/ldg/ - Local Diffusion General
Anonymous 01/18/25(Sat)12:26:30 | 356 comments | 113 images | 🔒 Locked
Narrow Minded Edition
Discussion of Free and Open-Source Diffusion models.
Last Time: >>103934948
Newfrien Guide: https://rentry.org/ldg-lazy-getting-started-guide
>UI
Metastable: https://metastable.studio
SwarmUI: https://github.com/mcmonkeyprojects/SwarmUI
Forge: https://github.com/lllyasviel/stable-diffusion-webui-forge
reForge: https://github.com/Panchovix/stable-diffusion-webui-reForge
ComfyUI: https://github.com/comfyanonymous/ComfyUI
InvokeAI: https://github.com/invoke-ai/InvokeAI
>Models, LoRAs, & Upscalers
https://civitai.com
https://tensor.art/
https://openmodeldb.info
>Training
https://github.com/Nerogar/OneTrainer
https://github.com/kohya-ss/sd-scripts/tree/sd3
https://github.com/derrian-distro/LoRA_Easy_Training_Scripts
>HunyuanVideo
Comfy: https://github.com/kijai/ComfyUI-HunyuanVideoWrapper/
Windows: https://rentry.org/crhcqq54
Training: https://github.com/kohya-ss/musubi-tuner | https://github.com/tdrussell/diffusion-pipe
>Flux
Forge: https://github.com/lllyasviel/stable-diffusion-webui-forge/discussions/1050
ComfyUI: https://comfyanonymous.github.io/ComfyUI_examples/flux
DeDistilled: https://huggingface.co/TheYuriLover/flux-dev-de-distill-GGUF/tree/main
Pixelwave: https://civitai.com/models/141592/pixelwave
>Misc
Share Metadata: https://catbox.moe | https://litterbox.catbox.moe/
Img2Prompt: https://huggingface.co/spaces/fancyfeast/joy-caption-alpha-two
Archive: https://rentry.org/sdg-link
Samplers: https://stable-diffusion-art.com/samplers/
Digital Art Software: https://krita.org/en/
Txt2Img Plugin: https://kritaaidiffusion.com/
Collagebaker: https://www.befunky.com/create/collage/
Video Collagebaker: https://kdenlive.org/en/
>Related
>>>/aco/csdg
>>>/b/degen
>>>/c/kdg
>>>/d/ddg
>>>/e/edg
>>>/h/hdg
>>>/trash/slop
>>>/vt/vtai
>Local Sexting
>>>/g/lmg
>Maintain thread quality
https://rentry.org/debo
Discussion of Free and Open-Source Diffusion models.
Last Time: >>103934948
Newfrien Guide: https://rentry.org/ldg-lazy-getting
>UI
Metastable: https://metastable.studio
SwarmUI: https://github.com/mcmonkeyprojects
Forge: https://github.com/lllyasviel/stabl
reForge: https://github.com/Panchovix/stable
ComfyUI: https://github.com/comfyanonymous/C
InvokeAI: https://github.com/invoke-ai/Invoke
>Models, LoRAs, & Upscalers
https://civitai.com
https://tensor.art/
https://openmodeldb.info
>Training
https://github.com/Nerogar/OneTrain
https://github.com/kohya-ss/sd-scri
https://github.com/derrian-distro/L
>HunyuanVideo
Comfy: https://github.com/kijai/ComfyUI-Hu
Windows: https://rentry.org/crhcqq54
Training: https://github.com/kohya-ss/musubi-
>Flux
Forge: https://github.com/lllyasviel/stabl
ComfyUI: https://comfyanonymous.github.io/Co
DeDistilled: https://huggingface.co/TheYuriLover
Pixelwave: https://civitai.com/models/141592/p
>Misc
Share Metadata: https://catbox.moe | https://litterbox.catbox.moe/
Img2Prompt: https://huggingface.co/spaces/fancy
Archive: https://rentry.org/sdg-link
Samplers: https://stable-diffusion-art.com/sa
Digital Art Software: https://krita.org/en/
Txt2Img Plugin: https://kritaaidiffusion.com/
Collagebaker: https://www.befunky.com/create/coll
Video Collagebaker: https://kdenlive.org/en/
>Related
>>>/aco/csdg
>>>/b/degen
>>>/c/kdg
>>>/d/ddg
>>>/e/edg
>>>/h/hdg
>>>/trash/slop
>>>/vt/vtai
>Local Sexting
>>>/g/lmg
>Maintain thread quality
https://rentry.org/debo
Anonymous 01/18/25(Sat)12:29:14 No.103944686
A noob question here, is there a way I could make a LoRA/text embedding/whatever off of a single image? I have a character I genned using Dall-E and want to try making different gens of them using local.
Anonymous 01/18/25(Sat)12:29:40 No.103944696
Blessed thread of frenship
Anonymous 01/18/25(Sat)12:31:59 No.103944718
>>103944686
>training off of a single image
Nothing is stopping you, I would imagine the Lora would be extremely limited or stiff, though. I feel like you need at least a handful of images.
>training off of a single image
Nothing is stopping you, I would imagine the Lora would be extremely limited or stiff, though. I feel like you need at least a handful of images.
Anonymous 01/18/25(Sat)12:32:33 No.103944723
>>103944686
absolutely you can, but the best it can do is recreate that one image you trained it on
absolutely you can, but the best it can do is recreate that one image you trained it on
Anonymous 01/18/25(Sat)12:34:11 No.103944743
>>103944686
>is there a way I could make a LoRA/text embedding/whatever off of a single image?
you could make variants of that image with Flux Redux
https://xcancel.com/bfl_ml/status/1859618500249002375#m
>is there a way I could make a LoRA/text embedding/whatever off of a single image?
you could make variants of that image with Flux Redux
https://xcancel.com/bfl_ml/status/1
Anonymous 01/18/25(Sat)12:36:43 No.103944776
Anonymous 01/18/25(Sat)12:38:23 No.103944792
>>103944718
>>103944723
I'm aware of the severe limitation of using only one image, which is why I'm wondering if there's a different technology/technique that can work off of a single image.
