Atmos
Semi-realistic model with cinematic feel.
If you like the model, consider giving it a thumbs up, appreciate you checking it out..
v1.1 Improvements
Fixed issues with head shapes and large hands
More contrast, color and background detail
Sharper details
Special thanks to @Sexiam
v1.1 examples are generated at 832 x 1216 and upscaled with hires fix: 2x scale, 15 steps (in Forge), 0.25-0.30 denoise
Recommended settings
Sampler: Euler a / DPM++ 2M Karras / DPM++ SDE Karras
Recommended CFG: 4-6, I usually use 5
Steps: 30-40 (35 is my go to)
VAE included
Quality tags
Your prompt...
masterpiece, best quality, amazing qualityFor more lifelike results add after your main prompt:
masterpiece, best quality, amazing quality, realistic, fine details, incredibly lifelikeNegative tags:
worst quality, bad quality, worst detail, bad anatomyAdd bad hands, missing finger/extra finger if you're getting bad results with hands
Prompting recommendations
Avoid using too many strong negative and positive quality tags. Using too many quality tags will often reduce the quality of your image. Unless you know what you're doing and use them with a very specific goal in mind.
By design simple prompts will produce simple (in other word - boring) results. Use specific tags to enhance your image:
For a subject: pose, mood, actions, limb positions
Camera angle - low angle or from below, high angle or from above, dynamic angle, dutch angle, pov, fish eye perspective and so on.
Framing: upper body focus, close up, focus on body part, out of frame, crooped, etc.
The model was built with focus on lighting, use it to your advantage with tags like: cinematic, atmospheric, dreamy, ethereal, time of day lighting (i.e. morning lighting), dappled light, light rays. Haze, fog, mist, dust and smoke will also add texture to your image.
Add depth to your image with clever use of particles or specs of dust, or other elements like flying leaves, petals and so on to create additional layers.
Using color tags can dramatically change the feel of your image - from cold and warm tones to using a specific color palette, for example pink, teal or red, black, white.
N.B. v1 is somewhat biased towards generating females, especially when using tags that are often assosiated with female or non-masculine characters like long or wavy hair, eyeshadow, specific clothing pieces, etc - even it is specifically prompted man, 1boy or male. This is something I will try to fix in future releases.
Description
FAQ
Comments (8)
Man. This peace is one of precise masterpieces. Great prompt adherence. Perfect proportions, and it makes the best backgrounds. Excellent work! Thank you! You're a true artist!
Thank you, it makes me really happy that you have good experience with the model
This model is very interesting. It seems to handle portraits with ease while adding subtle details to the scene. When pushed to creatures it roars with imagination. Something magical here!
Sounds interesting, I rarely generate creatures, let's call it a happy accident :P But something to keep in mind for future updates.
I could tell a real difference with this merge. I didn't have to fight the model for details and rarely ran into errors with props or background.
I did see what you mean with it having a hard time with younger men. One trick to add male anatomy and body types to a merge is to use Yaoi loras. They tend to have a wider range of body types and expressions.
Having the male LoRA also seems to mature the women and give them fuller bodies as a side effect.
Overall very nice merge, excellent colors, and a perfect blend of realism and style.
Thank you for the advice, I'll test it out. I have big plans for v2 but it's going to be a ton of work, so anything that would make it easier helps.
@fotonhugger v.1.0 is a very balanced and quite diverse blend. A wonderful and highly versatile model.
v.1.1 shifted toward significantly more contrasting tones and noticeably lost the smoothness of the lighting and tonal transitions. Well, you've reflected this in the description of the changes. Personally, I wouldn't want this to continue in version 2.0.
This is a very noticeable trend: by making further changes to the original model, most authors are creating more rigid and narrowing boundaries of possibilities; and in this context, individual improved parameters make it increasingly dull. I sincerely hope you avoid this trend.
And thank you very much for the wonderful model!
@windlike_gust Yes, after some cleaning up I was testing various merges that continued more cinematic look of 1.0 but they had problems in other areas that I couldn't resolve quickly. I decided to publush 1.1 as is, but for 2.0 planning to go back to the original feel. I have big plans for 2.0 and not sure how long it will take with my current workload. So it was a compromise. Thank you for the feedback!