✨ Immerse yourself in the mysterious world of Nymphs, the evolutionary sequel to Dark Pictures!
Hi everyone)))) Friends glad to introduce you a new model “Nymphs”. This model is a kind of reimagining of “Dark Pictures”, which received a lot of positive feedback, for which thank you all very much, I hope you will like this model too.
This time dark pictures acquire new facets of detailing, and each image becomes an embodiment of your most daring fantasies. “Nymphs” retains the magical charm of Dark Pictures, and became more accurate in following your creative ideas.
Unlike “Dark Pictures” I did not build in clip, encoders and wae into the model so that you can use those tools that you like more, so that the model weighs less and can be available for video cards with lower requirements.
⚡ Technical advantages:
- Breathtaking atmospheric images inspired by gothic and mysticism that will transport you to a world of the unknown;
- Clear and detailed results, even at low values, allowing you to capture high quality images;
- Versatile to use: from macabre illustrations to mysterious book covers and mysterious comic books, let your imagination run wild and create unique pieces;
- Perfect for creating breathtaking works of dark fantasy that will leave a lasting impression on all who see them.
⚙️ Recommended Settings:
- The model is based on the original fp8 Flux Dev;
- The FP8 version requires VAE, CLIP and text encoder to use the model, the AiOs and NF4 versions do not need them;
- Image size: 896x1152 and above, for maximum detail;
- Number of steps: 14-30 (20 recommended), for optimal balance between detail and speed;
- CFG Scale: 3.5, for a more pronounced and detailed image;
- Sampler: Euler, FLUX Realistic, DPM++2M, LMS, HEUN, DPM2 and others to create the most realistic and vivid images.
💡 Tips to improve image quality:
- Use extensions for Forge WebUi - A-detailer, Detail Daemon, as well as Hires.Fix , to improve the quality and detail of images.
- Experiment with different settings and parameters to find the perfect balance for your creative goals.
🎭 Let your imagination run wild with “Nymphs” - where darkness becomes art and every detail tells its own story. Also try my loras that improve the quality of the pictures - FLUX Enchanter and A magical dream on a moonlit night.
Create without boundaries! 🖤
If you liked my model, you can thank me on my Boosty blog ☕️🍨
Description
The model is based on the original fp8 Flux Dev;
VAE, CLIP, text encoder not required;
Image size: 896x1152 and above;
Steps: 14-30 (20 recommended);
CFG Scale: 3.5;
Sampler: Euler, FLUX Realistic. etc.
FAQ
Comments (11)
А в чем отличие от предыдущей aio-версии?
это единственная aio, предыдущая просто fp8
Great model! Hoping for a gguf variant in v2, not enough vram to run controlnetUnion and an FP8 model at the same time.
One of the biggest annoyances is the hard inclusion of artist signature/logo/name on 80% of the images.
With prompt you can avoid some, but then the model still generates 80% that have something blurry at the bottom right corner... wtaf.
A quick solution would be to crop the images.
Or do second pass with something intelligent that removes the annoyances. No idea what that could be though.
For the next version of model I suggest to remove the annoyances from the training data.
I don't quite understand what you mean, I work my datasets thoroughly in photoshop, they are 100% free of text and other defects, try to look for repeating text in your prompt or replace it, I also advise you to clear the cache in your browser, it should help, good luck 👍
@NevskiyArt It was a scheduler issue that guaranteed me a lot of signed images. I started using another scheduler and the amount of marked images has reduced drastically.
I really love your two Dark Picture models, this one and the another one. But with this one I have a problem. My images appear with a banding lines. I'm the only one with that problem?
Stripes appear on any FLUX model on weak video cards and with a small amount of RAM, reboot your pc more often and clear the cache in your browser, it should help).
@NevskiyArt I'll try, but it just happens with some models, this one (Not with the Dark Pictures v10) and with one more. I've 12 Gb VRAM and 64 Gb Ram. It's not a lot, but It's enough, I think.
@GodivasRubs In my case when I use Anime art Lora with DPM++ 2S Ancestral the banding lines are well visible, but with other schedulers less.. With larger images and upscaling they become more visible. Some say to use Beta scheduler to avoid banding lines,.
I have 16GB VRAM. No other LoRA produces banding lines for me. May be specifically an issue with the LoRA not designed for the scheduler.
@lyagushka420 I figure out the same issue when I use IPAdapter on Flux, maybe it has relation with that
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.