Sign In

Resources Worth Checking Out

106

Mar 23, 2025

(Updated: a day ago)

resource guide
Resources Worth Checking Out

SDXL/Pony

If you don't have the system requirements for FLUX, or don't want long generation times; then start here - Pretty much a LoRA to cover everything you could want already out there. Start with Pony Realism & CyberRealistic Pony first if you're new to AI.

S+ Tier

  1. Pony Realism 🔮 https://civitai.com/models/372465/pony-realism My favourite SDXL/Pony checkpoint. I like to rock both v2.2 & v2.1 (I think they're both worth checking out, but def start with v2.2, as 2.1 seems to produce more anomalies. I still regularly switch back to using 2.1 regardless though). Put this at the start of your positive prompts to instantly improve results "score_9, score_8_up, score_7_up"

  2. CyberRealistic Pony https://civitai.com/models/443821 (Realistic, but less versatile than Pony Realism. Quite similar though, & very good).

  3. DucHaiten-Pony-Real https://civitai.com/models/477851/duchaiten-pony-real (Still 'realistic', but heavily stylized, & I love the style. It's biased more towards Asians, but this actually works surprisingly well to its favour when you mix it with caucasian LoRAs or ReActor composite image blending (where you blend a bunch of different face models together).

  4. DucHaiten-Pony-XL (no-score) https://civitai.com/models/376450/duchaiten-pony-xl-no-score (Variant) I still really like v5.2, so maybe check that out after the newest versions if you like those.

  5. MeichiDarkMix_Reload https://civitai.com/models/643227/meichidarkmixreload Similar-ish (also heavily stylized like DucHaiten). I've had great results regarding versatility. I prefer this to DucHaiten, it also produces less deformaties/weird anatomy/busted fingers, etc. Also produces less abstract, sharper detailed images (especially backgrounds).

  6. 2DN https://civitai.com/models/520661?modelVersionId=933040 (I should preface this by saying I don't really like anime typically, but in terms of less realistic checkpoints; I find I often quite like the outputs on this one). If you're new; less realism usually = more forgiving/lax tweaking to achieve good results).

  7. GDP_PonyXL https://civitai.com/models/571319/gdpponyxl (Less realistic, stylized checkpoint). Similar to above.

  8. Dr. Pibb PonyXL https://civitai.com/models/499918/dr-pibb-ponyxl (Less realistic again, but this one just 'works'. The image outputs just look good without much effort, & it fills in all the blanks in the image prompt really well, & without tons of weird stuff going on in the image).

  9. Real Dream https://civitai.com/models/153568/real-dream (Similar to Pony Realism, but not as good).

  10. Realism_By_Stable_Yogi https://civitai.com/models/166609/realismbystableyogi Orientated towards realism, but I often find the outputs look slightly weird as a result/uncanny valley effect). Make sure you read guide/download Stable_Yogis_PDXL_Negatives, etc, otherwise outputs will suffer badly. Good inpainting.

  11. Jib Mix Realistic XL https://civitai.com/models/194768/jib-mix-realistic-xl (I haven't had the chance to use this one much, but I'm liking what I'm seeing so far. Great at mythical fantasy & surreal content)

  12. epiCRealism XL https://civitai.com/models/277058/epicrealism-xl

  13. Speciosa Realistica https://civitai.com/models/488361/speciosa-realistica (Very unique - You may fall in love with the style). Asians, Stylized Realism.

  14. Juggernaut XL https://civitai.com/models/133005/juggernaut-xl (Realistic / Apparently the most downloaded checkpoint ever).

Sadly there's a bunch of other models in my permanent collection I'd link, but the creators removed them ;(

A Tier

(Either limited in some way checkpoints, or just not as good as the S Tier list above, or they just lack that special something, something)

  1. ZavyChromaXL https://civitai.com/models/119229/zavychromaxl (Realism capable, Surreal + Fantasy Art - Doesn't do adult content)

  2. DreamShaper XL https://civitai.com/models/112902/dreamshaper-xl (Realism capable, Mythical, Fantasy Art - Doesn't do adult content)

  3. PowerPuffMix https://civitai.com/models/805817/powerpuffmix (Most similar to Pony Realism. I don't have anything particularly meaningful to say about it... it's just good)

  4. Jib Mix Pony Realistic https://civitai.com/models/515034/jib-mix-pony-realistic (Realism - This probably belongs more in S Tier, but I just never really seem to personally want to use it/prefer other options). J1B stuff is generally great across the board though.

  5. Crystal Clear XL https://civitai.com/models/122822/crystal-clear-xl (Realistic, Midjourney-looking, Fantasy-ish)

  6. ForRealXL https://civitai.com/models/432244/forrealxl-v10

  7. IDENTITY DISORDER (PONYXL) https://civitai.com/models/362232/identity-disorder-ponyxl-v25 (Somewhat abandoned/resources removed)

  8. Phony Pony Pepperoni Evolution https://civitai.com/models/1196991/phony-pony-pepperoni-evolution (I really liked phonyponypepperoni_v4 especially, but sadly the creator has removed a lot of their checkpoints from Civitai. You can still probably easily find them on other sites though. Very unique style).