>>103944743
>>103944776
Thanks, I'll check these out!
>>103944723
I'm aware of the severe limitation of using only one image, which is why I'm wondering if there's a different technology/technique that can work off of a single image.
>>103944743
>>103944776
Thanks, I'll check these out!
Anonymous 01/18/25(Sat)12:39:17 No.103944795
I think anon is just nostalgic about old Dall-e. It's not actually good. Neither is its newest version.
Anonymous 01/18/25(Sat)12:39:40 No.103944800
Anonymous 01/18/25(Sat)12:41:23 No.103944819
>>103944792
so you're just jerking me around, asking questions you already know the answer to as a pretext to ask a different question
I don't appreciate that
so you're just jerking me around, asking questions you already know the answer to as a pretext to ask a different question
I don't appreciate that
Anonymous 01/18/25(Sat)12:42:01 No.103944827
How can I make the AI surprise me?
I'm using Forge, and I want some sexy images of a particular character, but I don't want to go into deep detail, I would like to press generate and then get surprised of what the AI comes up with.
How to do that?
I'm using Forge, and I want some sexy images of a particular character, but I don't want to go into deep detail, I would like to press generate and then get surprised of what the AI comes up with.
How to do that?
Anonymous 01/18/25(Sat)12:42:11 No.103944830
>>103944819
Uh, what?
Uh, what?
Anonymous 01/18/25(Sat)12:43:45 No.103944847
Anonymous 01/18/25(Sat)12:48:36 No.103944890
>>103944827
Throw random and conflicting art styles into the prompt. Sometimes the solution riffs on a few of the styles with subtle motifs from the other tokens.
Throw random and conflicting art styles into the prompt. Sometimes the solution riffs on a few of the styles with subtle motifs from the other tokens.
Anonymous 01/18/25(Sat)12:53:03 No.103944949
Anonymous 01/18/25(Sat)13:02:02 No.103945048
>>103944686
I have done this few times. I used IP-adapter (controlnet module) to create few alternative versions of the image I started with. When I had 3-5 decent images I trained lora based on them with Prodigy optimizer. After that it was easy to create even more variants for a proper style lora.
I have done this few times. I used IP-adapter (controlnet module) to create few alternative versions of the image I started with. When I had 3-5 decent images I trained lora based on them with Prodigy optimizer. After that it was easy to create even more variants for a proper style lora.
Anonymous 01/18/25(Sat)13:04:29 No.103945071
>>103945048
Thanks, I'll try and figure out how to do this, never really messed with controlnet before, nor have I heard about Prodigy optimizer until now. Man, I really need to brush up on stuff.
Thanks, I'll try and figure out how to do this, never really messed with controlnet before, nor have I heard about Prodigy optimizer until now. Man, I really need to brush up on stuff.
Anonymous 01/18/25(Sat)13:05:17 No.103945083
Anonymous 01/18/25(Sat)13:07:30 No.103945113
>>103945071
Better make it a longer project then, take your time. Fun thing with loras is that you can always keep the dataset and improve it over time.
Better make it a longer project then, take your time. Fun thing with loras is that you can always keep the dataset and improve it over time.
Anonymous 01/18/25(Sat)13:23:40 No.103945281
>those anime gens
I'm obviously not prompting lighting correctly in Noob. I need to try harder.
Also, how can I do wildcards? I need a bunch of characters but they have to be different.
I'm obviously not prompting lighting correctly in Noob. I need to try harder.
Also, how can I do wildcards? I need a bunch of characters but they have to be different.
Anonymous 01/18/25(Sat)13:24:07 No.103945294
Anonymous 01/18/25(Sat)13:25:16 No.103945311
Yeah man who would make generated AI content of someone IRL, that would be fucking WEIRD.
Anonymous 01/18/25(Sat)13:38:51 No.103945511
Anonymous 01/18/25(Sat)13:40:13 No.103945521
Anonymous 01/18/25(Sat)13:45:32 No.103945575
Anonymous 01/18/25(Sat)13:51:32 No.103945640
Anyone have any experience using TangoFlux?
I keep getting this error. All of my models are in the right place. I'm not sure what's wrong exactly.
>Sizes of tensors must match except in dimension 1. Expected size 15 but got size 645 for tensor number 1 in the list.
I keep getting this error. All of my models are in the right place. I'm not sure what's wrong exactly.
>Sizes of tensors must match except in dimension 1. Expected size 15 but got size 645 for tensor number 1 in the list.
Anonymous 01/18/25(Sat)13:52:01 No.103945645
How would I go about upscaling manga pages using AI? Assuming that's the best way to get a good upscale for really REALLY small pages.
Anonymous 01/18/25(Sat)13:56:04 No.103945689
Anonymous 01/18/25(Sat)13:57:46 No.103945701
>>103945575
SICK!, catbox?
SICK!, catbox?
Anonymous 01/18/25(Sat)14:00:23 No.103945732
>>103944650
i haven't been doing any proompting for months and months so please forgive such a question, but what is the best way to have recurring faces/characters in realistic prompts which are not using a character Lora?
i haven't been doing any proompting for months and months so please forgive such a question, but what is the best way to have recurring faces/characters in realistic prompts which are not using a character Lora?
Anonymous 01/18/25(Sat)14:00:48 No.103945739
Anonymous 01/18/25(Sat)14:03:16 No.103945758
Anonymous 01/18/25(Sat)14:11:44 No.103945849
Anonymous 01/18/25(Sat)14:15:46 No.103945884
>>103945849
composite?
composite?
