Big Love Klein3 is also offered for $10 on Tensor Art. You can get Big Love Eryn1, Ultra5, Photo6, Zuna1 or Zia1 for $5.50 and Gwen1 or Hyper1 for $10 on Fanvue. Click 'Follow', buy a post, check your Fanvue DMs! A subscription includes the latest release and new ones within 30 days. Any problem? Contact me via chat.
To create up to 85 free images per day in highest quality you can use Big Love here. For discussing photorealistic image generation join me on Discord.
Big Love Klein is a fine-tune of Flux 2 klein. It can do txt2img and also edit images with prompts. It only needs 4 steps, so is quite fast. Check the "About this version" box on the right for info on the individual variants. Some are up to 3x faster.
It is recommended to use it with ComfyUI. Put the downloaded model into the diffusion models sub folder of ComfyUI. You also need qwen_3_8b_fp8mixed.safetensors in the text_encoders sub folder (Qwen3-8B-Q5_K_M.gguf wth 8 GB VRAM) and flux2-vae.safetensors in the vae sub folder.
Settings are 4 steps (2 steps for upscaling), cfg1, Euler sampler, Beta scheduler, 832x1216, 1024x1536, 1280x1920 or 1536x2240 pixels
Big Love Gwen is a finetune of Qwen Edit 2511 and can do NSFW image editing. It produces more realistic and sharp images with better skin texture & genitals than other Qwen checkpoints/loras. Txt2img is less reliable, but generates nice results too.
Big Love Gwen1 allows commercial usage without additional fees (if no auto-generation involved) unlike Big Love Klein. With its better anatomy it is easier to get good editing results. As fast at 6 steps as Klein at 4 steps. Check the "About this version" box on the right for info on the individual variants.
It is recommended to use ComfyUI. The Gwen1 workflows offer a skin enhance feature that is not available in Forge Neo. Move the Gwen1 model file into the diffusion_model sub folder of the models folder of ComfyUI, qwen_2.5_vl_7b_fp8_scaled.safetensors into the text_encoder sub folder and qwen_image_vae.safetensors into the vae sub folder.
For training a lora read here for Gwen1.
Settings are 6 steps (3 steps for upscaling), cfg1, Euler sampler, Beta scheduler, 832x1216 or 1024x1536 pixels.
Big Love Eryn is a fine-tune of Ernie-Image(-Turbo). You can run it locally with ComfyUI and Forge Neo. The fp8 version works with 8 & 12 GB VRAM. Use the (pruned) bf16 version if you have 16 GB VRAM or more. The "Full" versions are only meant for training a lora (coming soon!). Put the downloaded model into the diffusion models sub folder of ComfyUI or the StableDiffusion sub folder of Forge Neo. You additionally need to place ministral-3-3b.safetensors in the text_encoder sub folder and flux2-vae.safetensors in the VAE sub folder.
Settings are 4-6 steps (3-4 steps for upscaling), cfg1, Euler sampler, Beta scheduler, 832x1216 or 1024x1536 pixels.
Big Love Hyper supports generating at image sizes of 1664x2232 (4 megapixel) and higher with SDXL. Can be easily upscaled to 9 megapixel unlike the normal low-res images. Hyper currently has not fully stable anatomy, so needs more runs. Some prompts work better than others. It produces highly detailed and life-like images close to real photos. It has a special quality that cannot be achieved by just upscaling. The Hyper versions of Big Love are pay-only and special license conditions apply (see below).
Settings are DMD2 lora with strength 1.0, LCM Exponential, 8 steps, cfg 1, clip skip 2, 1664x2432 or 1536x2240. Upscale with img2img with DMD2 lora with strength 1.0, same prompt, LCM Exponential, 1.25-1.5x, 4-8 steps, cfg 1, 0.3-0.5 denoise, clip skip 2.
Big Love ZT, Zia or Zuna are fine-tunes of Z-Image(-Turbo) on the Big Love dataset. You can run it locally with ComfyUI and Forge Neo. The fp8 version works faster with 8 GB VRAM but produces more noise so may sometimes look more realistic, but details are not as good. Use the (pruned) bf16 version if you have 12 GB VRAM or more. The "Full" version is only meant for training a lora. Put the downloaded model into the diffusion models sub folder of ComfyUI or the StableDiffusion sub folder of Forge Neo. To make Big Love ZT/Zia/Zuna work you additionally need to place qwen3_4b.safetensors in the text_encoder sub folder and ae.saftendors in the VAE sub folder.