  9. Magica Pony Realism https://civitai.com/models/560814/magica-pony-realism (Fantasy)

  10. Pony Diffusion V6 XL https://civitai.com/models/257749/pony-diffusion-v6-xl (I really don't care for this checkpoint. I think the base model is rather poor by itself, & I'm not a fan of 'anime' either...but it's very popular, so I'd be remiss if I didn't at least mention it. I'm inclined to think its popularity is due to a massive furry following though, which isn't something I have any interest in). I'd advise not starting with this if you're new; it's fickle generating consistently good images.

LoRAs (Essentials)

FLUX

  1. https://civitai.com/models/686814 Jib Mix Flux (If I'm using FLUX, it's probably this one most of the time).

  2. https://civitai.com/models/778691 Fluxmania

  3. https://civitai.com/models/978314/ultrareal-fine-tune UltraReal Fine-Tune

  4. https://civitai.com/models/118111/copax-timeless Copax TimeLess

  5. https://civitai.com/models/938037/copax-flux-realistic Copax flux realistic

  6. https://civitai.com/models/843551/fluxdev-dedistilled FLUX.DEV DEDISTILLED (Dedistilled = expect generation times to increase noticeably)

  7. https://civitai.com/models/720719/project-gaia-flux1d-uncensored Project Gaia Flux1.D (Uncensored)

  8. https://civitai.com/models/1094778/hyper-flux-dedistilled Hyper Flux Dedistilled (Dedistilled FLUX Hyper variant) = expect those added generation times to go back down again.

  9. https://civitai.com/models/662112/gguf-flux-unchained GGUF: Flux Unchained

  10. https://civitai.com/models/141592/pixelwave PixelWave

  11. https://civitai.com/models/684924/real-horny-pro-v3 Real Horny Pro V3 (Dated, & should be the last option you check out on this FLUX list, but still interesting outputs/aesthetic)

Chroma

(Uncensored NSFW FLUX spin-off)

An officially licensed open-source FLUX spin-off model, built on the FLUX.1-schnell architecture. It's uncensored & Open, & way better with licensing with regards to anything generated. Images generated with FLUX Chroma are covered under the Apache 2.0 license, like with the model itself. A lot of money was spent training Chroma, & it's quite a big deal compared to something like Krea (which is only really just a artistic style alteration).

https://civitai.com/models/1330309?modelVersionId=2104870

https://huggingface.co/lodestones/Chroma

SD 3.5

Don't bother with SD3, it sucks. SD3.5 is also a bit of a running joke/meme as a result; but I think SD3.5 medium can actually generate some really good image ouputs (much faster than FLUX/lower system requirements also). Third party LoRA support suffers due to the lingering stigma though (& iffy licensing). It's way more versatile/flexible than XL with regards to using weird resolutions/aspect ratios, etc, & not getting broken anatomy/weird deformed hands as a result. In terms of the actual outputs I think it's actually really capable (at times great, even), but its 'leper' status probably means you don't want to invest in something with poor 3rd party community support treating it like a pariah.

I generated this in SD 3.5 medium https://civitai.com/images/45274249

It's a curious thing to ponder whether Stable Diffusion will really be able to recover from FLUX stealing the limelight.

SD 3.5 has recently undergone some community alterations in bypassing its overzealous censorship, & reversing it.  Perhaps it will breathe some new life into it.

Illustrious

More open, & very versatile, with usually better prompt adherence than SDXL/Pony. You're more likely to be able to prompt a random popular character, & get a good likeness back without even needing to use a LoRA from my experience. Fast generation times, too. Seems to be the best option in terms of business licensing. Great third party community support. It scales/performs really well at high resolutions without needing to use hires fix, as well.

(SDXL/Pony LoRAs aren't compatible with Illustrious checkpoints btw).

