CivArchive
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined
    Preview undefined

    Shameless plug for my deviant art page https://www.deviantart.com/le-razzz
    Follow me there for great Ai video content.



    Realism Engine Krea 2 v3


    This version is bring more concepts, better realism and a new workflow using ComfyUI-VAE-Utils node and wan 2.1 VAE https://huggingface.co/spacepxl/Wan2.1-VAE-upscale2x/blob/main/Wan2.1_VAE_upscale2x_imageonly_real_v1.safetensors

    Realism Engine Krea 2 v2


    This version is pretty in line with Ideogram 4 v5 knowledge.


    Realism Engine Krea 2 v1



    Due to high demand I decided to do a Krea 2 training. There's probably improvements to be made as I learn the quirks of the model.

    The gallery represent pretty much what the V1 can do.


    V5 it out! It fix a lot of the issues V4 had.


    Not many concept added but the previous ones are reinforced.

    Better styles, more realism even with euler which make everything faster.

    The default settings for the Lora's are 0.8 and 0.3-0.5 for the unconditional.

    If you like a cleaner image you can increase the unconditional value but it might start to look more plastic.

    Added concepts:

    • sucking breasts

    • deepthroat

    • soaking wet pussy


    V4. New concepts

    • no bra pokies

    • anal fisting

    • vaginal fisting

    • big hanging boobs

    • bondage

    • spread pussy

    • futanari

    • anal prolapse

      I recommend .7 for the Lora and .3 for unconditional. Euler look more realistic now so you can get faster realistic generation.

      V3 add a bunch of sexual position concepts

      Here's the full list of all the concepts so far:

    • nude woman

    • nude man

    • pov vaginal sex

    • pov doggy style vaginal sex

    • reverse cowgirl vaginal sex

    • sideview doggy style vaginal sex

    • doggy style anal sex

    • pov anal sex

    • reverse cowgirl anal sex

    • pov blowjob

    • blowjob

    • sloppy blowjob

    • handjob

    • titjob

    • vaginal dildo

    • lesbian pussy licking

    • huge natural breasts

    • face cum

    • hairy pussy

    • pussy insertion (name your object)

    • puffy nipples

    • big fake boobs

    • upskirt

    • ass spread

    • creampie

    • gaping pussy

    • gaping anus

    • self fisting

    Note: There might be a bit of pussy and anus confusion sometime. In that case put a bounding box where the other hole should be.

    V2 add men and a bunch of new concepts.

    Don't go higher than .90 or it will start creating bad quality image.
    Keep the steps at 18 instead of 12 as you may get generations error from time to time.

    Sweet spot is between 0.50-0.90.

    res_2s = best definition

    res_2m = most realistic and faster than 2s

    euler = fastest but less realistic

    https://huggingface.co/RazzzHF/Realism_Engine_Ideogram_4/resolve/main/rei4_v2_workflow.json

    V1 - First RE Ideogram 4 release. Trained on female DS only so far. Male is coming in the next versions.

    • Sweet spot is between 0.50-0.60. Higher will degrade the quality.

    • MAKE SURE YOU USE JSON PROMPTS.

    • res_2s res_2m is recommended.

    Description

    This version bring the knowledge to the level of RE Ideogram v5

    FAQ

    Comments (65)

    DonutsDeliveryJun 26, 2026
    CivitAI

    krea2 best shit ever

    GPUPoorChadJun 26, 2026
    CivitAI

    Didn't have much luck with Ideogram and it was super slow, perhaps a skill issue here, but this feels like Ideogram but consistently good and fast this btfos SDXL slop finally I think if you are smart with other LoRAs and in-painting.

    If I had more VRAM, I had interesting idea a while ago to code up OpenCode/Codex but for images, just prompting until VLLM says it has image that adheres to prompt, and maybe use impainting or LoRAs or even other models, etc. Just until it gets the desired result. This model combined with the insane amount of SDXL fine-tunes and LoRAs seems perfect for that idea, maybe someone richer than me can make it all work, a uncensored Gemma model would be a good VLLM for this

    makiaeveliJun 26, 2026

    if you can sample one at a time you can sample both. There are several iterations of this, some were mentioning running a Hermes Agent instance and communicating with it however you wanted. It'd run the LLM, generate a comfy workflow, and run it, providing you an output.