Settings are 8 steps, cfg1, Euler or DPM++ 2s a RF sampler, Normal or Simple or Beta scheduler, 832x1216, 1024x1536 or 1280x1920 pixels. For upscaling 8 steps, cfg1, Euler sampler, Normal or Simple or Beta scheduler, 1.5x upscale.
Big Love Ultra supports generating at image sizes of 1280x1870, 1360x1984 and higher. Can be easily upscaled to 6 megapixel unlike the normal low-res images. More photorealistic and detailed. A special quality that cannot be achieved by just upscaling. Ultra produces more details & sharpness when upscaling and detailing low-res images. The Ultra versions of Big Love are pay-only and special license conditions apply. See below.
Settings are DMD2 lora with strength 1.0, LCM Exponential, 8 steps, cfg 1, clip skip 2, 1280x1864 or 1360x1984. Upscale with img2img with DMD2 lora with strength 1.0, same prompt, LCM Exponential, 1.25-1.5x, 4-8 steps, cfg 1, lora strength 1.0, 0.3-0.5 denoise, clip skip 2.
Big Love Photo and Insta1 are finetuned versions of XL. Lust1 is a finetuned version of Lustify merged with Photo. They were trained on thousands of images and dozens of new concepts. Why they are the most realistic and versatile. For more details and prompting Photo1 read this article.
Settings are DMD2 lora with strength 1.0, LCM Exponential, 8 steps, cfg 1, clip skip 2, 832x1216 or 1024x1496. Upscale with img2img with DMD2 lora, same prompt, LCM Exponential, 1.25-1.5, 4-8 steps, cfg 1, lora strength 1.0, 0.3-0.5 denoise, clip skip 2.
Big Love XL is a combination of 5 different training branches of SDXL: original SDXL, bigASP, NatVis, Anteros and Pony. With a bit of Pony in it, some Pony loras (poses, characters) work too. It outputs a more photorealistic & creative look than the Pony versions and can also do paintings & cartoon.
Settings are DMD2 lora with strength 1.0, LCM Exponential, 8 steps, cfg 1-1.5, clip skip 2, 832x1216 or 1024x1496. Upscale with img2img with DMD2 lora, same prompt, LCM Exponential, 1.5-2x, 8-12 steps (XL1/XL2), 4-8 steps (XL2.5-XL4), cfg 1, lora strength 1.0, 0.3-0.5 denoise, clip skip 2.
Big Love Pony produces different images than both Pony & SDXL-based models. Pony tags are understood, but interpreted a bit differently. Some images lean more in the one or other direction depending on the prompt. It works with Pony loras as well as normal SDXL ones. It can turn anime/cartoon into photorealistic images in img2img. Pony3 includes a bit of Illustrious.
Settings are DMD2 lora with strength 1.0, LCM Exponential, 8 steps, cfg 1-1.5, clip skip 2, 832x1216 or 1024x1496. Upscale with img2img with DMD2 lora, same prompt, LCM Exponential, 1.5-2x, 8-12 steps, cfg 1, lora strength 1.0, 0.3-0.5 denoise, clip skip 2.
ComfyUI Workflow:
Download one of the following images and drag'n'drop it onto Comfy:
Klein, Eryn and Gwen workflows are embedded in the showcase images.
Z-Turbo/Z-Base txt2img & Upscale Workflow (recommended)
SDLX DMD2 txt2img & Upscale Workflow (recommended)
SDXL DMD2 txt2ing & Upscale for Big Love Ultra
SDXL DMD2 img2img Workflow
SDXL Lightning Workflow
Trained Concepts
Newer versions of Big Love include the concepts of older versions, e.g. Photo4 can do everything that Photo1 to Photo3 can do and more. The only exception are the Insta concepts, but they have been partially added to never versions too.