  1. https://civitai.com/models/959736/rillusm Rillusm by ZyloO

  2. https://civitai.com/models/1125067/cyberillustrious-or-cyberrealistic CyberIllustrious | CyberRealistic

  3. https://civitai.com/models/1152359/reijil reijIL

  4. https://civitai.com/models/1053104/illustreijl IllustreijL

  5. https://civitai.com/models/1025051/illustrij Illustrij

  6. https://civitai.com/models/1772230   Illustrij EVO

  7. https://civitai.com/models/1123090/posilustrij posILustrij

  8. https://civitai.com/models/917668/perfectrsbmix-illustrious-real-anime-sfw-nsfw PerfectRSBmix (Illustrious Real-Anime SFW-NSFW) (This one has really caught my eye lately)

  9. https://civitai.com/models/1309123/shift Shift

  10. https://civitai.com/models/926443/ntr-mix-or-illustrious-xl-or-noob-xl NTR MIX | illustrious-XL | Noob-XL

  11. https://civitai.com/models/1180280/one-illustrious-mix One illustrious mix

  12. https://civitai.com/models/1239956 DarkMix Mojito Illustrious - 3D Semi-Realistic

  13. https://civitai.com/models/1030120/alchemist-mix-illustrious-realism Alchemist Mix (Illustrious Realism)

  14. https://civitai.com/models/1058624 AnBan_Illus

  15. https://civitai.com/models/1228973/illustrioussat-ior-imitation-of-reality IllustriousSat-IOR (Imitation Of Reality)

  16. https://civitai.com/models/1094522/circusmix CircusMix

  17. https://civitai.com/models/860856/illustriousxlmmmix illustriousXL_MMMix

  18. https://civitai.com/models/1110783/ilustmix iLustMix

  19. https://civitai.com/models/1224788/prefect-illustrious-xl Prefect illustrious XL

  20. https://civitai.com/models/1078146/steinillustrious SteinIllustrious

  21. https://civitai.com/models/1196991?modelVersionId=1360903 Phony Pony Pepperoni Evolution (RealillustriousPhonyPony)

  22. https://civitai.com/models/520661?modelVersionId=1175889 2DN

  23. https://civitai.com/models/140272 Hassaku XL (Illustrious)

  24. https://civitai.com/models/827184/wai-nsfw-illustrious-sdxl WAI-NSFW-illustrious-SDXL

  25. https://civitai.com/models/795765/illustrious-xl Illustrious-XL (Not a fan, but it's popular)

Illustrious LoRAs

SD 1.5

If you're on an entry level system, this might be your only realistic option to generate images locally. That, or you might get lucky with sub 5 second image generations that can still look good at times. Otherwise, avoid it if you're new; because SD 1.5 is a headache to deal with, & will constantly output broken images with weird anatomy/multiple hands, etc, causing new users needless frustration (consider 1.5 to be like the Dark Souls of checkpoints, with FLUX being the equivalent to easy mode in terms of getting good results easily lol).

Does comic styles well, & there's some obscure/unique LoRAs out there no one bothered porting over to anything else. You can also spam the quick image generations as a basic template to get roughly what you want, then remaster again using something like FLUX to expand complexity & detail (a bit like hires fix).

If you notice most of your images are looking trash, then try sticking to these resolutions; 512x512, 768x768 (square), 768x512, 768x576 (you can switch these around to go from landscape/portrait), 912x512 (16:9 widescreen). Widescreen resolutions/aspect ratios tend to cause it to double characters, or cause a mess with anatomy. SD1.5 is just very fussy & prone to breaking easily, but avoid using weird custom resolutions/aspect ratios to avoid creating more issues! I feel the resolutions SD1.5 was trained on too restrictive & low, this is why SDXL is such an improvment (optimized around 1024x1024 pixels).

  1. https://civitai.com/models/22922?modelVersionId=72396 Lyriel

  2. https://civitai.com/models/4384/dreamshaper DreamShaper

  3. https://civitai.com/models/151774/westerncomicmix-by-mrmonster-or-model WesternComicMix by Mr_Monster | Model

  4. https://civitai.com/models/240267/realistic-comic-book Realistic Comic Book

  5. https://civitai.com/models/4201/realistic-vision-v60-b1 Realistic Vision V6.0 B1

  6. https://civitai.com/models/81458/absolutereality AbsoluteReality

  7. https://civitai.com/models/147927/ownwaifureal3dmix OwnWaifu_Real3dmix

Hunyuan

Video Generator - But set it to one frame, & you have a very realistic, uncensored image generator comparable to FLUX in terms of realism (which is what it's based on). LoRA support is now quite widespread, & really taken off.

With temporal tiling support, the system prerequisites for using this are now massively reduced. If you can run FLUX comfortably, then you can probably run this (at least some quantized variant of it). Still, expect to spend 5-10mins generating a few seconds long video on a 12GB Nvidia card. Frame interpolation (similar to Nvidia DLSS Frame Generation, but for video AI) helps with generation times.

It also has a "Fast" variant for quicker generating.

