Today I would like to introduce to you a SDXL-ControlNet, a universal control model: Anytest. It is very easy to operate and does not require a preprocessor. You can use it directly.
The basic functions include generating images based on line drawings and transferring the style of images. There is nothing much to say about this.
Anytest has several interesting uses, and can generate good quality images for compositions with unclear outlines.
Not only that, Anytest can also help you improve incomplete sketches, including re-creation based on pictures.
You may not feel much after hearing this, so let’s just look at the results!
We are still ComfyUI The workflow is the basic ControlNet workflow.
The author provides many Anytest models, and recommends that we use the two models in the red box because these two models have good compatibility.
But since there are V3 and V4, theoretically V4 should be better, but it cannot be so absolute!
Let’s compare these two models!
However, these two models are a bit large. If you don’t have enough video memory, you can download the dim256 version. You can see that there are am and pn in the model names. am represents the large anime model, and pn represents the large model based on the Pony model. You can download it as needed.
The model download link and network disk will be placed at the end of the article.
OK, now that we have introduced the model, let’s take a look at its specific effects.
Let’s start with a simple example of generating an image based on a drawing. Both models accurately recognized the line drawing and generated an image. I personally think that the output effect of v4 is better.
Prompt words: analog film photo One girl, casual, faded film, desaturated, 35mm photo, grainy, vignette, vintage, Kodachrome, Lomography, stained, highly detailed, found footage
Generate images based on pictures. This time, I feel that the images generated by the two models are not very different, and the reproduction of the characters is perfect. Of course, we can also use different Lora models for style transfer.
Prompt words: (masterpiece), (best quality), (ultra-detailed), One girl, ahanfu, illustration, disheveled hair, detailed eyes, perfect composition, moist skin, intricate details, earrings, by wlop
Now that we have introduced the basic functions, let’s look at something different.
Even with very rough outlines of characters, Anytest can draw characters and scenes very well. However, it should be noted that the details of the characters and scenes in the picture should be reflected in the prompt words as much as possible, otherwise the effect will not be very good!
The output effects of different prompt words are quite different. If the effect is not good, you can try to adjust the prompt words.
Prompt words: (masterpiece), (best quality), (ultra-detailed), 1 girl, blouse, stand on the bridge and watch the fireworks from far away., illustration, disheveled hair, detailed eyes, perfect composition, moist skin, intricate details, earrings, by wlop
There are also multi-person ones like this.
Prompt words: (masterpiece), (best quality), (ultra-detailed), 2girl,summer_dress,, illustration, disheveled hair, detailed eyes, perfect composition, moist skin, intricate details, earrings, by wlop
Even if you upload an incomplete line drawing, the line drawing can be automatically completed and the image can be generated.
Prompt words: High Fashion, 1girl, , dynamic, dramatic, haute couture, elegant, ornate clothing, High Fashion
There is another very interesting usage.
We find a sketch we like and draw a few strokes on it. For example, the author below drew a few thick lines on the basic sketch on the left picture, and finally integrated them into the picture perfectly through Anytest.
Let's use this picture to try out the effect. Let me say here that the newly added lines can be drawn with the normal brush tool, not the mask!
Prompt words: art illustrating insane amounts of raging elemental energy turning into 1girl, avatar of elements. magical surrealism, wizardry. best quality, high resolution
The effect is still very good.
Not only can you do secondary creation on sketches, you can also do secondary creation on pictures and generate new pictures.
Prompt words: surrealist art 1girl, Scarves. dreamlike, mysterious, provocative, symbolic, intricate, detailed
If you want to convert anime characters or line-drawn characters into real people, you can just find a large model that supports real portraits. However, V3 still looks a bit like anime style, while the real portrait style of V4 is more obvious. From this point of view, V4 has better generalization.
Prompt words: glamorous photo 1girl, best quality, masterpiece, realistic, high fashion, luxurious, extravagant, stylish, sensual, opulent, elegance, stunning beauty, professional, high contrast, detailed
Okay, that’s all for today’s sharing. If you are interested, go and try it!
Model download address:
https://huggingface.co/2vXpSwA7/iroiro-lora/tree/main/test_controlnet2
Network disk link:
https://pan.quark.cn/s/ebd83f47edcb