Photo6/Ultra5: ai-style, bodyscape, butt plug, candid amateur, cumshot, illustrious-style, outdoor sex, partner fingering, photoart, innie pussy, ray-style, rebellious, reflection, sensual, sexy ass, sweetheart, wedgie
Photo5/Ultra4/Hyper1: alluring, anal sex, artistic photo, artistic pinup, charming, cum portrait, cute asian, cute pinup, cute portrait, double oral, dreamy-style, face pov, fantasy pinup, fantasy-style, kai-style, lowkey, mj-style, outdoor flashing, photo art, pole dance, public indecency, public nudity, rock climbing, scify-style, sword pose, vaginal sex
Photo4/Photo4.5/Ultra2/Ultra3: 2girls kissing, acrobatic sex, average face, beautiful face, amateur pinup, anal fingering, artistic portrait, cuddling, cum swapping, doggystyle pov, even skin color, fashion photo, female masturbation, female pov, finger licking, food porn, footjob, french kiss, full nelson, funny movie, funny tongue out, insta cute, legs pov, low contrast, mirror portrait, nude pose, oral pov, orgasm face, outdoor nude, penis licking, pinup photo, pro photo, pro portrait, pussy licking, sex pov, sexy portrait, vaginal fingering, waterfall
Photo3/Lust1/Ultra1: 360 degree photo, 3d selfie, 69 position, adorable, alt beauty, anal gape, anal sex, average face, beach life, beautiful, beautiful face, candidness, cum, cute, deepthroat, dildo insertion, double exposure, enthusiastic, extreme pose, faces, fashion photo, female pov, fisting, follow me pov, girl next door, gorgeous, handjob, heroin chic, huge feet, huge hands, huge nipples, light rays, light streaks, lighting, low/mid/high contrast, lowkey/midkey/highkey, low/mid/high saturation, middle finger, motion blur, object insertion, detailed eyes, teeth, pro photo, pussy gape, real life, rooftop, round ass, sad, sexy pose, skin texture, skin tone, spread pussy, squirting, street portrait, testicle licking, testicle sucking, vaginal sex, voluminous hair, water splash
Insta1: amateur photo, fashion photo, insta selfie, pinup photo, pro photo, real life, big ass, bimbo, cute, luxurious, adorable, enthusiastic, average face, beautiful face, detailed eyes, sexy legs.
Photo2: 3d selfie, 69 position, adorable, alt beauty , amateur photo, ball licking, ball sucking, beach life, bubble butt, low/mid/high contrast, dildo insertion, double exposure, enthusiastic, extreme pose, female pov, follow me pov, forced perspective, girl next door, gorgeous, handjob, heroin chic, huge feet, huge hands, huge nipples, light rays, light streaks, lighting, middle finger, motion blur, no tan lines, object insertion, pussy gape, rooftop, sad, low/mid/high saturation, sexy pose, squirting, street portrait, water splash.
Photo1: amateur photo, pro photo, lighting, lowkey, midkey, highkey, beautiful, cute, candid, real life, average face, remarkable face, voluminous hair, skin tone, skin texture, 360 degree photo, spread pussy, deepthroat, fisting, anal gape, anal sex, vaginal sex.
Training a (character) lora
My advice is this: Take 10-30 high quality images (at least 50-100 for styles or poses), download OneTrainer, click the SDXL lora preset, choose Big Love as the base checkpoint, add your images as a concept, add tags on the Tools tab, and start training. If you train on Big Love Ultra change the training resolution to 1536 and provide 1248x1832 images.
You will only get the full Big Love quality by training on real photos or extremely photorealistic images, which is usually not the case with a fictional character. It is all about image quality and not choosing boring images. Don't add posing images to a character lora, because Big Love provides the posing later for it. Focus mainly on good portraits with a few upper and full body images. Rather fewer higher quality images than a lot of average images. Read here for traing a lora on Big Love ZT2, ZT3, Zia1, Zuna1 or Eryn1.
Fixing Anatomy
If you want to keep the seed and repair anatomical problems in an image, activate the Extra checkbox in A1111/Forge and set Variation Strength to 0.01 to 0.1. Then generate until anatomy is fine. In Comfy there are sampler nodes that support variation seed (e.g. Inspire Pack), which do the same thing. Also possible to change steps or add commas to the prompt, but I do not recommend it. Extra/Variation Seed is more powerful and convenient as it gives you endless variations with just one click. There is always a variation which looks similar or even better with great anatomy.
You don't need negative prompts for this. They are rather ineffective. You can also change the positive prompt to suppress things. Anatomy problems are also created by the prompt if you prompt contradicting poses that the model cannot combine. Better to fix the prompt first then.
