青禾_Realistic-Reset Fantasy

青禾-Realistic：
The first time I did a realistic large model, and it took a long time, iterated through many versions, finally, the version I'm more satisfied with came out. To complain: realistic is really much harder to control than anime. I've been making changes for over half a month.
【Introduction】：
This is not a specialized model (simply put, it is not born just to draw 1 girl), it adapts to many styles, sceneries, and objects, with a wide range, even your realistic flora training can use it as a base film. Other big shots often use this 1.5 for testing during XL tests, and the feedback is not bad.
【Simple Features】（Read the following for helpful tips）：
The default base film face of this model is European and American, not limited to races. Asian, European, Black, these can still be recognized. The model does not lock the face, so when you don't like the European and American base film face, you can switch to a face model you like, put it in AD Face Edit. I already have a face model ready, but thinking about the locked face, I simply didn't fuse it.
<img src="https://liblibai-online.liblib.cloud/img/820ee8b4e9904f87aa4e525e4d78ec55/92142ba9b87fc33a897624d9a40102ea59df5d5eec1685db41bfe07339bc7848.jpg?x-oss-process=image/resize,w_1290,m_lfit/format,webp" alt="" data-href="https://liblibai-online.liblib.cloud/img/820ee8b4e9904f87aa4e525e4d78ec55/92142ba9b87fc33a897624d9a40102ea59df5d5eec1685db41bfe07339bc7848.jpg?x-oss-process=image/resize,w_1290,m_lfit/format,webp" style=""/>
【It can also recognize some specific styles very well】, for example: retro, fantasy, futuristic, technological, and other similar style words. (I try to include them in the main image examples for reference.)
【It can also express some artistic terms very well】, for example: paper-cut art, etc. (examples in the main image)
【This model is not only good at character portrayal】, it also provides excellent feedback for pencil sketches, watercolors, ink paintings, and, with the correct use of prompts, can even produce a 2D style. I will provide the prompt method at the end.
For example, if it's just pure sci-fi, it might turn out confusing and even abstract, but if you add words like cyberpunk, technology, holograms, futuristic elements, special effects, neon lights, it will provide very good feedback.
【It provides very good feedback for close-ups of characters and photography】. For close-ups, you can add some photography quality terms and realistic style words, which I'll provide at the end.
However, it's not the same as other models that are only good at drawing 1 girl, it can't produce good results if used with just 1 girl, you have to describe everything in detail behind the girl, like clothes, etc. (#^.^#)
The model embeds VAE, so you don't need to add VAE when using it.
That's enough introduction, let's keep it brief.
 
【How to Use - Very Simple】：
You can directly ask GPT for prompts, and it can recognize most prompts in natural language. After all, it's version 1.5. I won't praise it too much, but if used correctly, you will definitely get a satisfying answer.
【Attention with Prompts】：
【First Point】: If you don't need to add quality terms when drawing characters, terms like 'masterpiece,' 'best quality,' etc., are not necessary. Just add negative terms like (worst quality: 1.4), (low quality: 1.4) if needed.
Avoid adding too many negative terms, these two are enough, unless you want to avoid something positive, add negative terms. It's simple. Otherwise, you can add quality terms when drawing things other than characters.
【Second Point】: If you're drawing certain style categories like paper-cut art, simple sketches, abstract lines, etc., you can remove negative terms and leave it blank. Explanation: Because those styles may be classified as low quality, adding negative terms might filter them out. Personally tested.
【Third Point about NSFW】: I won't go into details here. Locally, I don't check, but if you're using it online and find images violating rules, make sure to describe the clothes when drawing people to avoid nudity. If you accidentally missed a violation, you can add 'nipples' backwards.
That's it, just these three points.
【Recommended Parameters】
<img src="https://liblibai-online.liblib.cloud/img/820ee8b4e9904f87aa4e525e4d78ec55/31331f71f78de48ef225c59cc1cc1af68a32f702c36b93754375aa58a2df19dd.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" alt="" data-href="https://liblibai-online.liblib.cloud/img/820ee8b4e9904f87aa4e525e4d78ec55/31331f71f78de48ef225c59cc1cc1af68a32f702c36b93754375aa58a2df19dd.png?x-oss-process=image/resize,w_1290,m_lfit/format,webp" style="width: 722.67px;height: 525.67px;"/>
 
【Some Realistic Flattery Terms to Share】：
High-Quality Photography (best used for products and non-people): realistic, photograph, photorealistic, Commercial photography, e-commerce photography, ultra-high definition, RAW photo,
 
Realistic Style (character portrayal, usually only used for close-ups): photorealistic, 8k, uhd, ultra-sharp focus, highly detailed, internationally-acclaimed photograph, RAW photo,
 
Cinematic Depth of Field (used for characters): cinematic photo, cinematic light, film, depth of field, blurry background, bokeh, gloom,
 
Aesthetic Portrait Soft Light (used for characters): soft light, warm color, light leak, film photography, old-school 35mm film, filmg
 
Light Effects (like the Tyndall Effect): extreme details, 8k, RAW photo, (colorful), highly detailed CG illustration, an extremely delicate, (defuse lighting: 1.2), (ambient lighting: 1.2), (rim lighting: 1.2), best shadow, drop shadow, deep shadow, (Tyndall Effect: 1.2),
 
Anime Drawing Style (used when the realistic model triggers an anime style, both positive and negative): positive: cartoonish, flat illustration, Illustration, painting, anime, comic, negative: ad image, worst quality, bad drawing, (intricate details, messy details, 3d, c4d, realistic, photo: 1.3),
Lastly, hope everyone provides feedback. If you have nice pictures, share them, and forgive any shortcomings!!! Thank you. (#^.^#)

Files with SHA256: 34f9e7362ae526ac6e4513bdb3dca6080b0f2f144a14a2e498c7e3b0758b7555