Anonymous 01/18/25(Sat)14:21:33 No.103945939
>>103945884
Nope. Using noobaiXLNAIXL_vPred10Version and I trained some loras for it for the style. Just testing
Nope. Using noobaiXLNAIXL_vPred10Version and I trained some loras for it for the style. Just testing
Anonymous 01/18/25(Sat)14:23:20 No.103945962
Anonymous 01/18/25(Sat)14:25:47 No.103945991
i2v when
Anonymous 01/18/25(Sat)14:26:15 No.103946002
>>103945939
Looks good, the style works.
Looks good, the style works.
Anonymous 01/18/25(Sat)14:27:12 No.103946012
>>103945689
This is very ugly. I don't understand the appeal.
This is very ugly. I don't understand the appeal.
Anonymous 01/18/25(Sat)14:28:27 No.103946027
So when is Comfy going to release his operating system?
Anonymous 01/18/25(Sat)14:33:00 No.103946084
>>103946027
and will it run on NVIDIA Project DIGITS?
and will it run on NVIDIA Project DIGITS?
Anonymous 01/18/25(Sat)14:34:09 No.103946094
How can I get rid of shadows? I just want completely flat colors.
Anonymous 01/18/25(Sat)14:36:49 No.103946119
>>103946094
put shading in the negative prompt
put shading in the negative prompt
Anonymous 01/18/25(Sat)14:37:28 No.103946127
>>103946094
https://civitai.com/models/1114224/featureless-flat-mix-noob
This isn't a solution but using a model like this might help assuming you're already putting the appropriate tags in the negative.
https://civitai.com/models/1114224/
This isn't a solution but using a model like this might help assuming you're already putting the appropriate tags in the negative.
Anonymous 01/18/25(Sat)14:37:40 No.103946131
>>103945991
Never.
Never.
Anonymous 01/18/25(Sat)14:39:22 No.103946155
>>103946002
Keeps breaking fingers + other small detail.
Keeps breaking fingers + other small detail.
Anonymous 01/18/25(Sat)14:42:44 No.103946194
>>103946155
nice
nice
Anonymous 01/18/25(Sat)14:46:38 No.103946244
any of you come up with a robust way of doing sequential actions in hunyuan? seems to be a roll of the dice
Anonymous 01/18/25(Sat)14:47:09 No.103946255
>>103946094
Minimalism loras?
883572-minimalist-style-illustrious
930648-minimalist-flat-color-style-illustriousxl-lora
Minimalism loras?
883572-minimalist-style-illustrious
930648-minimalist-flat-color-style-
Anonymous 01/18/25(Sat)14:48:37 No.103946270
>>103946255
Trying to get results without loras
Trying to get results without loras
Anonymous 01/18/25(Sat)14:49:23 No.103946280
>>103946270
That's like still very much checkpoint dependent or something. Not really not pointless.
That's like still very much checkpoint dependent or something. Not really not pointless.
Anonymous 01/18/25(Sat)14:51:30 No.103946306
>>103946244
i2v solves this
i2v solves this
Anonymous 01/18/25(Sat)15:03:29 No.103946431
>>103946244
I've had limited luck using timestamps, but if it don't know it then it don't know it
I've had limited luck using timestamps, but if it don't know it then it don't know it
Anonymous 01/18/25(Sat)15:03:53 No.103946433
Anonymous 01/18/25(Sat)15:17:38 No.103946548
>>103946433
that's a clean upscale
that's a clean upscale
Anonymous 01/18/25(Sat)15:19:10 No.103946559
>>103946548
Thanks
Thanks
Anonymous 01/18/25(Sat)15:21:57 No.103946580
Anonymous 01/18/25(Sat)15:28:58 No.103946633
>>103946580
I'm using PAG, SEG, an image quality lora, and the Siax 200k upscaler.
https://files.catbox.moe/32w3n2.png
I despise doing any post-gen work, so I keep messing with stuff to get better pure txt2img output.
I'm using PAG, SEG, an image quality lora, and the Siax 200k upscaler.
https://files.catbox.moe/32w3n2.png
I despise doing any post-gen work, so I keep messing with stuff to get better pure txt2img output.
Anonymous 01/18/25(Sat)15:48:08 No.103946798
Anonymous 01/18/25(Sat)16:48:16 No.103947416
>>103945689
Very well done and very interesting
Very well done and very interesting
Anonymous 01/18/25(Sat)16:53:17 No.103947454
>>103947416
It looks like cheap stock art you would buy to decorate an Airbnb
It looks like cheap stock art you would buy to decorate an Airbnb
Anonymous 01/18/25(Sat)16:54:35 No.103947460
>>103947454
And maybe that was the prompt.
And maybe that was the prompt.
Anonymous 01/18/25(Sat)16:59:05 No.103947506
>>103947416
teehees at you queefingly
teehees at you queefingly
Anonymous 01/18/25(Sat)17:02:37 No.103947533
Anonymous 01/18/25(Sat)17:08:30 No.103947577
Anonymous 01/18/25(Sat)17:22:53 No.103947717
I beg of you anon, do NOT use NTRMix. Whatever you do, DO NOT use it. Under no circumstances.
Anonymous 01/18/25(Sat)17:24:55 No.103947740
Anonymous 01/18/25(Sat)17:25:25 No.103947745
>>103947717
elaborate
elaborate
Anonymous 01/18/25(Sat)17:26:45 No.103947769
>>103947745
It's unironically just shit. He thinks he's smart with his low effort reverse psychology attempt.
It's unironically just shit. He thinks he's smart with his low effort reverse psychology attempt.
Anonymous 01/18/25(Sat)17:28:04 No.103947778
>>103947577
Pretty cheeky
Pretty cheeky
Anonymous 01/18/25(Sat)17:28:22 No.103947781
Anonymous 01/18/25(Sat)17:40:50 No.103947913
Anonymous 01/18/25(Sat)17:42:45 No.103947936
Anonymous 01/18/25(Sat)17:52:05 No.103948043
Anonymous 01/18/25(Sat)17:53:30 No.103948057
Does reForge support Flux GGUF? Or that is a Forge only thing
Anonymous 01/18/25(Sat)17:58:01 No.103948104
>>103948057
forge only
forge only
Anonymous 01/18/25(Sat)18:11:52 No.103948261
Is there a point to using character lora if Forge already knows her really well?
