sdupdates
A mega collection of all resources and news related to Stable Diffusion. Focused around AUTOMATIC1111's webui (https://github.com/AUTOMATIC1111/stable-diffusion-webui) (by questianon)
sdupdatesl
By questianon
sdupdates | sdupdatesl | |
---|---|---|
13 | 2 | |
512 | - | |
- | - | |
4.9 | - | |
about 1 year ago | - | |
The Unlicense | - |
The number of mentions indicates the total number of mentions that we've tracked plus the number of user suggested alternatives.
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
sdupdates
Posts with mentions or reviews of sdupdates.
We have used some of these posts to build our list of alternatives
and similar projects. The last one was on 2022-12-06.
-
What happened while I was gone?
This will get you caught up to the 2nd of March: https://github.com/questianon/sdupdates
- best way to learn the latest goings on?
- This technology has grown so big! where to start?
-
We really need a new SD resource goldmine
From sdupdates3 and GitHub.
-
Stable Hands - Hands comparison in 23 models
Apparently, Stability AI and other companies are testing distributed AI training, from what I read on SD Update GitHub.
- Rare Tokens For DreamBooth Training Stable Diffusion...
-
Getting into the warpmobile
There is also an Unstable Diffusion Discord server with much information. I found a big markdown list: - https://rentry.org/sdupdates/ - https://rentry.org/sdupdates2 - https://github.com/questianon/sdupdates
- [Certain Magical Index] Services no longer required
- Stable Diffusion links from around November 3, 2022 that I collected for further processing
-
Stable Diffusion resources (pile of links and up-to-date news)
See also the GitHub repository for tracking changes; https://github.com/questianon/sdupdates
sdupdatesl
Posts with mentions or reviews of sdupdatesl.
We have used some of these posts to build our list of alternatives
and similar projects. The last one was on 2022-12-06.
-
Stable Hands - Hands comparison in 23 models
Tested models (recipes and references 1,2,3): model_1.4_full.ckpt [7460a6fa] v1-5-pruned-emaonly.ckpt [3e16efc8] wd1-2_sd1-4_merged.ckpt [2037c511] wd-v1-3-float32.ckpt [4470c325] Berry's Mix.ckpt [579c005f] Elysium_V1.ckpt [44bf0551] f222.ckpt [44bf0551] r34_e4.ckpt [6e45cf2c] novelai-animefull-final-pruned-model.ckpt [925997e9] (NAI-FULL_0.8+TRI-ART-2_0.2)_0.9+TRIN-ART-IT4-V1_0.1.ckpt [62b2fadd] twintrin-5050-chinai-65-sd15-20.ckpt [55fd32ec] Anything-V3.0-pruned-fp32.ckpt [1a7df6b8] SXD-v0.8-pruned.ckpt [487e8ff0] SXD+Anything_0.3.ckpt [b268912b] cafe-instagram-unofficial-test-epoch-9-140k-images-fp32.ckpt [50b987ae] Elysium_Anime_V1.ckpt [44bf0551] novelai-animesfw-final-pruned-model.ckpt [1d4a34af] HassanBlend1.4_Safe.safetensors [ee1aa2b8] copeseethemald_style_10000.ckpt [32186669] samdoesbimbos_midjourney_f222_Chimera_Ghibli_PitBimboAgain_SXD_w015.safetensors [d61f229a] SamDoesArt-V3.ckpt [d1c6e030] 768-v-ema.ckpt [2c02b20a] 512-base-ema.ckpt [09dd2ae4]
-
Stable Hands - Hands Comparison in 23 models
**Intro** I wanted to test hands and fingers for as many models as I could test, to see if certain patterns are appearing and if certain models performed extremely well or poorly in comparison to others. To make the test as consistent as possible I used the same original picture in img2imd, with different denoising strengths and different seeds. **Analysis of results** Particularly bad results: Worst hands and worst images were created by SXD (I had to censor them). R34 produced the second worst results. Both of these models also significantly deformed the face, which other models didn't do. SD v2 512px couldn't connect the elbow many times, therefore it got a third place from the end. WD 1.3 produced bad results too. Other models didn't show consistently good results, with extra, missing, deformed, finders, wrong direction, wrong position of rind, mashed fingers, and wrong side of the hand. If comparing only vanilla SD v1.4 vs v1.5 vs v2.0, improvement is very minor. Hasan blend v1.4 and SameDoesArt didn't show significant and consistent improvement if any at all. I'm not sure if this is statistically true, but I think that the right hand (that appears on the left side of the image) is generated in a slightly better way than the left hand (that appears on the right side of the image). If this effect is indeed real, I can think of two ways to explain it. First, the differences in position, angle, and state of