    GPUPoorChadJun 26, 2026· 2 reactions
    CivitAI

    For some reason Kirea 2 version really struggles with pubic hair, and it takes wet pussy a bit too literally everything is wet, has bias for clothing on guys, doesn't do cowgirl well. Otherwise however, insanely impressed, it's crazy this quality of image generator would have blown people away if some major lab released it proprietary not too long ago, but I have it open source* on my dinky little RX 6800 somewhat uncensored

    Agent_SmthJun 26, 2026· 3 reactions

    weird, i often got the opposite, only pubic hair that i can't remove.

    firemanbrakeneckJun 26, 2026

    That's kind of how it's been for a while. Open weights are about / at most a year behind closed api. Commercial models are only in business because most people are too lazy / technically challenged to run open models, and the obscene prices for good pc parts thanks to our ferengi business culture. And probably moreso hype generating funding where demand is nonexistent.

    At least with llms, the api has a somewhat reasonable free tier, and competitive open models really are server class. Not so with imagen.

    fronyaxJun 26, 2026· 6 reactions
    CivitAI

    KREA 2's Goon capability absolutely blows ZIT/ZIB out of the water 😂😂

    GPUPoorChadJun 26, 2026· 3 reactions

    SDXL is finally completely cooked for the most part, could still be better, but this is best I've seen personally

    fox23vang226Jun 26, 2026· 2 reactions

    Yup ZIT/ZIB has been dead to me for months now, its a very dumb model and learns details very poorly in training. Ideogram and Krea2 is now king.

    fronyaxJun 26, 2026· 3 reactions

    @GPUPoorChad Yeah hate to say this, SDXL era is mostly over 😭 RIP.

    Anima, KREA2, Klein 9B (image edit), IDE4 takes the crown. Can't wait for KREA2 image edit.

    bigASP 3.0 is still in progress tho. Continuing SDXL spicy model (bigLove,Lustify) legacy.
    https://huggingface.co/fancyfeast/bigasp-3

    GPUPoorChadJun 28, 2026

    @fronyax Love to say*

    VariousDog9932615Jun 26, 2026· 5 reactions
    CivitAI

    Is Krea2 the real model that finally make SDXL fully obsolete? so far i'm impressed by coherence and overall prompt understanding, also not many critical anatomical flaws to notice in first few tries... image 'pixel sharpness' though is still a touch under the best sdxl models if I want to 'bitch' about something though. And your lora seems capable or roughly anything that you throw at it... very well done, can't wait for a v3 or v4 if you feel like pushing it even to greater quality.

    New_NameJun 26, 2026
    CivitAI

    Recommended strength for this LORA?

    I can literally see zero difference turning this on and off. I have tried strength from 0.4 - 1.5.

    razzz
    Author
    Jun 26, 2026· 1 reaction

    usually 0.9 - 1.0 unless it's very suborn with what you want to achieve then you can go over 1.0

    sumsen22Jun 26, 2026
    CivitAI

    Your V5 version for IdeogramV4 is already great. Interestingly, it works the better the higher the resolution is. Got very good results in 6 MP without the hazzle of upscaling and tiled refining. (Disclaimer: Did only test a subset of the promised concepts so far, and sometimes some limitations show up. The rendering at smaller resolution is a little bit too "harsh" for my taste, but anyhow the overall package is already great.)

    SD_STORMJun 26, 2026
    CivitAI

    Problems with krea2 lora v2 and v1:

    and other blocks too:

    [WARNING] lora key not loaded: diffusion_model.single_blocks.9.linear1.lokr_w2

    [WARNING] lora key not loaded: diffusion_model.single_blocks.9.linear2.alpha

    [WARNING] lora key not loaded: diffusion_model.single_blocks.9.linear2.lokr_w1

    [WARNING] lora key not loaded: diffusion_model.single_blocks.9.linear2.lokr_w2

    makiaeveliJun 26, 2026

    This has more to do with your lora loader. Generally you need to update your inference application.

    SD_STORMJun 26, 2026

    @makiaeveli comfyui is uptodate.