Anonymous 01/18/25(Sat)18:12:38 No.103948271
Anonymous 01/18/25(Sat)18:14:04 No.103948285
Anonymous 01/18/25(Sat)18:15:20 No.103948303
>>103948285
We all start somewhere
We all start somewhere
Anonymous 01/18/25(Sat)18:16:26 No.103948311
Anonymous 01/18/25(Sat)18:16:50 No.103948315
Anonymous 01/18/25(Sat)18:17:10 No.103948321
Anonymous 01/18/25(Sat)18:18:22 No.103948332
>>103947577
prompt?
prompt?
Anonymous 01/18/25(Sat)18:23:21 No.103948386
>>103948332
>A pov shot of a woman sitting across from the viewer at a table inside a dimly lit restaurant. She is sitting at a table wearing a floral dress showing mild cleavage. Her short, wavy, brown hair adds a playful touch to her look. She gives a coy look as she takes a sip from her wine glass. The warm light from the candle on the table in front of her flashes revealing her beautiful face. The framing focuses on her upper body, capturing her delighted expression and the texture of her dress, while the soft lighting adds a cozy, heartwarming feel to the scene. In the background a waiter is seen tending to other diners in the busy restaurant.
>A pov shot of a woman sitting across from the viewer at a table inside a dimly lit restaurant. She is sitting at a table wearing a floral dress showing mild cleavage. Her short, wavy, brown hair adds a playful touch to her look. She gives a coy look as she takes a sip from her wine glass. The warm light from the candle on the table in front of her flashes revealing her beautiful face. The framing focuses on her upper body, capturing her delighted expression and the texture of her dress, while the soft lighting adds a cozy, heartwarming feel to the scene. In the background a waiter is seen tending to other diners in the busy restaurant.
Anonymous 01/18/25(Sat)18:25:11 No.103948401
Any talk of Auto/Forge getting video support? I don't use either I'm just curious
Anonymous 01/18/25(Sat)18:30:27 No.103948437
>>103948303
Start in comfy
Start in comfy
Anonymous 01/18/25(Sat)18:39:17 No.103948527
Anonymous 01/18/25(Sat)18:40:13 No.103948533
Anonymous 01/18/25(Sat)19:04:15 No.103948785
Anonymous 01/18/25(Sat)19:23:02 No.103948962
Anonymous 01/18/25(Sat)19:24:07 No.103948973
>>103945311
I'm thankful to be a vramlet so that I do not fall prey to such temptations videographically speaking. It's really not any "worse" than images but it feels as such in my heart of hearts.
I'm thankful to be a vramlet so that I do not fall prey to such temptations videographically speaking. It's really not any "worse" than images but it feels as such in my heart of hearts.
Anonymous 01/18/25(Sat)19:26:21 No.103948997
>>103948973
I feel the same way when looking at some posted video of a tiktok woman dressed like a disney princess playing with her hair for millionth fucking time, but people call me a faggot
I feel the same way when looking at some posted video of a tiktok woman dressed like a disney princess playing with her hair for millionth fucking time, but people call me a faggot
Anonymous 01/18/25(Sat)19:31:08 No.103949041
I suddenly am reminded of that one British guy who got arrested for having AI videos of pizza. How? Did his clients really get off to 5 seconds of video, or did he have a super video model, several years ahead of its time, capable of more?
Anonymous 01/18/25(Sat)19:32:49 No.103949060
>>103949041
v2v w/ faceswap probably, that's been a thing for a lot longer than local t2v quality
v2v w/ faceswap probably, that's been a thing for a lot longer than local t2v quality
Anonymous 01/18/25(Sat)19:33:50 No.103949073
>>103947577
thanks, it works fine. if you have any other prompts, please feel free. hunyuan is so great
thanks, it works fine. if you have any other prompts, please feel free. hunyuan is so great
Anonymous 01/18/25(Sat)19:38:19 No.103949107
>>103949073
No problem bro, have another:
>A mature older woman with gray hair with a vanilla popsicle in her mouth. She is licking the popsicle. The popsicle is going in and out of her mouth.
No problem bro, have another:
>A mature older woman with gray hair with a vanilla popsicle in her mouth. She is licking the popsicle. The popsicle is going in and out of her mouth.
Anonymous 01/18/25(Sat)19:41:59 No.103949136
>>103949060
Unlikely. If I understand the story correctly, his client could even commission specific scenarios. I don't think the dude trawled the web for that exact scenario in a porn video to faceswap.
Unlikely. If I understand the story correctly, his client could even commission specific scenarios. I don't think the dude trawled the web for that exact scenario in a porn video to faceswap.
Anonymous 01/18/25(Sat)19:42:59 No.103949144
>>103949041
like the other anon said, if it was before 2023 then it was almost certainly a faceswap deepnude using a real child face
>>103949136
>his client could even commission specific scenarios
now i remember this and no, you could commision faceswaps on specific onlyfans whores iirc
like the other anon said, if it was before 2023 then it was almost certainly a faceswap deepnude using a real child face
>>103949136
>his client could even commission specific scenarios
now i remember this and no, you could commision faceswaps on specific onlyfans whores iirc
Anonymous 01/18/25(Sat)19:44:29 No.103949160
>>103949136
then witchcraft, I don't know
then witchcraft, I don't know
Anonymous 01/18/25(Sat)19:46:10 No.103949170
Just a heads up, any time you spend wasting in forge is well... wasted.
You will all eventually need to use comfy and stay there.
You will all eventually need to use comfy and stay there.