Reducing Saturation, Warm Colors and Sharpness
These SDXL loras reduces the saturation and change the color temperature without reducing image quality. The lora strength defines the effect. Negative values increase saturation or add more warmth. To make an image softer use a lower strength for the DMD2 lora, e.g. 0.95, 0.9, 0.85.
Big Love Hyper/Ultra/Photo Sizes
Big Love Hyper supports these image sizes:
1664x2432 (recommend)
1536x2240
Big Love Ultra supports these image sizes:
1360x1984
1280x1872 (recommend)
Big Love Photo supports these image sizes:
1024x1496 (recommend)
832x1216
You can also use the 16:9 or 2:3 versions of these image sizes as well as landscape orientation, but the above should be more reliable.
The higher the image size, the lower the probability of good results. If you get too many problems with a higher image size, switch to the lower one. Some prompts do higher sizes fine, some do not. Do not use a lower image size than the recommended one, otherwise image quality will be lower. With higher image sizes anatomy problems occur too often.
Remove fill words and unusual tags in the prompt. They can cause a quality reduction with Big Love Ultra/Hyper as they were not trained. Shorter prompts should work better as it is less likely that bad or untrained tags are in it.
SFW Output - No Tits Please
"Is there a trick so that Big Love doesn't always generate raised shirts or visible breasts?" I get this asked so often, while it is so obvious. Big Love knows what men want so it gives it to them. Negatives don't work that well for it. So you got to speak to it in the positive prompt like you were taking to a nun. "Breasts? How dare you speak such an obscene word to me." Don't mention breasts or any of these sexy parts. Instead describe clothes a lot. Don't write down your impure thoughts. God forbid! Yeah, I know, so hard. Describe her body shape, which also implies certain tits underneath the clothes.
XL Prompting Advice
SDXL prompts & prompts from various SDXL checkpoints work with it. Pony prompts too, but they create a more photorealistic look. Natural language prompts tend to create artistic glamour images with less skin detail while normal tags do more photo-style images. Big Love XL gives individual words more attention, so you don't need to rework prompts like Big Love Pony requires sometimes.
XL1/XL2 has a tendancy towards cuteness and bright skin. If you get too much of it, remove words like cute, sweet, pale/fair skin & flash from the prompt. It also creates realistical analog photos with reduced image quality. Sometimes better to remove terms like analog or grain to increase quality. Decrease cfg to get a more natural look.
XL2.5 can sometimes get too sharp. Reduce cfg or use less steps in img2img to make images softer.
Pony Prompting Advice
Pony prompts & prompts from other checkpoints work with it. Score tags and fill words like masterpiece or perfect skin have as good as no effect usually. A negative prompt is often unnecessary unless you want to make something vanish or avoid underage, but it can also change the style of the image.
Pure booru tags and some other samplers may produce Pony style. Photographic terms or a simple prompts ensure a photo look. Sometimes it refuses to produce medium shots, portraits or close-ups unless you weigh the term heavily or remove feet, shoes, high heels, boots etc. from the prompt. wavy hair sometimes does not look so great.
If it refuses to do porn, try simplifying the prompt. If you want to do a portrait, don't describe surroundings too much. If you want to do porn, describe the action mainly and don't focus on the people or surroundings in the prompt. The largest part of the prompt should be about the main subject of the image.
If a prompt does not produce the intended image, you often only need to modify it a bit to make it work perfectly.
Using Character Loras
Big Love Pony supports Pony characters, but they look like real humans. There were some complaints about them not working, but the problem was not the model itself so far. So here are some tips:
* Try Pony as well as SDXL character loras
* Use a lora weight up to 1.5 if it does not reduce quality
* If it is a well known character, add the name in the prompt and also give it a higher weight if necessary.
* Check your negative prompt for problems. Delete it if necessary and rewrite it from scratch.
* Render a lot of images to be able to pick those with the best resemblance.
More SDXL Photorealism:
You can highly improve the images generated with this model by using img2img, upscaling, detailing etc. The SDXL Lightning lora enhances the look and constrast despite only 8 steps. Alternatively, the SDXL DMD2 lora produces a sharp, but more natural look with less contrast. Also check out the Subtle Style loras, which nicely enhance images.