the right hand vs left hand, in this particular image, make the right hand easier for SD to "understand" and recreate. The second explanation is a bias in the training data, due to the fact that 90% of the human population is right-handed. I'm not sure which models were using mirroring of training data if any at all. If my observation, regarding a better right hand, is real, it gives me hope that better training could produce better hands sooner than later. Initially, I wanted to create a methodic count of hands and fingers, with scores. But very fast I understood that this would take me way too much time, and I wasn't sure that I could keep my sanity in the end, so I present the results as is. Anyone can judge and decide for themselves. I hope that img2img comparisons could be used to produce better comparisons than txt2image, and easier ways to diagnose changes between models. Possibly a better way to diagnose the results of such comparison would be a crowdsourced effort, in a double-blind experiment. My idea to do so is to show 2 random photos from the results (without model names), and allow only to select which picture the hands are better. This would allow to rank all images and associate scores for each model. **Tested models** (recipes and references [1](https://rentry.org/sdmodels),[2](https://rentry.org/lftbl),[3](https://github.com/questianon/sdupdatesl)): model_1.4_full.ckpt [7460a6fa] v1-5-pruned-emaonly.ckpt [3e16efc8] wd1-2_sd1-4_merged.ckpt [2037c511] wd-v1-3-float32.ckpt [4470c325] Berry's Mix.ckpt [579c005f] Elysium_V1.ckpt [44bf0551] f222.ckpt [44bf0551] r34_e4.ckpt [6e45cf2c] novelai-animefull-final-pruned-model.ckpt [925997e9] (NAI-FULL_0.8+TRI-ART-2_0.2)_0.9+TRIN-ART-IT4-V1_0.1.ckpt [62b2fadd] twintrin-5050-chinai-65-sd15-20.ckpt [55fd32ec] Anything-V3.0-pruned-fp32.ckpt [1a7df6b8] SXD-v0.8-pruned.ckpt [487e8ff0] SXD+Anything_0.3.ckpt [b268912b] cafe-instagram-unofficial-test-epoch-9-140k-images-fp32.ckpt [50b987ae] Elysium_Anime_V1.ckpt [44bf0551] novelai-animesfw-final-pruned-model.ckpt [1d4a34af] HassanBlend1.4_Safe.safetensors [ee1aa2b8] copeseethemald_style_10000.ckpt [32186669] samdoesbimbos_midjourney_f222_Chimera_Ghibli_PitBimboAgain_SXD_w015.safetensors [d61f229a] SamDoesArt-V3.ckpt [d1c6e030] 768-v-ema.ckpt [2c02b20a] 512-base-ema.ckpt [09dd2ae4] **Setup:** Using the same(keywords needed by certain dreambooth models added when needed) prompt. Each model was tested with 4 denoising strength and 4 seeds. Tested seeds: 747450140-747450143 Denoising strength (X-axis): 0.55,0.65,0.75,0.85 Used script: X/Y Plot All models(besides [09dd2ae4] and [2c02b20a]) tested with VAE: vae-ft-mse-840000-ema-pruned.vae. Keywords "SamDoesArt" added to [d1c6e030] and [d61f229a]. Keywords "copeseethemald style" added to [32186669]. **Prompt:** Jennifer Lawrence, woman, solo focus, sitting, black dress, sleeveless dress, short hair, blond hair, beautiful eyes, detailed eyes, expressive eyes, perfect hands, cleavage, small breast, hands up, open hands, hands palms, soft lighting, reflections, office, night, necklace, earrings, eyelashes, eyeshadow, lips, lipstick, jewelry, makeup, nose, parted lips, white nail polish, simple background,blurred background, masterpiece, best quality, highest quality, extremely detailed, focused, realistic, hyper realistic, extremely detailed, ultra realistic, photo realistic, sharp focus, absurdres, photography, insanely detailed, intricate **Negative prompt:** nipples,undressed,topless,naked,nude,black and white,border, frame, lowres, jpeg artifacts, low quality, worst quality, poorly drawn, error, abstract, asymmetrical, cropped, disconnected, duplicate, extra, missing, signature, text, username, watermark, amputee, bad anatomy, deformed, disfigured, disproportionate, fused, malformed, morbid, mutated, mutation, mutilated, bad hand,extra limbs,missing limb,extra leg,missing leg,missing hand,extra hand Steps: 60, Sampler: DPM++ SDE Karras, CFG scale: 13, Seed: 747450140, Size: 768x1024, ENSD: 31337, Mask blur: 4
What are some alternatives?
When comparing sdupdates and sdupdatesl you can also consider the following projects:
Animator - Animator script for SD Web UI
tutorials - A collection of tutorials about training and generating with Stable Diffusion.
dreambooth-tokens
stable-diffusion-webui - Stable Diffusion web UI
hs610-info - Information on HUION's HS610 tablet
stable-diffusion - A latent text-to-image diffusion model