The model inherits the license from Dev. The settings set in workflow are optimal. LoRA on generation acceleration is applied, sometimes 12 steps are enough for acceptable result. CFG scaling was shifted from 1.0 to 1.75. Two methods of de-distillation are applied simultaneously, so you can use standard workflow and write negative prompt without “Flux Guidance” node (you can use it too, but the result will be very different, besides it is difficult to find the right value because of CFG scaling shift).
Модель наследует лицензию от Dev. Выставленные в воркфлоу настройки являются оптимальными. Применена LoRA на ускорение генерации, иногда 12 шагов достаточно для приемлемого результата. Масштабирование CFG было сдвинуто с 1.0 на 1.75. Применено одновременно два способа де-дистиляции, поэтому можно пользоваться стандартными воркфлоу и писать негативный промпт без ноды "Flux Guidance" (можно и с ней, но результат будет сильно отличаться, к тому же сложно найти правильное значение из-за сдвига масштабирования CFG).
Description
FAQ
Comments (13)
Thank you
Impressive skin texture rendering and exceptional Lora compatibility!
I found your model quite amazing. It gives very good result.
P.S. - Thanks for the workflow, buddy
Doesn't work. I got an error trying run the model.
Show me the error
Thanks, is it possible to add the diffusers files to huggingface who be able to train lora's on your model ?
I don't have them, I've seen a few conversion scripts, maybe some of them work correctly and it might be interesting for someone to check it out
@Z_phyr thanks!
i want to know wich sampler and Schedule type is the best for your checkpoint?
I'm thinking deis or ipndm, beta
Where do i need to put the model after i downloaded it?
ComfyUI\models\unet and ComfyUI\models\text_encoders
@Z_phyr thank you for the fast answer
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.