Big Love License
Big Love SDXL versions are licensed under CreativeML Open RAIL++M, the Big Love Z-Image, Ernie & Qwen versions are licensed under Apache 2.0, and the Big Love Klein versions are licensed under the FLUX Non-Commercial License with the following additional terms:
For clarity, the following version tiers are defined:
Restricted versions: Ultra, Hyper, Klein, Zia, Zuna, Eryn, Gwen
Open versions: Pony, XL, Photo, Insta, Lust, ZT
1. You may use Big Love without crediting the creator.
2. You may sell the images that it generates.
3. You may only run the official Big Love versions of the SubtleShader account on Civitai and Tensor Art. Running it on other public or shared servers commercially (including other Civitai and Tensor Art accounts) requires a separate license or permission from the creator.
4. You may not sell this model or merges of this model.
5. You may not merge, share merges of, or share LoRA extractions incorporating Restricted versions. You may merge and share merges of Open versions.
6. When sharing the allowed merges of Open versions, as well as finetunes of them, you must apply these same additional license terms.
7. Redistribution of Restricted versions is prohibited. They may only be obtained from official sources designated by the creator.
8. All use-based restrictions of the CreativeML Open RAIL++M license apply to SDXL versions and their derivatives. All use-based restrictions of the FLUX Non-Commercial License apply to Klein versions and their derivatives.
9. Any restrictions in these additional terms may be waived only with explicit written permission from the creator.
TLDR: This is basically the same license as before (just stated more clearly), so nothing changes for the Pony, XL, Photo, Insta, Lust and ZT versions of Big Love. However, restrictions were added for Ultra, Hyper, Klein, Zia, Zuna, Eryn and Gwen versions (no merging, no redistribution). The "Share merges" condition below the license on the right hand side was only deactivated because of the restricted versions. Merging is still allowed for the open versions.
Many thanks to the creators of SDXL, Z, Qwen, Flux klein, Pony, bigAsp and Lustify for making their wonderful models available. Big thanks to RaymondLuxuryYacht for allowing me to train on his fabulous images.
Description
Please note: The Zuna versions of Big Love are pay-only and special license conditions apply. See description!
Zuna1 produces the most realistic genitals of any Z model to date. It is also quite good at lighting. Zuna1 is Z-Base but only needs 8 steps and cfg1. It looks more Z-Turbo-like & not like the typical Z-Base checkpoint. It is better at NSFW than Zia1.
FAQ
Comments (67)
I’m really impressed with your new Zia1 model! I purchased it via Tensorhub and have been testing it quite a bit.
The flexibility in terms of lighting and posing is excellent. Also, all of my LoRAs that were trained for ZT3 work perfectly with it, which is a huge plus.
A small tip for others: I’m using DPM-2 as the sampler and Simple as the scheduler, and I’m getting very nice results — high contrast, realistic images, and the skin no longer looks rubbery.
Keep up the great work!
Thanks for the review! Appreciated.
Hi ! Yeah, Zia1 is great and produces really cute girls out of the box... Speaking of cute girls, love your "Bavarian" ones ! ;-)
@BretChampagne Hi! Thanks a lot for the kind words – really appreciate it 😊
I’ve been meaning to release an update for quite a while. After the PONY version, I originally planned to move to a Big Love SDXL-version, but I was never fully satisfied with the results – especially in terms of flexibility, which just didn’t meet my expectations.
I’ll take this as motivation to dive back in and work on a Zia1/ZT3-based version next. Let’s see where that goes!
Heya, I bought the Klein checkpoint and I tried it with your workflows. For the T2I workflows I keep getting errors because of supposedly some wrong configuration in the KSampler (inspire) node -
variation_strength, increment, could not convert string to float: 'increment'
variation_method: '0' not in ['linear', 'slerp']
internal_seed, linear, invalid literal for int() with base 10: 'linear'
And I could guess, but it would be best if you could let me know what are the actual correct values since you got amazing results according to what you shared!
As for the I2I flows... Image editing... I'm getting photos in terrible quality! It is very hard to believe the images you shared were really created from the workflows embedded into them! There doesn't seem to even be any upscaling in those workflows... could you please share the actual workflows used to make those very impressive image edit results? I bought this checkpoint for the sole purpose of getting better results with I2I editing while keeping facial and body details, but currently it is completely unusable with the quality that I am getting...
Thank you!
Please download one of the Pruned versions not Full. You downloaded the base version of Klein2 that is meant for training (and requires 25 steps with cfg 5). Of course, it looks terrible when used with only 4 steps and cfg1.