Anonymous 01/18/25(Sat)19:46:29 No.103949176
Anonymous 01/18/25(Sat)19:49:06 No.103949196
>>103949170
this is true but only because almost everyone will switch to genning videos eventually and will have no need for inpainting or anything thats easier with webui derivatives
this is true but only because almost everyone will switch to genning videos eventually and will have no need for inpainting or anything thats easier with webui derivatives
Anonymous 01/18/25(Sat)19:49:54 No.103949198
Anonymous 01/18/25(Sat)19:50:25 No.103949201
>>103949176
cool, where do I stick it?
cool, where do I stick it?
Anonymous 01/18/25(Sat)19:51:14 No.103949212
>>103949201
between the thighs like a little Greek boy
between the thighs like a little Greek boy
Anonymous 01/18/25(Sat)19:52:11 No.103949220
>>103949212
it should really have a proper bed or a nesting box
it should really have a proper bed or a nesting box
Anonymous 01/18/25(Sat)20:01:48 No.103949313
>>103949201
Fool, do not be tempted by the eldritch boggart's wiles. You're much better off serving the Monarch of Decay here.
Fool, do not be tempted by the eldritch boggart's wiles. You're much better off serving the Monarch of Decay here.
Anonymous 01/18/25(Sat)20:11:57 No.103949406
>>103949198
It's OBJECTIVELY ugly
It's OBJECTIVELY ugly
Anonymous 01/18/25(Sat)20:16:20 No.103949443
Anonymous 01/18/25(Sat)20:17:26 No.103949451
i had plans to go have drinks tonight but id rather stay in and generate
Anonymous 01/18/25(Sat)20:21:01 No.103949483
>>103948315
Moody
Moody
Anonymous 01/18/25(Sat)20:21:20 No.103949488
>>103944800
Loras/style? Really like this
Loras/style? Really like this
Anonymous 01/18/25(Sat)20:34:08 No.103949614
>>103949488
3d, Suprematism ,Constructivism , data, graph, by Lucien Rudaux gouache ,abstract expressionism, Landsat image , by Daido Moriyama, by Beksinski,
Steps: 5, Sampler: Euler, Schedule type: Simple, CFG scale: 1, Seed: 1964760198, Size: 1024x1024, Model hash: 4038c907c8, Model: flux1-schnell, Denoising strength: 0.67, Version: f2.0.1v1.10.1-previous-636-gb835f24a, Diffusion in Low Bits: float8-e5m2, Module 1: clip_l, Module 2: ae, Module 3: t5xxl_fp16
3d, Suprematism ,Constructivism , data, graph, by Lucien Rudaux gouache ,abstract expressionism, Landsat image , by Daido Moriyama, by Beksinski,
Steps: 5, Sampler: Euler, Schedule type: Simple, CFG scale: 1, Seed: 1964760198, Size: 1024x1024, Model hash: 4038c907c8, Model: flux1-schnell, Denoising strength: 0.67, Version: f2.0.1v1.10.1-previous-636-gb835f24
Anonymous 01/18/25(Sat)20:37:37 No.103949661
>>103949614
Awesome, thanks, that you're using Flux is convenient too
Awesome, thanks, that you're using Flux is convenient too
Anonymous 01/18/25(Sat)20:41:57 No.103949709
Anonymous 01/18/25(Sat)20:42:00 No.103949710
Anonymous 01/18/25(Sat)20:57:02 No.103949863
Anonymous 01/18/25(Sat)20:59:21 No.103949889
Can you bros help me out and inpait that hand off from her head?
Anonymous 01/18/25(Sat)21:00:46 No.103949902
>>103949889
https://github.com/Sanster/IOPaint
https://github.com/Sanster/IOPaint
Anonymous 01/18/25(Sat)21:00:53 No.103949905
>>103949889
yeah
yeah
Anonymous 01/18/25(Sat)21:10:37 No.103949985
>>103948785
Excellent stuff, Mr. Dorf
Excellent stuff, Mr. Dorf
Anonymous 01/18/25(Sat)21:19:33 No.103950071
>>103949902
Thanks
Thanks
Anonymous 01/18/25(Sat)21:34:11 No.103950212
https://xcancel.com/aiwarper/status/1880658326645878821
hunyuan is impressive
hunyuan is impressive
Anonymous 01/18/25(Sat)21:35:30 No.103950222
if someone trained a new model today that met all the following criteria:
- based on the underlying SDXL UNET Architecture
- had the exact same number of parameters as SDXL
- used a single absolute SOTA text encoder (no additional CLIP or anything) while being trained on an exceptionally well-captioned dataset
- used a native 16-channel VAE
- was trained for a wide resolution range, like say 0.5mp up to 2mp
I unironically think this model would be completely competitive with Flux in every practical way that a person could actually see and define IRL based solely on output results.
TLDR I have become based on experience completely convinced over time that both "moar params" and moving from UNET to DiT architecture have literally no practical IRL benefits whatsoever at all, and that 100% of tangible improvements post-SDXL are thanks solely to better VAEs and better text encoders.
If someone can explain like what I'm missing or how I'm wrong in a way that makes sense I'd be happy to hear it though.
- based on the underlying SDXL UNET Architecture
- had the exact same number of parameters as SDXL
- used a single absolute SOTA text encoder (no additional CLIP or anything) while being trained on an exceptionally well-captioned dataset
- used a native 16-channel VAE
- was trained for a wide resolution range, like say 0.5mp up to 2mp
I unironically think this model would be completely competitive with Flux in every practical way that a person could actually see and define IRL based solely on output results.
TLDR I have become based on experience completely convinced over time that both "moar params" and moving from UNET to DiT architecture have literally no practical IRL benefits whatsoever at all, and that 100% of tangible improvements post-SDXL are thanks solely to better VAEs and better text encoders.
If someone can explain like what I'm missing or how I'm wrong in a way that makes sense I'd be happy to hear it though.
