Selfora v2_1 — realistic amateur checkpoint (Krea 2 Turbo, uncensored)

About this version Selfora v.2.1

Selfora v2.1 Night Fix is up.

Night scenes and low-light scenes now generate correctly. Ask for a room lit

only by a phone torch and that is what you get: the light falls off, the rest

of the room stays black, and nothing gets added that you did not ask for. The

old version would light the whole room anyway and sometimes put a bedside lamp

in it that was never in the prompt. One component in the blend was doing that.

It is gone from this version, and the amateur styling weights moved with it.

int8 is live right now. fp8 is uploading, GGUF comes after.

Same prompt and same seed will look different from v2. That is on purpose.

v2 stays on the page for anyone who built workflows around it.

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

### 🆕 What's new in v2

v2 is rebuilt from scratch on a clean RAW Krea 2 base (not a merge stacked on another merge), with a fresh set of style LoRAs baked in at my own balanced weights. What changed since v1:

* Better realism and sharper skin texture. Pores, fine hairs, real unevenness.

* Age finally obeys the prompt. Ask for 40s or 50s and you get it, not a defaulted 25-year-old. Give it clear age cues for the older end and it holds.

* Faster, because every style LoRA is baked in and there's no per-step LoRA overhead.

Three builds this time: bf16 master, fp8 scaled, and int8 (see Which file* below).

Everything you liked about v1 is still here, just cleaner.

---

### 💬 Questions or setup issues?

Join the [Artfat AI Discord](https://discord.gg/r9tepyNSw). I help with installation, workflows and prompting there, and it's where the community hangs out.

Solo project, so the free stuff stays free ☕ Want new models early plus a Discord role? Patreon's open, totally optional → https://www.patreon.com/c/ArtfatAI

---

Selfora is a realism-first checkpoint on Krea 2 Turbo built for one thing: believable amateur, shot-on-a-phone photos. Real skin with pores and texture, candid lighting, natural bodies. The kind of frame that looks like a selfie off someone's camera roll, not a glossy studio render. It's made for AI-influencer and Instagram-style model content.

If you want that look with zero guesswork, pair it with my companion workflow. The presets are already tuned for this exact amateur-selfie output → Selfism AIO v1.1-https://civarchive.com/models/2844184/selfism-aio-krea-2-v11-or-llm-prompter-or-krea2-edit-or-controlnet-or-facestyle-presets-or-luts-or-colorgrade-or-more-50-promt-presets?modelVersionId=3211045

Every style LoRA is baked straight into the checkpoint. Always on, nothing to stack or juggle. Fully uncensored, and just as strong on clean SFW. And it's Turbo, so you get there in a handful of steps.

Selfora started as a personal project, built for my own use so I could get this exact look without stacking LoRAs every time. It came out well enough that I'm sharing it.

### Look & feel

* Amateur / smartphone photoreal. Real skin, real light, no plastic.

* Candid "phone camera" energy, not editorial studio gloss.

* Made for AI Instagram models and virtual influencers.

* Dialed in out of the box, consistent without touching a single LoRA.

### Bring your own character LoRA

This is what a lot of you actually want it for, so here's the honest rundown. Selfora takes character LoRAs on top just fine: the baked style stays, your character's identity comes through. A few things make it reliable.

* Load your character LoRA into both sampler passes, not just the first. If it's only in the base pass, the refine pass repaints the face and you lose the likeness on close-ups.

* Strength around 0.8 is the sweet spot. Push it too high and hair detail breaks up first (that's the classic over-stack tell). If identity is weak, nudge it up. If hair frays, nudge it down.

* A weak or undertrained LoRA will drift. The baked weights sit a little differently than the clean base your LoRA trained on, so a thin identity signal gets outvoted. A well-trained one punches right through.

* Put the ethnicity and look in the prompt too, don't lean on the LoRA alone. And skip trigger words that are real dictionary words, because Krea 2's text encoder reads language and the prompt can fight your character.

### Which file (bf16 / fp8 / int8)

* bf16 is the master: max quality and max detail. It's big (24 GB+ range) and it's the best pick if your character LoRA needs every bit of identity.

* fp8 scaled gives you that quality at about half the size. The everyday pick.

* int8 is the smallest and the fastest on low-VRAM cards, since it loads lighter and offloads less. Great on 12 GB.

