Krea 2 Masterclass Workflows Updated: Uncensored Encoder, Sharper Skin, Better Prompt Adherence (SFW v3.1 + NSFW v2) 🔓 (Patreon)

If you've been generating with the AI Influencer Masterclass image workflows, this update is worth grabbing. Your prompts land more accurately, skin looks more detailed, and you get more realistic variety from seed to seed. The SFW workflows move from v3 to v3.1, and the NSFW workflows move from STACKED to v2.

What You Need to Download

Huihui-Qwen3-VL-4B-Instruct-abliterated-fp8_scaled.safetensors (all workflows, replaces qwen3vl_4b_fp8_scaled in the CLIP Loader)

skindetails_krea2_loraholic.safetensors (all workflows)

fedor_bypass.safetensors (SFW only, NSFW already used it)

If fedor_bypass or skindetails show up as missing in the SFW workflows, reselect them in the Power Lora Loader.

Changes in Every Workflow

New uncensored text encoder. The abliterated version has its built-in censorship stripped out, so your prompt reaches the model as written. That means better prompt adherence, and NSFW prompts that don't get watered down.

Krea2T Enhancer removed. It's outdated and no longer works properly. One less custom node to install.

No Face Detailer. The Face Detailer pass doesn't function properly with Krea 2, so it's out.

Eye Detailer denoise lowered to 0.08. The old strength was too strong and did the same damage as the Face Detailer: freckles and too much invented random detail. At 0.08 it sharpens the eyes and leaves the face alone.

AuraFlow shift raised from 3 to 6.

Smart Seed Variance set to 📸 Realistic. Variation between seeds now leans toward photorealism.

SFW text2img: v3 → v3.1

👉 Get the workflow here

Removed

Face Detailer and its face_yolov8m detector

Krea2T Enhancer

canon_krea2 LoRA

Changed

Text encoder → Huihui abliterated

Steps: 8 → 12

Resolution: 768 x 1344 → 720 x 1280

AuraFlow shift: 3 → 6

Eye Detailer denoise: 0.10 → 0.08

bloomgirls realism LoRA: 0.4 → 0.3

Smart Seed Variance direction: None → 📸 Realistic

Film Grain bypassed by default (0.07 if you turn it back on)

Added

skindetails LoRA at 0.7

fedor_bypass LoRA at 1.0

SFW img2img: v3 → v3.1

👉 Get the workflow here

Removed

Face Detailer and its face_yolov8m detector

Krea2T Enhancer

canon_krea2 LoRA

Changed

Text encoder → Huihui abliterated

Sampler / scheduler: er_sde / simple → dpmpp_2m / sgm_uniform

Denoise: 0.40 → 0.37

Reference resize (longer edge): 1344 → 1280

AuraFlow shift: 3 → 6

Eye Detailer denoise: 0.10 → 0.08

bloomgirls realism LoRA: 0.4 → 0.3

Smart Seed Variance direction: None → 📸 Realistic

Film Grain bypassed by default (0.07 if you turn it back on)

Positive prompt now ends with a realism line for a candid, iPhone-style look

Added

skindetails LoRA at 0.7

fedor_bypass LoRA at 1.0

NSFW text2img: STACKED → v2

👉 Get the workflow here

Removed

Krea2T Enhancer Advanced

Empty negative prompt box

Example character LoRA (add your own as a new slot)

Changed

Text encoder → Huihui abliterated

Steps: 10 → 12

AuraFlow shift: 3 → 6

Eye Detailer denoise: 0.15 → 0.08

Famegrid Natural: 0.8 → 0.4

Krea2_NSFW_V4: 0.5 → 0.7

blowjob_krea2_v1: 0.8 → 0.6

Save folder: T2I → NSFW/T2I

Added

Smart Seed Variance (Balanced, 📸 Realistic) for more variety between seeds

ConditioningZeroOut as the negative, the cleaner way to run a blank negative at CFG 1

skindetails LoRA at 0.55

NSFW img2img: STACKED → v2

👉 Get the workflow here

Removed

Second CLIP Loader (one abliterated encoder now handles both generating and captioning)

Regex Replace node (the caption now goes straight to Preview)

Example character LoRA (add your own as a new slot)

Node groups

Changed

Text encoder → Huihui abliterated, so captions describe your reference fully instead of refusing

Captioning instructions rewritten:

Clothed or SFW images get a neutral description, nothing sexual invented

Nude images with no act get described as they are

Explicit scenes get described in full

The man is described first, your character's permanent features stay out so your LoRA controls her look

No hedging phrases like "it appears," and watermarks or on-screen text are ignored

CR Prompt List adds a realism line to the end of every caption

Steps: 10 → 12

AuraFlow shift: 3 → 6

Eye Detailer denoise: 0.15 → 0.08

LoRA strengths match NSFW text2img (Famegrid 0.4, NSFW_V4 0.7, blowjob 0.6)

Save folder: ComfyUI → nsfw/images/img2img

Added

Smart Seed Variance (Balanced, 📸 Realistic)

skindetails LoRA at 0.55

Direct Links

SFW Text 2 Image v3.1

SFW Image 2 Image v3.1

NSFW Text 2 Image v2

NSFW Image 2 Image v2

Update your files, grab the new encoder, and drop your results in the Discord. 👇