Anonymous 01/18/25(Sat)21:37:30 No.103950239
>>103950222
moar parameters is just easier and people will always try the easy path
moar parameters is just easier and people will always try the easy path
Anonymous 01/18/25(Sat)21:38:32 No.103950257
>>103950212
The facial expressions unfortunately never quite translate, do they
The facial expressions unfortunately never quite translate, do they
Anonymous 01/18/25(Sat)21:45:36 No.103950316
Just started playing Made in abyss Binary Star and felt I should try genning some stuff like it since it seems like there's horror frens in the thread.
Recommended artists for eldritch abominations?
Recommended artists for eldritch abominations?
Anonymous 01/18/25(Sat)21:48:56 No.103950334
https://xcancel.com/bfl_ml/status/1879913014578811270#m
>Look goyim, now you can finetune our API model, and the one that makes the best finetune gets a candy... what do you mean you'd want to get the model back, that's not how it works goyim, you make the training and we keep the model and we'll be using it
the worst part is that people will fall for it
>Look goyim, now you can finetune our API model, and the one that makes the best finetune gets a candy... what do you mean you'd want to get the model back, that's not how it works goyim, you make the training and we keep the model and we'll be using it
the worst part is that people will fall for it
Anonymous 01/18/25(Sat)21:51:19 No.103950348
>>103950334
i will never use flux locally out of protest for this
i will never use flux locally out of protest for this
Anonymous 01/18/25(Sat)21:53:11 No.103950360
>>103950334
so they provide the model, the hardware, and you provide the legal liability for the dataset?
sign me up!
so they provide the model, the hardware, and you provide the legal liability for the dataset?
sign me up!
Anonymous 01/18/25(Sat)21:55:24 No.103950378
Anonymous 01/18/25(Sat)21:55:27 No.103950380
Anonymous 01/18/25(Sat)21:56:28 No.103950388
>>103950334
looks like he used flowedit but I always got blurry shit when using it
https://github.com/logtd/ComfyUI-HunyuanLoom/blob/main/example_workflows/example_hy_flowedit.json
looks like he used flowedit but I always got blurry shit when using it
https://github.com/logtd/ComfyUI-Hu
Anonymous 01/18/25(Sat)21:57:16 No.103950394
>BFL receives a finetune
>I receive the police at my door
>I receive the police at my door
Anonymous 01/18/25(Sat)22:05:14 No.103950472
>>103949985
Thanks
Thanks
Anonymous 01/18/25(Sat)22:10:30 No.103950514
https://github.com/chengzeyi/Comfy-WaveSpeed?tab=readme-ov-file#comfy-wavespeed
I hope he'll make it, this is a very important feature imo
I hope he'll make it, this is a very important feature imo
Anonymous 01/18/25(Sat)22:13:54 No.103950546
Couldn't find a better place to ask: there used to be multiple places for genning Magic: The Gathering AI cards, but it seems like all of them have disappeared? Does anyone know what happened, or if any are still working?
Anonymous 01/18/25(Sat)22:16:58 No.103950571
Anonymous 01/18/25(Sat)22:28:06 No.103950661
>>103950546
I vaguely remember that generator. There are loras on civitai for magic the gathering cards but for creating the text and which model was used for the original generator I don't know.
I vaguely remember that generator. There are loras on civitai for magic the gathering cards but for creating the text and which model was used for the original generator I don't know.
Anonymous 01/18/25(Sat)22:32:46 No.103950696
>>103950514
What kind of results are we expecting?
What kind of results are we expecting?
Anonymous 01/18/25(Sat)22:32:58 No.103950697
https://github.com/chengzeyi/Comfy-WaveSpeed/issues/87
for those who want to optimise first block cache and have better quality, this post is interesting
for those who want to optimise first block cache and have better quality, this post is interesting
Anonymous 01/18/25(Sat)22:33:47 No.103950706
>>103950661
Wow that guy looks ugly, what's wrong with his teeth, and does he know his hair looks like a cow?
Wow that guy looks ugly, what's wrong with his teeth, and does he know his hair looks like a cow?
Anonymous 01/18/25(Sat)22:39:01 No.103950736
>>103950661
draenei yes
draenei yes
Anonymous 01/18/25(Sat)22:43:15 No.103950771
Anonymous 01/18/25(Sat)22:50:54 No.103950851
>>103950771
Finally a correct render.
The basic idea is very simple. Does anyone else on here know what a mind palace is?
This is one part of a set of places I use for my own personal checklist.
This is
pebble: people
i-10: itenerary
two dewy rags: todo
camera & tripod: camera gear
Finally a correct render.
The basic idea is very simple. Does anyone else on here know what a mind palace is?
This is one part of a set of places I use for my own personal checklist.
This is
pebble: people
i-10: itenerary
two dewy rags: todo
camera & tripod: camera gear
Anonymous 01/18/25(Sat)22:52:47 No.103950873
>>103950851
Right now I'm putting such things in Anarchy Arcade, because that's the easiest way to make a virtual gallery. Unfortunately, it doesn't really have gilt or modern frames, and it doesn't have a way to set the height precisely or repeatedly.
Right now I'm putting such things in Anarchy Arcade, because that's the easiest way to make a virtual gallery. Unfortunately, it doesn't really have gilt or modern frames, and it doesn't have a way to set the height precisely or repeatedly.
Anonymous 01/18/25(Sat)23:01:20 No.103950945
>>103950697
https://ai-image-journey.blogspot.com/2025/01/wavespeed-quality.html
>Since these settings can be complex, I asked ChatGPT for its recommended values.
wtf, based ChatGPT?
https://ai-image-journey.blogspot.c
>Since these settings can be complex, I asked ChatGPT for its recommended values.
wtf, based ChatGPT?