    Vyxen808Jun 26, 2026
    CivitAI

    (Ideogram 4) Has anyone successfully mixed this LoRA with a character LoRA without affecting the character's likeness?

    ainewb14Jun 26, 2026· 2 reactions

    Yes, apply your character lora to base strength 1 and unconditional strength 0.3

    onionhulu781Jun 26, 2026· 5 reactions
    CivitAI

    After some testing I prefer V1 over V2 for Krea 2. V2 improved in some ways, but totally killed the variety/creativity I was getting with different seeds on V1. V2 has more of that classic AI same face problem

    fronyaxJun 26, 2026· 3 reactions

    I thought it's just me, can confirm the seed variance is downgraded vs V1.

    V2's seeds variety reminds me of ZImage Turbo which is very bad.

    aising23Jun 26, 2026

    Using keywords like "random" in prompts, adjusting noise etc. adds another level to variety.

    marth4200Jun 27, 2026

    same

    arsibaltJun 29, 2026

    IMO nowhere near as severe as ZIT, but I agree there is some effect. Worth it for the expanded knowledge, especially since there are straightforward ways to play with variance via conditioning and so on

    secretofmanaJul 4, 2026

    yeah i made some try today and was generating a girl with a prompt using some aging word like "mature woman" and i had the same woman generated before without the mature tag but older with same face while being in random seed mode. I didnt try v1, i may try v1 to see if i get face variation thought i wonder if it's not an issue of the base model itself. And since i was a fan of chroma model by Lodestone which was a model for me that offered one of best face variation because maybe of wise training data strategy. I have hard times to suffer the lack of variety in generation and seek randomness a lot in body and face features. Lot of models are like this lacking diversity but maybe they were trained like that as a base and if you want to add variety you have to make lora or finetuned. So maybe we need that.

    secretofmanaJul 5, 2026

    ok so after many tests, i have way more face variation with v1.

    zdenekh91278Jun 26, 2026
    CivitAI

    Good one but changing face from another LORAs

    ss9999Jun 26, 2026
    CivitAI

    Wish list: Analingus, testicles size, pubic hair/shadow control, anal fingering, height/size difference, male gyno (nipple size control)

    doorajar24Jun 27, 2026
    CivitAI

    great work with krea2! only issue: v2 is lacking face expressions, it makes it neutral. v1 doesn't override it so hard.

    doorajar24Jun 27, 2026

    I was wrong, v2 is much better with emotions. For some reason, v2 + unlocking weights and loras gives worse emotions, but pure krea2 + realism engine do everything great.

    potatometer350Jun 27, 2026· 5 reactions
    CivitAI

    This keeps generating people with really long legs. Like Belters from The Expanse. I do like the lora, though.

    Belters? lol

    DocueiJun 29, 2026

    You forgot to do a long legs Pomni reference.

    -4589 Karma points for you.

    potatometer350Jun 30, 2026

    This weird phenomenon stopped after restarting comfyui, it seems.

    15976095556410Jun 27, 2026· 2 reactions
    CivitAI

    The krea2 version will affect the face

    marth4200Jun 27, 2026

    yeah, it changes the face a lot whereas v1 of the Krea2 didn't affect it much.

    UdimuzJun 27, 2026· 6 reactions
    CivitAI

    This Lora makes people look much older

    potatometer350Jun 30, 2026

    When I first started using Krea2 I had that problem sometimes. Try different combinations of samplers and schedulers, and don't use those rebalancing nodes.

    DonutsDeliveryJun 27, 2026· 1 reaction
    CivitAI

    A/B shows V1 with better realism

    StoneDawgJun 27, 2026· 1 reaction
    CivitAI

    I for the life of me cannot get no bra pokies to work correctly. it always has the nipples coming through the fabric. any help on how to properly prompt for it??

    BadensteinJun 29, 2026

    You could try "erect nipples". That works most of the time if the model is not too NSFW trained.

    tuguphyJun 29, 2026· 10 reactions
    CivitAI

    Very, very great. I can't live without the Lora you posted now. It's amazing. I'm looking forward to your next version and thank you for your contribution to the community :D.

    ArthurJackBroJun 29, 2026· 2 reactions
    CivitAI

    We need fingering brotha!!! Great work nonetheless!!!

    EuphoricusJun 29, 2026· 4 reactions
    CivitAI

    I tried two poses and had difficulties with both.

    First was masturbation using dildo, while she is holding and moving the dildo. It never came out right.