You probably use an older version of the Inspire Pack. Please update it to fix the problem.
The embedded workflows are what I used. In two cases I used txt2img and then a image editing workflow. As an image cannot have 2 workflows, the txt2img one is embedded. Of course, the editing workflows require certain images to replicate the results which I don't provide. But they are meant as examples and Inspiration, not for reproduction.
@SubtleShader Hey, thanks for your fast response!
I double checked and I did try both the PRUNED bf16 and fp8 variants - both gave me poor results somehow...
As for the Inspire Pack, I just tried again after updating it and making sure it is up to date but the values in the fields itself are still mismatching and I'm getting the same errors even with the most recent version...
@talafek96 Right click on the node and choose "Fix Node" or "Recreate Node". You can also replace it with a normal sampler and recreate the connections & settings. I used the Inspire sampler because it offers variation seed for easily fixing problems.
What do you mean with poor results? Can you upload an image somewhere for me to see?
@SubtleShader I can send you privately?
@talafek96 So the main point of our conversation was: Only use high quality input images as the model will replicate the low quality in the output. If you do not have better images, try to improve them with SVR2, img2img or other methods.
WOW! Why didn't I try this sooner? Klein2 is absolutely fantastic! Great details and composition, fantastic pussies! Cum looks good, but placement is erratic. And I see what you're talking about the instability of text-to-image. Well, still fantastic and can't wait for the future updates!
Big Love Zuna1 Lora Training
If you want to use your lora with Big Love Zuna1, don't train on Z-Base or another Z-Base checkpoint. Train on Big Love Zuna1. Z is not as flexible as SDXL in this regard.
OneTrainer Instructions
Download the Full Model bf16 version above to train a lora with One-Trainer. Choose the z-Image LoRA preset in OneTrainer. On the model tab click the ... button next to Override Transformer / GGUF and choose the downloaded safetensors file. Add your concept and click Start Training.
Please note: OneTrainer has 512 resolution set on the training tab. If you have 24 GB VRAM, you may want to set resolution to 1024. If you want to train 1024 resolution with 16 GB VRAM, use the z-Image LoRA 8 GB preset and set resolution to 1024. In both cases it will take longer to train.
AI-Toolkit Instructions
Contact me for the Zuna1 de-turbo diffusers zip file. Unzip it. Choose the Z-Image option in AI-Toolkit. Then in the "Name or Path" text box replace Tongyi/Z-Image with the file path to the unzipped folder.
General formula for steps:
Image count 100 epochs / batch size
So, 25 images 100 epochs / batch size 2 = 1250 steps. Save every 10 epochs (every 125 steps here) and compare the last few saved versions.
With a batch size above 1 calculate a new learning rate as
lr = base_lr sqrt(batch size)
For example: 0.0001 sqrt(batch size 2) = 0.00014
Spectacular!
Thanks! I assume you mean Zuna1.
Ayo, Zimage/Flux look kinda plastic, but SDXL’s still straight fire. Pleaseee keep cookin’ more SDXL models, man! Great work!
Photo6, Ultra5 and Hyper2 coming up next! You can always run SDXL over Z & Klein for more realism. Z has the best anatomy and composition, Klein is great for image editing (even SDXL images). So all are useful in their own way.
I would pay good money for a Klein 4b version of bigLove. I know it's a dumber model but it's still incredibly capable. Not enough people train for it. 9B Klein license sucks so I stay away.
You mean that you have to pay BFL for commercial usage of 9B?
I tried training klein 4B (with the better license) but it is as good as untrainable. I probably tried more than anyone else and only gave up after 45 days. It breaks too quickly. Image editing deteriorates after 20 epochs. 20 epochs is not enough to teach it.
@SubtleShader ah. alright. sad to hear it failed. there's a 4b model here: unStable Revolution F2K 4b alpha
that's pretty decent but that being said I haven't really pushed it to do anything that isn't softcore. Was hoping you would have more success
I also was only really using it as a polish pass on top of an existing gen. not as the true base gen. I'd settle for it to be just a good SDXL image-gen / i2i model. Klein does fine details and texture well.
@QualityControl The original 4B is fine. Just does not do NSFW. 9B is still a lot better. The quality discepancy between my klein 4B and 9B version was huge. Made no sense to release a 4B version.