Anonymous 01/18/25(Sat)23:19:10 No.103951120
When I generate with the same seed in the KSampler, ComfyUI makes a different image. Why doesn't it keep the gen the same?
Anonymous 01/18/25(Sat)23:20:55 No.103951140
Anonymous 01/18/25(Sat)23:21:09 No.103951141
>>103951120
it should give you the same image, can you make a screen of your workflow, something must be wrong with your settings
it should give you the same image, can you make a screen of your workflow, something must be wrong with your settings
Anonymous 01/18/25(Sat)23:22:10 No.103951148
>>103951120
Maybe denoise isn't exactly the same?
Maybe denoise isn't exactly the same?
Anonymous 01/18/25(Sat)23:23:22 No.103951163
>>103951141
>>103951148
Workflow with prompt shamelessly included. Do tell if I've got something retarded going on, it's my first foray.
>>103951148
Workflow with prompt shamelessly included. Do tell if I've got something retarded going on, it's my first foray.
Anonymous 01/18/25(Sat)23:23:53 No.103951167
>>103951140
Lowkey one of the best posters in the general
Lowkey one of the best posters in the general
Anonymous 01/18/25(Sat)23:25:08 No.103951180
>>103951163
>sampler with "sde"
maybe that's the issue, sde gives you additional noise so that the output doesn't converge and has more details (supposedly), but maybe that makes the whole thing non deterministic
>sampler with "sde"
maybe that's the issue, sde gives you additional noise so that the output doesn't converge and has more details (supposedly), but maybe that makes the whole thing non deterministic
Anonymous 01/18/25(Sat)23:29:43 No.103951206
>>103951163
what does those weird brackets mean?
what does those weird brackets mean?
Anonymous 01/18/25(Sat)23:32:20 No.103951237
>>103951206
no idea, but i'm working off of scraps and random prompts I've found and they seem to improve results
no idea, but i'm working off of scraps and random prompts I've found and they seem to improve results
Anonymous 01/18/25(Sat)23:34:24 No.103951259
>>103951206
>>103951237
https://www.aidemos.info/what-are-parentheses-and-brackets-in-stable-diffusion/
>>103951237
https://www.aidemos.info/what-are-p
Anonymous 01/18/25(Sat)23:35:40 No.103951268
Anonymous 01/18/25(Sat)23:37:48 No.103951285
Anonymous 01/18/25(Sat)23:42:20 No.103951319
Yes!
Anonymous 01/18/25(Sat)23:44:01 No.103951331
>>103951319
Yes, finally something we all asked for! Rf inversion fixes!
Yes, finally something we all asked for! Rf inversion fixes!
Anonymous 01/18/25(Sat)23:45:11 No.103951336
Anonymous 01/18/25(Sat)23:47:27 No.103951355
>>103951336
>kijai doesn't have GGUF
>Comfy core doesn't have block swap and Enhance-A-Video
we'll never get the perfect shit innit? :(
>kijai doesn't have GGUF
>Comfy core doesn't have block swap and Enhance-A-Video
we'll never get the perfect shit innit? :(
Anonymous 01/18/25(Sat)23:50:09 No.103951382
>>103951355
Comfy has some kind of auto memory management but it's really fucky for me and freezes my shit up. If he implements proper block swapping I'll move over, for now I'm sticking with Kijai
Comfy has some kind of auto memory management but it's really fucky for me and freezes my shit up. If he implements proper block swapping I'll move over, for now I'm sticking with Kijai
Anonymous 01/18/25(Sat)23:51:45 No.103951396
>>103951382
>Comfy has some kind of auto memory management but it's really fucky for me and freezes my shit up. If he implements proper block swapping I'll move over
the alternative is to use the --reserve-vram flag, it forces the model to put some of its weights to the ram, dunno if that's less efficient than block swap but you should try that out
>Comfy has some kind of auto memory management but it's really fucky for me and freezes my shit up. If he implements proper block swapping I'll move over
the alternative is to use the --reserve-vram flag, it forces the model to put some of its weights to the ram, dunno if that's less efficient than block swap but you should try that out
Anonymous 01/18/25(Sat)23:55:42 No.103951423
>>103951396
I'll check it out
I'll check it out
Anonymous 01/19/25(Sun)00:08:55 No.103951516
>>103950697
When is this guy going to implement parallel GPUs?
When is this guy going to implement parallel GPUs?
Anonymous 01/19/25(Sun)00:10:04 No.103951523
>>103951516
I don't think that's an easy task at all, I mean, the imageGen boom started more than 2 years ago and we still haven't implemented that yet
I don't think that's an easy task at all, I mean, the imageGen boom started more than 2 years ago and we still haven't implemented that yet
Anonymous 01/19/25(Sun)00:13:38 No.103951544
>>103951523
It's possible with hyvid and therefore flux but the people who know how to have it paywalled. You can look at hyvids repo and see they have it too. The question is how do you get it into comfyUI
It's possible with hyvid and therefore flux but the people who know how to have it paywalled. You can look at hyvids repo and see they have it too. The question is how do you get it into comfyUI
Anonymous 01/19/25(Sun)00:17:55 No.103951585
Anonymous 01/19/25(Sun)00:24:09 No.103951635
>>103951585
doesn't olama have this stuff already? Is this just a matter of script kiddies not knowing systems programming?
doesn't olama have this stuff already? Is this just a matter of script kiddies not knowing systems programming?
Anonymous 01/19/25(Sun)00:25:23 No.103951649
>>103951635
image diffusion and LLMs are two totally different beasts nigga
image diffusion and LLMs are two totally different beasts nigga
Anonymous 01/19/25(Sun)00:25:46 No.103951654
>>103951635
>doesn't olama have this stuff already?
llama.cpp have this stuff for almost 2 years now lol
>doesn't olama have this stuff already?
llama.cpp have this stuff for almost 2 years now lol
Anonymous 01/19/25(Sun)00:30:06 No.103951698
>use noob
>gens start looking like Koikatsu
If I wanted Koikatsu, I'd use Koikatsu.