### Recommended settings (Krea 2 Turbo)

The cleanest results come from the Selfism AIO v1.1-https://civarchive.com/models/2844184/selfism-aio-krea-2-v11-or-llm-prompter-or-krea2-edit-or-controlnet-or-facestyle-presets-or-luts-or-colorgrade-or-more-50-promt-presets?modelVersionId=3211045— it's tuned specifically for this model, with everything already dialed in. Already on Selfism v1? Just download v2 and load it.

Building your own graph? Two passes, both CFG 1. Important: set your generation resolution to roughly what you actually want out — don't start too small. The polish pass runs at the same resolution as the base (there's no latent upscale between the two passes), so whatever you generate at is your working resolution. On a tiny base the face and eyes get too few pixels and the detail stays poor — and a final upscale model can't invent detail that was never generated. Portrait 832×1216 or 1024×1024 is a good floor; go higher for more facial detail. This is the RES4LYF /

ClownsharKSampler setup:

* Pass 1, base: res_2s (exponential), scheduler beta, 6 steps, eta 0.45, denoise 1.0, bongmath on

* Pass 2, polish: deis_3m (multistep), scheduler bong_tangent, 2 steps, eta 0.60, denoise 0.20, bongmath on

* Resolution (optional): add it with an upscale model, not another full sampler pass. A few that pair well:

* 4xNomosWebPhoto_RealPLKSR — amateur/web-photo look, keeps grain and pore texture

* 4x-ClearRealityV1 — clean realism, balanced

* 4xPurePhoto-RealPLKSR — very clean, smoother skin (use if you want polished)

* 1xSkinContrast-High-SuperUltraCompact — 1x, layer on top to push skin texture back in

How to think about the two passes: the first pass is a rough draft that locks the overall shape, pose and composition. Glance at it, and if you don't like the framing or something mutated, skip it right there before spending time on the finish. The second pass builds the final on top of it, adding the detail and cleaning it up. So pass 1 is your preview, pass 2 is the keeper.

That's 8 steps total, and the second pass is a light finish, not a heavy hires. A heavy high-denoise second pass repaints the frame, which is what causes skin droplets on close-ups or the face shifting between passes. Keeping it light and letting the upscale model add resolution avoids that.

At CFG 1 the negative prompt does nothing, so say what you do want, positively.

### Prompting

No trigger words for the style, it lives in the weights. Just describe the scene. Krea 2's text encoder is language-based, so it reads best with plain natural sentences instead of comma-piles of tags. Write it the way you'd describe a photo to a friend.

None of these are required, since the amateur look is already baked in, but they help you dial a specific vibe. Weave a couple into a sentence rather than dumping the whole list.

* Amateur / candid: amateur iphone selfie, candid expression, casual everyday, unposed, story-upload quality

* iPhone / camera look: shot on iPhone, iPhone 15 Pro photo, warm iphone color, soft HDR, slight sensor grain, front-camera selfie

* Real skin: natural unretouched skin texture, visible pores and freckles, minor blemishes, bare face, realistic catchlights

* Grittier / night: low light phone photo, sensor noise, slight motion blur, handheld, warm tungsten light

* Candid energy: mid-laugh, genuine unposed expression, slightly tired eyes

Avoid these, since they push it toward studio and AI-gloss: professional photography, studio lighting, DSLR, cinematic, retouched, 8k. At CFG 1 the negative is ignored, so just keep them out of the positive.

💡 Don't want to type any of this? The Selfism_Krea2_optimal_v1 workflow has all of it dialed in as one-click presets, tuned specifically for Selfora → Selfism AIO v1.1-https://civarchive.com/models/2844184/selfism-aio-krea-2-v11-or-llm-prompter-or-krea2-edit-or-controlnet-or-facestyle-presets-or-luts-or-colorgrade-or-more-50-promt-presets?modelVersionId=3211045

### Content

Uncensored, NSFW and SFW both. It leans toward realistic amateur NSFW, but clean portraits and everyday shots come out just as well.

### Best with my Selfism_Krea2_optimal_v1 workflow

Selfora runs best in my Selfism_Krea2_optimal_v1 workflow. It ships with ready-made presets for the selfie / amateur look plus a full set of face-detailer presets, and the two work together: the scene preset sets the shot, the face preset finishes the skin. It's powered by two custom nodes, Artfat LLM Prompter (image or text to prompt, on-device) and Artfat Face Preset (one-click phone-camera face presets). Load Selfora, pick your presets, shoot.

### License & usage

Personal, non-commercial use only. Don't sell the model or anything made with it, and don't use it commercially. Don't re-upload or pass this checkpoint off as your own.