This has undergone some major advancements. Among them; Hunyuan Custom.

https://github.com/Tencent-Hunyuan/HunyuanCustom

https://huggingface.co/tencent/HunyuanCustom

WAN 2.2

A groundbreaking open-source text-to-video & image-to-video generation framework that's optimised to work on lower end hardware (locally). Think of it as like SwarmUI, ComfyiUI, AUTOMATIC1111, etc; but specifically optimised & built for AI Video Generation instead.

https://github.com/Wan-Video/Wan2.2 https://huggingface.co/Wan-AI

T2V-A14B Text-to-Video https://huggingface.co/Wan-AI/Wan2.2-T2V-A14B-Diffusers

I2V-A14B Image-to-Video https://huggingface.co/Wan-AI/Wan2.2-I2V-A14B-Diffusers

TI2V-5B Text+Image-to-Video https://huggingface.co/Wan-AI/Wan2.2-T2V-A14B

Looking at those links, you're probably naturally wondering "Why wouldn't I just download the Text+Image-to-Video combined....? Well, the answer is it that it does both of them to a slightly lower quality than the standalone versions. You will still need a lot of VRAM btw, but with GGUF/quantized variants; you can still probably find something reasonable to suit your hardware/needs.

As it stands right now, this whole platform isn't very accessible or user friendly btw. If you struggled to set up something like ComfyiUI before (on your own machine locally), then you're definitely going to struggle here. If you're patient, then it's not so bad; but you will have to be very comfortable with computers to get this working for the time being. Though I'm sure there's detailed guides or third-party 'one-click installers' already out there if you really want to jump into the action, otherwise just wait for something easier to materialize.

Upscalers

  • Remacri https://civitai.com/models/147759/remacri (Old, but gold upscaling option. It still looks great).

  • https://openmodeldb.info/   Great site to download archived upscalers.  Drop the downloaded file into ESRGAN folder inside your Swarm/Forge/AUTOMATIC1111 install directory (exact location will depend on what you're using). 

These are trained for specific tasks!  Trying to use something suited to anime or photo-realism will generally yield poor or sub-optimal results if used on the incorrect thing.

Below, I will link some of my favourites/popular ones.  Upscaling images is a great way to increase detail/resolution when your hardware is hitting the VRAM ceiling (it's also quicker).  Personally, I like to experiment with spamming low footprint images, then cherry picking the best for hi-res upgrades/edits.

These aren't big downloads, so just experiment using different ones on the same image seed/prompt.

  • SUPIR

Flux Tools

Consisting of Flux Fill, Depth, Canny & Redux.

Especially FLUX Fill (One of the best inpainting tools around). You can use this suite to replicate artistic styles, fill in missing details, or massively alter existing ones (among other things). This really requires an entire dedicated guide though, so I won't go into details here.

Redux - Can be used to remix images.

Flux Kontext

https://huggingface.co/black-forest-labs/FLUX.1-Kontext-dev  It's kind of like photoshop for image editing, but using prompts instead to make edits/changes.  If you quickly want to change hair color, add a moustache, change the background, etc; then this is going to be useful & particularly well suited.

Style LoRAs

XL/Pony (LoRAs)

Character

FLUX (LoRAs)

Misc.

Krita

Photoshop alternative basically (but without the overpriced DRM/subscription BS).

Open Source, & no strings attached free (actually designed more around digital painting, but a powerful photo editor, nonetheless). But what's noteworthy is that you can get plugins to enable Stable Diffusion/A.I. support from within the photo editing enviroment itself! If you ever thought inpainting was a bit **** in whatever you're currently using, then this will enable really advanced fine control to edit.

Matrix-Game 2.0

https://huggingface.co/Skywork/Matrix-Game-2.0 Matrix-Game 2.0: An Open-Source, Real-Time, & Streaming Interactive World Model

Really not useful or practical at all, but I thought I'd include it because it's dope as fk :D

It's basically a real-time AI Video generator that's tantamount to a video game stream that can be played in theoretically 'real-time' on local hardware. Essentially, it AI generates video streams of computer games it's trained on (Like Grand Theft Auto V), that can actually be played 'real-time'!  For now, it's a laggy as hell slide show requiring insane hardware, but it demonstrates proof of concept & theoretical potential.  With dedicated hardware acceleration built around it one day...

All the DLSS/Frame Gen haters that be like "fake frames suck, AI sucks" will probably have a seizure when this is eventually implemented mainstream lol. Nvidia are already working on implementing dynamic AI generation into computer games; albeit compartmentalised to things like in-game character faces. That combined with LLMs should create some extrememly realistic NPCs in the not too distant future.

These are definitely exciting times we live in ^^ Imagine a future in which you can create your own computer games like an image prompt.... It's a far flung future, indeed, but an exciting one to potentially imagine nonetheless.

I designed this article for people to quickly middle mouse click through groups of links/tabs at a time, presented here a rough personal ranked order/set of relevant categories; but here they are presented again (Civitai links only) in a thumbnail collection (without notes/ranking/sometimes bad picture thumbnails that don't do them justice). https://civitai.com/collections/8511267

Checkpoint Collection (all mixed together FLUX/SDXL/Pony/SD3.5/SD1.5) https://civitai.com/collections/8511178

(This whole guide is an on-going work in progress, continually updated).

106

Comments