@SubtleShader yeah - but my uses are commercial so 9b is dead to me. Thanks for the insight
@QualityControl Try Qwen Edit or wait for Z Omni/Edit
@SubtleShader yeha. my main pipeline is qwen edit. but I'm always shopping around. currently running qwen + a klein 4b polish pass to crisp up details. Works well enough but nsfw details could always be better. Hence my questions about 4b training here
Zuna1 is HERE!! YAY!! Downloading now!! Thank you my friend!! :-D
Big Love Zia1 Lora Training
After some tests it turned out that training loras on ZT3 works much better than training on the original Zia1 training files.
Here are the ZT3 training instructions!
Original Zia1 training notes in case you disagree...
How does zuna work with character loras?
I made a few Zuna character Loras with AI-Toolkit, they look very nice!
@GenoMachino Did you have to train them specifically on this model?
@mobdik17378 Sorry, just catching up with the comments! I don't know if I "had to" create them with Zuna, but I did and they worked great. I imagine other Z-Base loras would also work with Zuna... my Z-image Turbo Loras are not compatible with Zuna. If you have other Z-Base loras give them a try, I'm interested to hear if they work.
Could you elaborate on the difference between the BF16 pruned vs Full for klein? is it purely just pruning or is it merged with the r128 turbo lora or something like that.
if so the full model + the turbo lora should be "relatively" similar in results if someone wants to have an experience more closer to the base model right?
Full is only for training a lora and has no distillation so needs 20+ steps instead of just 4. Base/Full plus turbo lora is similar but lower quality than just using pruned/distilled. Of course, you can use turbo lora below 1.0 strength with base and use 4-12 steps. But I optimized it for 4 steps. With more steps you get more texture but it usually is too much. Feel free to experiment though.
Big Love Ultra1 For Free
I made Ultra1 freely downloadable. Click here and enjoy its hi-res images. It is still unique in what it offers compared to most other SDXL models.
Ultra1 was my first restricted model in Dec 2025. I'm still collecting data and will come up with a plan to gradually make the others available. With only 2 weeks of Early Access my work on Big Love would not be sustainable.
Great models and loras, love your work! Any chance you are working on a Flux2-dev version?
@username2313 I plan to check out Flux 1 dev. Cannot guarantee that it will convince me. It was always hard to train on NSFW. Maybe that changed. Flux 2 dev is a monster. Too much effort & costs.
Are you trying to make a living from this hobby?
@fcukit It would not be possible to deliver this quality this fast if I would not do it full-time.
Very kind, thank you! 🩷⭐
It takes lots of time, effort and also costs to do this stuff.. so understandable that gotta have some forms of funding to keep going. Community gratitude warms the soul, but community gratitude and a few dollars/euro warms the soul and also pays maybe the electric bill and costly GPU upgrades and so on. 👍
I bought early access to Klein2 for 20k ... and now I can't download the model now that it's marked as generation only...
Please contact me privately via Civitai chat
Hello everyone, first of all a big thank you to @SubtleShader for all these amazing models
I’d love to get your feedback and discuss training LoRA on Big Love (ZT3, ZIA, ZUNA). On my side, I’m using Ai-toolkit.
Is anyone else using it and willing to share some tips or settings, whether for the dataset or Ai-toolkit itself? Also, do you have any testing protocols in ComfyUI?
Thanks!
Expand the Big Love description above and read these sections: ComfyUI Workflows & How to train a (character) lora.
I have my testing protocols for checkpoints but not for loras. I rarely do loras. Best if you do at least 20 images with identical prompts & seeds on multiple lora versions. Then compare.
Thanks for the feedback—I had already read the recommendations, but I was more looking for real-world experience rather than doing lots of long and tedious tests :D
For example:
In Ai-toolkit, which optimizer should I choose? Adafactor, Adam8bit? I’ve even seen that you can switch to Prodigy in the JSON...
I selected 512, 768, and 1024—does that serve any purpose?
For the dataset, should I use trigger only or trigger + WD14 captions?
And for testing, at what LoRA weight should I avoid going beyond so as not to “overwrite” the model?
@adri510 Use Adam8bit. Adafactor makes more sense in OneTrainer with Fuse Backpass for saving a lot of VRAM. AI-Toolkit does not have that option. I saw no real difference between optimizers so far except for speed and VRAM. 512px is fine for lora unless you want it to learn your character's skin texture. You can use more if your VRAM allows it, but will also be slower. I would use trigger and captions/tags. Use the same standard description when you prompt. Lora strength of 0.75 is usually fine with Z. Test more or less and check image quality & likeness.