### Credits

Full credit to the creators of the LoRAs merged into this checkpoint (merged at my own balanced weights):

* [PornMaster Krea2 (Uncensored)](https://civarchive.com/models/2768638/pornmaster-krea2-uncensored)

Selfora v2 — realistic amateur checkpoint (Krea 2 Turbo, uncensored)

---

### 🆕 What's new in v2

v2 is rebuilt from scratch on a clean RAW Krea 2 base (not a merge stacked on another merge), with a fresh set of style LoRAs baked in at my own balanced weights. What changed since v1:

* Better realism and sharper skin texture. Pores, fine hairs, real unevenness.

* Age finally obeys the prompt. Ask for 40s or 50s and you get it, not a defaulted 25-year-old. Give it clear age cues for the older end and it holds.

* Faster, because every style LoRA is baked in and there's no per-step LoRA overhead.

Three builds this time: bf16 master, fp8 scaled, and int8 (see Which file* below).

Everything you liked about v1 is still here, just cleaner.

---

### 💬 Questions or setup issues?

Join the [Artfat AI Discord](https://discord.gg/r9tepyNSw). I help with installation, workflows and prompting there, and it's where the community hangs out.

Solo project, so the free stuff stays free ☕ Want new models early plus a Discord role? Patreon's open, totally optional → https://www.patreon.com/c/ArtfatAI

---

Selfora is a realism-first checkpoint on Krea 2 Turbo built for one thing: believable amateur, shot-on-a-phone photos. Real skin with pores and texture, candid lighting, natural bodies. The kind of frame that looks like a selfie off someone's camera roll, not a glossy studio render. It's made for AI-influencer and Instagram-style model content.

If you want that look with zero guesswork, pair it with my companion workflow. The presets are already tuned for this exact amateur-selfie output → Selfism AIO Krea 2 v.1.1 (https://civarchive.com/models/2844184/selfism-aio-krea-2-v11-or-llm-prompter-or-krea2-edit-or-controlnet-or-facestyle-presets-or-luts-or-colorgrade-or-more-50-promt-presets?modelVersionId=3211045)

Every style LoRA is baked straight into the checkpoint. Always on, nothing to stack or juggle. Fully uncensored, and just as strong on clean SFW. And it's Turbo, so you get there in a handful of steps.

Selfora started as a personal project, built for my own use so I could get this exact look without stacking LoRAs every time. It came out well enough that I'm sharing it.

### Look & feel

* Amateur / smartphone photoreal. Real skin, real light, no plastic.

* Candid "phone camera" energy, not editorial studio gloss.

* Made for AI Instagram models and virtual influencers.

* Dialed in out of the box, consistent without touching a single LoRA.

### Bring your own character LoRA

This is what a lot of you actually want it for, so here's the honest rundown. Selfora takes character LoRAs on top just fine: the baked style stays, your character's identity comes through. A few things make it reliable.

* Load your character LoRA into both sampler passes, not just the first. If it's only in the base pass, the refine pass repaints the face and you lose the likeness on close-ups.

* Strength around 0.8 is the sweet spot. Push it too high and hair detail breaks up first (that's the classic over-stack tell). If identity is weak, nudge it up. If hair frays, nudge it down.

* A weak or undertrained LoRA will drift. The baked weights sit a little differently than the clean base your LoRA trained on, so a thin identity signal gets outvoted. A well-trained one punches right through.

* Put the ethnicity and look in the prompt too, don't lean on the LoRA alone. And skip trigger words that are real dictionary words, because Krea 2's text encoder reads language and the prompt can fight your character.

### Which file (bf16 / fp8 / int8)

* bf16 is the master: max quality and max detail. It's big (24 GB+ range) and it's the best pick if your character LoRA needs every bit of identity.

* fp8 scaled gives you that quality at about half the size. The everyday pick.

* int8 is the smallest and the fastest on low-VRAM cards, since it loads lighter and offloads less. Great on 12 GB.

### Recommended settings (Krea 2 Turbo)

The cleanest results come from the Selfism v2 workflow — it's tuned specifically for this model, with everything already dialed in. Already on Selfism v1? Just download v2 and load it.

