Developments around AI image generation continue to advance. We have mobile apps, depth effects, video and animation, but we must not forget interfaces for local model execution. In the specific case of Stable Diffusion, the solution of Automatic1111 is the most compatible and flexible, but if you prefer an intuitive and easy-to-use environment with excellent features to expand images beyond their original size, you need to look closely at InvokeAI.
There are at least three paths to generate images with artificial intelligence. On one hand we have DALL-E, with Craiyon as one of its best free online options. Then Midjourney appears, extremely popular on social networks, but it requires Discord and the purchase of tokens. Finally, we come to Stable Diffusion, which besides offering a free online mode, also enables its offline execution, with a graphics card.
In the early days, everything worked in the system console, but Automatic1111's project gave rise to a web interface for Stable Diffusion with a high level of compatibility and flexibility. However, we must also recognize that the state of Auto1111 is a bit chaotic, with more than 1,300 issues open on GitHub, and documentation that is not entirely optimal. Now, if you have enough firepower to run Stable Diffusion offline and you're looking for an alternative to Automatic1111's interface, all arrows point to InvokeAI.
InvokeAI: One of the Best Web Interfaces for Stable Diffusion
One of the most positive aspects of InvokeAI is its installation. The main requirements are three: the installation of Python (3.10.6 is a good option since it maintains compatibility with Auto1111), the application of a registry patch that enables longer unit paths, and about 25 GB of disk space. A double click on install.bat will start the process itself, but it is not 100% automatic. InvokeAI will ask for the destination of the files, and the models to install (the recommended list is a good starting point). Obviously, this means InvokeAI will download very large files, so be patient.
By far, the most powerful feature of InvokeAI is the Unified Canvas, which combines functions of inpainting, outpainting, text-to-image, and image-to-image. Both inpainting (modify or regenerate specific parts of an image) and outpainting (expand an image with additional details) leave much to be desired in the Automatic1111 interface, but it is clear that InvokeAI has hit the mark with its latest edition.
Despite these impressive functions, InvokeAI has limitations that we must highlight. The first is the number of tokens (in other words, if our prompt is too long, InvokeAI ignores the excess), but the most important is its lack of efficiency in VRAM usage, to such an extent that I recommend ignoring InvokeAI completely if the graphics card has 6 GB of VRAM or less available. In that case, the best option is to return to Automatic1111 and activate the commands --medvram or --lowvram, which work very well (beyond the reduction in performance).
Official site: Click here.
