the system with its own cost and supported aspect ratios, so choosing the right one should feel more intuitive.

Upscaling That Respects Your Image

We had a mistep during our Upscaler upgrade, we discovered a persistent issue with the Clarity Upscaler changing things a little too much, especially faces. We've replaced the default with a Topaz-based method that is much more conservative. Faces keep their structure, but the overall resolution is still bumped up with care.

Behind the Scenes: A Cleaned-Up Core

To support all these new models, we had to clean up how things work under the hood. The old image generation logic was getting cluttered, so we refactored it. That mostly means:

• More consistent ways to validate and handle inputs

• Fewer one-off model tweaks—it's easier to add new models now

• Shared structure and clear separation between billing, generation, and image storage

If you're just using the app like usual, you probably won't notice. But as we keep adding new capabilities (like inpainting or future video features), this makes everything easier to maintain and extend.

Face Forge → Forge

We renamed Face Forge to just Forge. It’s a small change, but one that reflects how our thinking about custom models has evolved. The big idea from our observation of the early adopters is to make custom models more flexible so they can be used for niche uses cases. We also made a few interface updates to reflect that and cleaned up old links in the process.

More Useful Error Messages

If something breaks now, you should get a clearer message about what went wrong. Before, errors could be vague or buried in layers of logs. Now, the system looks deeper into responses and gives you something more descriptive—whether you're a user submitting a prompt or a developer building with the API.

What’s Next

The team is already working on a few other things:

• Forge Upgrades: more advanced control and training tools powered by Flux Pro Trainer

• Flux Kontext: a for editing existing images using natural language.

• Improved browse page & onboarding: We want to improve usability, especially for new users.

Wrapping Up

This cycle was about depth and reliability: adding some of the best ai art generator models out there, tightening up our workflow, and making sure the output you get feels more predictable. We’ve also made space for more experiments ahead—custom AI models, smart editing tools, and more community-facing work are all on the table.

If you’ve been generating with us already: thank you. And if you’re just exploring, this is honestly a great time to jump in. Remember you can always follow the live development progress on the public Trello Board.