Alright, thanks for the info. I’ll run some more tests then.
In my case, using a dataset + trigger only, the LoRA weight needs to be around 0.95 or even 1.0 to start getting a noticeable resemblance. And even then, you could say that about 4 out of 10 images still aren’t similar enough—it really depends on the seed.
@SubtleShader If I want to train at 1024 in AI Toolkit, should I also keep 512 and 768 checked?
@adri510 AI-Toolkit does not upscale images, so better do that if you have low res images. Will likely make.no difference. 1024px takes 4x longer to train than 512px.
Alright, I’ll crop to 1:1 to get 1024×1024 images and set the resolution to 1024 only in the AI toolkit. Thank you.
@adri510 No, don't crop anything. That is BS from SD1.5 times. Some myths never die.
Pfff, it's true that you read all kinds of stuff on the internet lol. Anyway, the results at 512 aren't very good... I have a training in progress at 1024, wait & see (without cropping :D)
@adri510 90% of what I read about lora training did not work for me. Often advice is too generalized from very specific cases.
I don’t know where I’m messing up, but at 0.75 it doesn’t look like the subject at all :/. I must be preparing my dataset incorrectly—I don’t see any other explanation. Still, I do have between 10 and 30 fairly varied photos (upper body, full body, portrait).
@adri510 Maybe you do not train enough epochs/steps?
@SubtleShader Yet I followed your recommendations—unless I made a mistake. I have 22 photos, I run training for 1100 steps (batch size 2, LR 0.00014), and I save every 110 steps.
@adri510 Try 2200 steps. 100 epochs.
@SubtleShader Is your example not correct then? oO
General formula for steps:
Image count × 100 epochs ÷ batch size
So, 25 images × 100 epochs ÷ batch size of 2 = 1250 steps. Save every 10 epochs (every 125 steps here) and compare the last few saved versions.
I’ll run a test tomorrow and give feedback.
Thanks
@adri510 Sorry, overlooked the batch size 2. 1100 steps is correct.
@SubtleShader yes, so I do have an issue with my dataset then. At 0.75 the resemblance isn’t great. On the other hand, I pushed it to 2200 steps at 0.75, and in the final steps the resemblance isn’t too bad with a fixed seed and a LoRA weight around 0.8, but yeah the facial expression is frozen and there are visible artifacts in the generated images…
Hi, quick update. I ran several tests with the same dataset, with and without captioning, all at 1024. The tests were done on ZT3, ZIA1, and ZTURBO (with adapter v2).
Result: with captioning, ZT3 and ZIA1 produce even worse generations than using only the trigger; the resemblance is only good from 0.9 to 1.0. On the other hand, with ZTURBO, using captioning or not makes very little difference, but I can lower the LoRA weight to 0.75–0.8 and the resemblance is much better. Of course, I need to use the base ZTURBO checkpoint.
@SubtleShader Hello, while rebuilding my dataset I started wondering about captioning. Do you use both JoyCaption and tags, or just one of them? Is it better to be very detailed (person, hairstyle, hair and eye color, environment, etc.), or to keep things concise?
@adri510 I don't do character loras very often. I suggest you add a detailed description of her and also use the same description when prompting.
Question,is Zimage and Flux 2 the king of realism best models while SDXL/Pony is lower quality overall,
SDXL still looks the most realistic to me. Z is catching up and has superior composition. Klein is best for editing and just copies the style that you give it. The photorealistic versions of Pony were always somewhat limited in expression.
@SubtleShader I agree, yet I still want to give Kudos to your amazing Klein checkpoints and I am not only happy with the editing but also with the generation results. You really did something great here.
@SaburoDio Thanks. I'm.planning to make Klein3 more flexible with posing. Should improve txt2img as well as editing. Will also try to reduce anatomy problems.
My first ever comment on Civit, but I have to write this: Zuna1 is by far the best ZiB model I've used so far, SFW and NSFW.
There isn't _any_ negative impact on my Z-base trained character Loras, except that Zuna1 enhances their realism reagarding skin texture and skin tones, very nice. Also very nice balanced colors, yet a touch to much contrast for my personal taste).
Brilliant work, thank you!
How do you deal with detailless skin?