I recently discovered a new text-to-image AI generator named Leonardo that's free to use, though there are paid pricing tiers available. I don't think most users will need to go the paid route, however, since everyone is given 150 credits each day. Even though several credits are used for each image generation and more for features such as upscaling, I still found I could get a good number of images without exhausting my daily quota of credits. But those who feel the need for a larger number of credits, the ability to keep generated images private, and more control using such features as Alchemy, now in beta, which offers a number of style variations and generally results in higher quality images, may well want to consider upgrading to a paid plan.
The great advantage to Leonardo for those who do not have extensive experience with AI generative models is its ease of use, especially when compared to such apps as Midjourney which require a certain level of expertise. At the most basic level all one need do is choose a model - there are several featured platform models as well as an extensive library of community-created models - that will determine the general look of the generated image(s), choose between Leonardo Style (which will usually result in greater realism) and No Style, and enter a prompt. Leonardo will do the rest. It even has a prompt generator that will improve on a submitted prompt within the image generation window itself.
As mentioned, there are a number of features available to those who want more control over their images. In regard to the image generation itself, one can first of all (at a cost of extra credits) choose the output size and the number of variations. For myself, I've found two variations on each image and the default 768x768 size to be sufficient. If I need a larger size I can always upscale later in Topaz Labs Gigapixel.
Even without the afore-mentioned Alchemy, one can the choose between Leonardo Style and no style at all. Prompt Magic is turned on by default and its guidance set to .4. I haven't had a chance to fully explore this feature myself but it is defined quite simply as follows: "Prompt Magic enables higher prompt adherence and greater deviance from training data." High Contrast is also turned on by default. A nice touch is the ability to enter negative prompts and thereby exclude from generated images items one does not wish to see in them.
Beyond this, there are several features one would not expect to find in a free app. A Guidance Scale, for example, determines how closely a given prompt should be adhered to when generating images. One can also upload images for use in image-to-image prompting while ControlNet "allows you to influence your generations with input images."
One can also edit within the generated image canvas itself, allowing one ""to visually correct and refine details, modify styles, seamlessly extend content, create composite artwork and striking photomontages through inpainting and outpainting."
Finally, and perhaps most importantly, Leonardo offers users the ability to train their own models in only a few clicks.
As far as image quality, Leonardo provides excellent results, though I believe that, unless one has access to Alchemy, Midjourney, Lexica, and the new Stable Diffusion XL may hold a slight advantage. I did try using Crisp HD Upscale on a few photos, but it didn't really seem to appreciably improve the overall image quality.
In summary, I think that once Alchemy comes out of beta, Leonardo will be a viable alternative to Midjourney as far as quality is concerned. It's certainly easier to use - one is no longer tied to the Discord server for one thing - and the pricing on the paid tiers is comparable to what Midjourney charges.
No comments:
Post a Comment