Workflow plus custom node for Krea 2 NSFW.

Credits: https://github.com/nova452/ComfyUI-ConditioningKrea2Rebalance

To activate the custom node just copy ComfyUI-ConditioningKrea2Rebalance folder (included in the zip) to your ...\ComfyUI\custom_nodes\ folder and restart ComfyUI.

Change the resolution if it doesn't work for some specific prompt and try again. The NSFW training data is in the model out of the box (even very hardcore), just try different seeds/resolutions/prompts and it should work about half of the time. You could also experiment with the "multiplier" option - anything from 0.80 to 20 (2 or 4 is ok). Higher resolutions (like 1344x2048, 1920x2048, 1536x2560, 1792x3072) seems to require stronger multiplier (2.75 or more) - you could fix a seed and find out what is the minimal value that works for a specific resolution. Haven't done much testing, so I'm not even sure if the resolution and the multiplier are connected, but this is what I get.

You could also add to the workflow any NSFW Krea 2 LoRa from this site (like this one, check the list below) and set both the LoRa strength and the multiplier to 0.45 or 0.5 (or 0.60 + 0.40 or vice versa, or slightly higher if it fails) to get some very exciting results. Neither a LoRa, nor the conditioning node would normally work separately at such a weak strenght, but combined they work. The idea is not to let the LoRa hijack the model's aesthetics/fine details, but simply add its NSFW action.

Kudos to the Krea 2 creators - they ain't stupid/afraid, which is a rare exception nowadays.

Sampler + schedule:

Euler + sgm_uniform

Euler + simple

Er_sde + sgm_uniform

Er_sde + simple

Max (best) resolution: 1792x3072

-------------------------------------------

Model:
krea2_turbo_bf16.safetensors (26.3 GB)
or
krea2_turbo_fp8_scaled.safetensors (13.1 GB)

or RAW:
krea2_raw_bf16.safetensors (26.3 GB)
or
krea2_raw_fp8_scaled.safetensors (13.1 GB)

8-step LoRA (use it with the RAW model - 8 steps/strengt 1.00 or 12 steps/strength 0.60:
krea2_turbo_lora_rank_64_bf16.safetensors (469 MB)

vae:
https://civarchive.com/models/2729562 (likely the best) (484 MB)
or
qwen_image_vae.safetensors (254 MB)
or
wan_2.1_vae.safetensors (254 MB)

Text encoders:
https://civarchive.com/models/2731465 (4.5 GB)
or
Huihui-Qwen3-VL-4B-Instruct-abliterated-fp8_scaled.safetensors (5.24 GB)

-------------------------------------------
LoRAs (optional):

https://civarchive.com/models/2728234 (Krea2FilterBypass) 160 B
https://civarchive.com/models/2725430 (KNP) 218 MB
https://civarchive.com/models/2688234 (Realism Engine Krea 2) 1.46 GB
https://civarchive.com/models/2728365 (Krea2-realism) 1.46 GB
https://civarchive.com/models/2730896 (Krea 2 Enhancer) 223.89 MB
https://civarchive.com/models/2729908 ([KREA 2] Detail Slider) 10.99 MB
https://civarchive.com/models/2727636 (Realistic Slider - (Krea-2) (Rank1 7mb)) 6.87 MB
https://civarchive.com/models/2732306 (Krea2 AIO NSFW LoRA) 218 MB
https://civarchive.com/models/2728644 ([KREA 2] Mystic XXX) 218 MB
https://civarchive.com/models/2729550 (Clothes On And Off (Krea2)) 218 MB
https://civarchive.com/models/2173844 (Facial Cumshot - Krea 2) 218 MB
https://civarchive.com/models/2733144 (Krea 2 Feet) 436 MB