Building your own graph? Two passes, both CFG 1. Important: set your generation resolution to roughly what you actually want out — don't start too small. The polish pass runs at the same resolution as the base (there's no latent upscale between the two passes), so whatever you generate at is your working resolution. On a tiny base the face and eyes get too few pixels and the detail stays poor — and a final upscale model can't invent detail that was never generated. Portrait 832×1216 or 1024×1024 is a good floor; go higher for more facial detail. This is the RES4LYF / ClownsharKSampler setup:

* Pass 1, base: res_2s (exponential), scheduler beta, 6 steps, eta 0.45, denoise 1.0, bongmath on

* Pass 2, polish: deis_3m (multistep), scheduler bong_tangent, 2 steps, eta 0.60, denoise 0.20, bongmath on

* Resolution (optional): add it with an upscale model, not another full sampler pass. A few that pair well:

* 4xNomosWebPhoto_RealPLKSR — amateur/web-photo look, keeps grain and pore texture

* 4x-ClearRealityV1 — clean realism, balanced

* 4xPurePhoto-RealPLKSR — very clean, smoother skin (use if you want polished)

* 1xSkinContrast-High-SuperUltraCompact — 1x, layer on top to push skin texture back in

How to think about the two passes: the first pass is a rough draft that locks the overall shape, pose and composition. Glance at it, and if you don't like the framing or something mutated, skip it right there before spending time on the finish. The second pass builds the final on top of it, adding the detail and cleaning it up. So pass 1 is your preview, pass 2 is the keeper.

That's 8 steps total, and the second pass is a light finish, not a heavy hires. A heavy high-denoise second pass repaints the frame, which is what causes skin droplets on close-ups or the face shifting between passes. Keeping it light and letting the upscale model add resolution avoids that.

At CFG 1 the negative prompt does nothing, so say what you do want, positively.

### Prompting

No trigger words for the style, it lives in the weights. Just describe the scene. Krea 2's text encoder is language-based, so it reads best with plain natural sentences instead of comma-piles of tags. Write it the way you'd describe a photo to a friend.

None of these are required, since the amateur look is already baked in, but they help you dial a specific vibe. Weave a couple into a sentence rather than dumping the whole list.

* Amateur / candid: amateur iphone selfie, candid expression, casual everyday, unposed, story-upload quality

* iPhone / camera look: shot on iPhone, iPhone 15 Pro photo, warm iphone color, soft HDR, slight sensor grain, front-camera selfie

* Real skin: natural unretouched skin texture, visible pores and freckles, minor blemishes, bare face, realistic catchlights

* Grittier / night: low light phone photo, sensor noise, slight motion blur, handheld, warm tungsten light

* Candid energy: mid-laugh, genuine unposed expression, slightly tired eyes

Avoid these, since they push it toward studio and AI-gloss: professional photography, studio lighting, DSLR, cinematic, retouched, 8k. At CFG 1 the negative is ignored, so just keep them out of the positive.

💡 Don't want to type any of this? The Selfism_Krea2_optimal_v1 workflow has all of it dialed in as one-click presets, tuned specifically for Selfora → Selfism AIO v1.1-https://civarchive.com/models/2844184/selfism-aio-krea-2-v11-or-llm-prompter-or-krea2-edit-or-controlnet-or-facestyle-presets-or-luts-or-colorgrade-or-more-50-promt-presets?modelVersionId=3211045

### Content

Uncensored, NSFW and SFW both. It leans toward realistic amateur NSFW, but clean portraits and everyday shots come out just as well.

### Best with my Selfism_Krea2_optimal_v1 workflow

Selfora runs best in my Selfism_Krea2_optimal_v1 workflow. It ships with ready-made presets for the selfie / amateur look plus a full set of face-detailer presets, and the two work together: the scene preset sets the shot, the face preset finishes the skin. It's powered by two custom nodes, Artfat LLM Prompter (image or text to prompt, on-device) and Artfat Face Preset (one-click phone-camera face presets). Load Selfora, pick your presets, shoot.

### License & usage

Personal, non-commercial use only. Don't sell the model or anything made with it, and don't use it commercially. Don't re-upload or pass this checkpoint off as your own.

### Credits

Full credit to the creators of the LoRAs merged into this checkpoint (merged at my own balanced weights):

* [PornMaster Krea2 (Uncensored)](https://civarchive.com/models/2768638/pornmaster-krea2-uncensored)

### VAE

Several VAEs work fine with this checkpoint, use whichever you have:

* the standard Qwen Image VAE

* Wan 2.1 VAE (both the fp32 version, Wan2_1_VAE_fp32, and the regular one)

* krea2RealVae (v10)

### Notes

Self-contained one-file checkpoints, just load and go. Three builds: bf16 master, fp8 scaled, and int8.

### Notes

Self-contained one-file checkpoints, just load and go. Three builds: bf16 master, fp8 scaled, and int8.

Files with SHA256: 914d02abed68fc9d7542dd8ee7554112845c791e7eb06f0554432f7d4921e65c