Any plans to release a Prompting guides for Lumina2?

[RamonGuthrie]

Lumina2 offers a uniquely different experience compared to previous models like SD and Flux. Its prompting behavior feels notably distinct, and I’m curious whether there are plans to release detailed prompting guides or tutorials tailored specifically for this model. This would help users better understand and harness its innovative features?

[ChinChyi]

This part of the content will be included in our upcoming technical report as guidance. We're not certain if it will fully meet user needs, so we may need feedback from dedicated users like you when it's released.

[RamonGuthrie]

This part of the content will be included in our upcoming technical report as guidance. We're not certain if it will fully meet user needs, so we may need feedback from dedicated users like you when it's released.

Great look forward to hearing more

[peteromallet]

This part of the content will be included in our upcoming technical report as guidance. We're not certain if it will fully meet user needs, so we may need feedback from dedicated users like you when it's released.

@ChinChyi, a bunch of people have been testing Lumina in our Discord, you're very welcome to join! Here's the link to the channel if you're interested.

[Johnz86]

I made some good examples with comfyui and posted in civitai. What models excel in my opinion are landscapes.
Photographs of people and monsters are average and had issues with irisis and teeth as usual in most models. I noticed that lot of times, the produced photography of people went into semirealistic style similar to dale-3. Negative prompting had not able to remove such effects.
I also tried to generate interiors. The photography quality is average, but usable. It fails mainly on regular patterning and has lot of artifacts for smaller object blending. Its easy to discern that images are generated with artifacts. I tried to improve the quality with different negative prompts and more precise descriptions of subjects. The quality drop is most noticeable on regular tiled patterns of carpets. Artistic expression in prompt make the object blending more sever and the patterns break more the photography feel. What worked best were simpler prompts for interior.

You are an assistant designed to generate superior images with the superior degree of image-text alignment based on textual prompts or user prompts. <Prompt Start > A profesional interior photograph of american livingroom. Warm atmosphere of summer morning. Large LCD in middle, big sofa, modern fireplace, clean environment. White marble floor tiles. Wide angle photograph shot with DSLR lens camera. Photo is from magazine advertisement.

Compared to landscape:

You are an assistant designed to generate superior images with the superior degree of image-text alignment based on textual prompts or user prompts. <Prompt Start> This expansive aerial view captures a majestic mountain range under the surreal atmosphere of dawn through a high-resolution DSLR lens. The sky, alive with stars and clouds, adds to the mystical ambiance as light gradients paint rugged craggy peaks in vibrant hues. Intricate textures and delicate alpine flora dot the landscape, while dynamic shadows dance across the scene. A wide-angle lens captures the vastness of the terrain, ensuring hyper-detailed clarity that vividly portrays every element—from rich cobalt blues of the sky to earthy terracotta accents of the terrain. The natural spatial composition guides the viewer's eye through layered mountain valleys, where crisp reflective streams shimmer in soft pastel dawn light. Ambient luminescence enhances the scene’s mesmerizing qualities, highlighting exquisite details that evoke the essence of a natural painting. Atmospheric perspective adds depth and intrigue, making this 8k photographic masterpiece a captivating exploration of nature’s splendor.

All images use res_multistep sampler and simple scheduler. I used default civitai worfklow, with slight adjustments to number of steps to 36. There is maybe mistake on my side, should be not present, in the prompt. But it did not make much difference in the outcomes.
The worst results for me were for realistic photographs of architecture and cars, which tended to be malformed color blobs, or tended to go to semirealistic style of dale-3.

I will try illustration next, when I have time.

[RamonGuthrie]

I made some good examples with comfyui and posted in civitai. What models excel in my opinion are landscapes. Photographs of people and monsters are average and had issues with irisis and teeth as usual in most models. I noticed that lot of times, the produced photography of people went into semirealistic style similar to dale-3. Negative prompting had not able to remove such effects. I also tried to generate interiors. The photography quality is average, but usable. It fails mainly on regular patterning and has lot of artifacts for smaller object blending. Its easy to discern that images are generated with artifacts. I tried to improve the quality with different negative prompts and more precise descriptions of subjects. The quality drop is most noticeable on regular tiled patterns of carpets. Artistic expression in prompt make the object blending more sever and the patterns break more the photography feel. What worked best were simpler prompts for interior.

You are an assistant designed to generate superior images with the superior degree of image-text alignment based on textual prompts or user prompts. <Prompt Start > A profesional interior photograph of american livingroom. Warm atmosphere of summer morning. Large LCD in middle, big sofa, modern fireplace, clean environment. White marble floor tiles. Wide angle photograph shot with DSLR lens camera. Photo is from magazine advertisement.

Compared to landscape:

You are an assistant designed to generate superior images with the superior degree of image-text alignment based on textual prompts or user prompts. <Prompt Start> This expansive aerial view captures a majestic mountain range under the surreal atmosphere of dawn through a high-resolution DSLR lens. The sky, alive with stars and clouds, adds to the mystical ambiance as light gradients paint rugged craggy peaks in vibrant hues. Intricate textures and delicate alpine flora dot the landscape, while dynamic shadows dance across the scene. A wide-angle lens captures the vastness of the terrain, ensuring hyper-detailed clarity that vividly portrays every element—from rich cobalt blues of the sky to earthy terracotta accents of the terrain. The natural spatial composition guides the viewer's eye through layered mountain valleys, where crisp reflective streams shimmer in soft pastel dawn light. Ambient luminescence enhances the scene’s mesmerizing qualities, highlighting exquisite details that evoke the essence of a natural painting. Atmospheric perspective adds depth and intrigue, making this 8k photographic masterpiece a captivating exploration of nature’s splendor.

All images use res_multistep sampler and simple scheduler. I used default civitai worfklow, with slight adjustments to number of steps to 36. There is maybe mistake on my side, should be not present, in the prompt. But it did not make much difference in the outcomes. The worst results for me were for realistic photographs of architecture and cars, which tended to be malformed color blobs, or tended to go to semirealistic style of dale-3.

I will try illustration next, when I have time.

I don't believe this relates to my first question, what are you asking for? You would be better off starting a new thread!