    Second was cowgirl position from mans point-of-view. It was cowgirl position, but not from PoV.

    Also, any mention of "wetness" results in streams of liquid flowing down the woman. Not just wet pussy or sheen of sweat, whole waterfall.

    tuguphyJun 30, 2026· 3 reactions

    argee with the wetness.

    bradcausa9919Jun 30, 2026

    concur on the abundant fluid flow.

    potatometer350Jul 2, 2026

    I am getting the liquid with or without the lora enabled, and I have never mentioned wetness or any variation of that word in a prompt.

    potatometer350Jun 30, 2026
    CivitAI

    Using Krea2 I have difficulty generating images with missionary sex. Missionary sex is difficult with other models and loras, too. At least the way I envision it. But with this lora, it occasionally nail it perfectly. Most of the time, though, it makes images with the man fisting the woman. Like 9/10 tries it's fisting, then all of the sudden it follows the prompt.

    Without the lora it won't do missionary, but it never does fisting.

    potatometer350Jun 30, 2026

    It won't let me edit. I'm using the V2 version for Krea2. Krea2convrot int8mixed. No other loras or rebalancing nodes. 1MP at 3:2. Lora looks great but has this hiccup. If I make an extremely detailed prompt to force missionary, I often get really weird results for a bunch of images and then, suddenly, it generates what I want.

    uncaringbear449Jun 30, 2026· 5 reactions
    CivitAI

    It's takes a ton of trial and error to prompt what you want consistently, but the results can be quite good. Appreciate all the effort in developing this lora!

    hellyeah354Jul 1, 2026· 1 reaction
    CivitAI

    Hey, looks like great work. I noticed you've listed the Ideogram ones under with a base model of "Other" I think Ideogram 4 is available as an option now and might help people find it.

    LP_AIJul 2, 2026· 1 reaction
    CivitAI

    This lora is great, please keep updating it with newer concepts! I am using the Krea version, it works quite well but occasionally there will be concepts it doesn't understand. Keep up the good work, this lora saved the model for me!

    MeemotJul 4, 2026
    CivitAI

    Assjob / buttjob would be a great concept to add

    BzzzDarklordJul 4, 2026· 7 reactions
    CivitAI

    just recently began to discover for myself wonders of KREA 2 model on multiple levels. thank you for your contribution, very promising LoRA to my modest opinion, GG's

    ShadowCellJul 5, 2026· 19 reactions
    CivitAI

    So far the best NSFW LoRA I've found for Krea 2 (of all the ones I've tried so far, only one other was what I'd really consider to be in an acceptable state tbh). Most of them ruin the image, anatomy, prompt adherence, facial likeness, fine realistic detail of Krea2 (quickly turning it into janky SDXL looking trash), etc.

    This banger LoRA is leagues above the majority attempting the same thing. Bravo <3

    ShadowCellJul 7, 2026

    PS - You should probably put "NSFW" in your LoRA name btw. It's what I (& I'm sure a lot of other people) use in the search box when they're looking. You deserve way higher ranking on the search page for Krea 2.

    Understandable if you were trying to be more low key, & subtle with it though lol.

    outagesgritty0t668Jul 7, 2026· 7 reactions
    CivitAI

    This Lora works really well for Krea2 Turbo but I’ve noticed it tends to add liquid in random places unprompted. I end up having to use the Raw model with negative prompt to get around it. Just wondering if anyone else has experienced it. Otherwise, it’s the only Lora that nails down the positions.

    IriduJul 7, 2026· 1 reaction

    You might want to use a wet slider lora if you haven’t already.

    redcreamsodaJul 7, 2026

    Yeah; it gives me the random liquid issue too. Other than that this model's awesome, but hopefully there's a V3 that fixes that.

    outagesgritty0t668Jul 7, 2026· 1 reaction

    @redcreamsoda The wet slider Lora with negative value fixed it for me.

    fronyaxJul 8, 2026

    Try negpip node, it enables negative prompt on turbo,

    prompt (liquid:-1) on positive prompt field.

    https://github.com/blue-pen5805/ComfyUI-krea2-negpip

    KalemFakesJul 8, 2026

    Same dude

    dyviumld881Jul 9, 2026

    @outagesgritty0t668 where can i find this wet lora please