>gens start looking like Koikatsu
If I wanted Koikatsu, I'd use Koikatsu.
Anonymous 01/19/25(Sun)00:30:20 No.103951702
When I try to inpaint an image using ComfyUI + Flux it messes around with the entire image and not just the masked pixels. It isn't super noticeable at first but after 3-4 sets of inpainting there's a pretty noticeable decline in quality. In particular, it gets very grainy.
Am I doing something wrong with my workflow or something? I thought that the way masking worked was that it kept everything else in the image exactly the same. I'm not resizing the image in any way or trying to do any kind of HD fix either.
Am I doing something wrong with my workflow or something? I thought that the way masking worked was that it kept everything else in the image exactly the same. I'm not resizing the image in any way or trying to do any kind of HD fix either.
Anonymous 01/19/25(Sun)00:32:32 No.103951721
Anonymous 01/19/25(Sun)00:33:28 No.103951730
is there any autocomplete extension for comfyui where i can use a booru tags csv?
Anonymous 01/19/25(Sun)00:33:37 No.103951734
Anonymous 01/19/25(Sun)00:34:43 No.103951746
Anonymous 01/19/25(Sun)00:45:31 No.103951818
https://xcancel.com/kimmonismus/status/1880253749911765326#m
I think veo 2 is as much revolutionary than GPT4 was almost 2 years ago
I think veo 2 is as much revolutionary than GPT4 was almost 2 years ago
Anonymous 01/19/25(Sun)00:55:43 No.103951879
>>103951818
buy all the google stocks
buy all the google stocks
Anonymous 01/19/25(Sun)00:56:50 No.103951889
you know they never said January 2025, we all just assumed
Anonymous 01/19/25(Sun)01:00:01 No.103951910
NEVER
IGNOBLE
GENERATIONS
IGNOBLE
GENERATIONS
Anonymous 01/19/25(Sun)01:00:13 No.103951912
>>103951889
there are TWO WEEKS left
there are TWO WEEKS left
Anonymous 01/19/25(Sun)01:01:05 No.103951920
>>103951889
>they never said January 2025
the internet never forgets anon
https://xcancel.com/TXhunyuan/status/1869314683922272365#m
>they never said January 2025
the internet never forgets anon
https://xcancel.com/TXhunyuan/statu
Anonymous 01/19/25(Sun)01:01:18 No.103951921
hello sarr
Anonymous 01/19/25(Sun)01:01:48 No.103951928
>>103951920
Did they say a year? Read the fine print horny gweilo.
Did they say a year? Read the fine print horny gweilo.
Anonymous 01/19/25(Sun)01:01:49 No.103951929
>>103951920
They never said WHICH January
They never said WHICH January
Anonymous 01/19/25(Sun)01:01:50 No.103951930
>>103951920
did I stutter
did I stutter
Anonymous 01/19/25(Sun)01:02:26 No.103951933
>>103951920
they said january, could be january 2026
they said january, could be january 2026
Anonymous 01/19/25(Sun)01:04:16 No.103951944
How do I print videos?
Anonymous 01/19/25(Sun)01:05:17 No.103951951
Where do I buy video paper?
Anonymous 01/19/25(Sun)01:05:59 No.103951954
>>103951951
I don't think they sell them in shops any more
I don't think they sell them in shops any more
Anonymous 01/19/25(Sun)01:08:51 No.103951974
>>103951954
That's not paper, that's ribbon, I want video PAPAER.
That's not paper, that's ribbon, I want video PAPAER.
Anonymous 01/19/25(Sun)01:13:34 No.103951995
Anonymous 01/19/25(Sun)01:15:00 No.103952003
It's amazing how IMPORTANT my gens are.
Anonymous 01/19/25(Sun)01:15:57 No.103952011
>>103952003
dunno if they are important because I can't see them lol
dunno if they are important because I can't see them lol
Anonymous 01/19/25(Sun)01:17:37 No.103952020
Anonymous 01/19/25(Sun)01:19:22 No.103952032
IMPORTANT? no, i say they are EXPORTANT because we need to EXPORT them to /sdg/
Anonymous 01/19/25(Sun)01:19:28 No.103952035
>>103952011
IMPORTANT.
GAME CHANGERS.
Revolutionizing the way humans interact live and interact with the world around them.
IMPORTANT.
GAME CHANGERS.
Revolutionizing the way humans interact live and interact with the world around them.
Anonymous 01/19/25(Sun)01:21:31 No.103952060
>>103952020
1girl garb
1girl garb
Anonymous 01/19/25(Sun)01:22:32 No.103952072
Anonymous 01/19/25(Sun)01:22:51 No.103952073
Meds
Anonymous 01/19/25(Sun)01:23:31 No.103952079
>no Flux gens
based.
based.
Anonymous 01/19/25(Sun)01:27:24 No.103952108
Okay how do you guys get Flux models to work? I've tried installing the model, quantization, text encoder. I get nothing, but errors. What am I missing?
Anonymous 01/19/25(Sun)01:27:28 No.103952109
>>103952079
The only baste I know is asmongold's terminal cancer.
The only baste I know is asmongold's terminal cancer.
Anonymous 01/19/25(Sun)01:28:08 No.103952116
>>103952060
If you can't handle me at my 1girl, then you don't deserve me at my 1other.
If you can't handle me at my 1girl, then you don't deserve me at my 1other.
Anonymous 01/19/25(Sun)01:28:21 No.103952119
Anonymous 01/19/25(Sun)01:34:09 No.103952176
>>103952109
I fear I am 50% wrong. There is 1 other baste: asmondgold's drug resistant HIV
I fear I am 50% wrong. There is 1 other baste: asmondgold's drug resistant HIV
Anonymous 01/19/25(Sun)01:37:03 No.103952200
>>103952119
But i've been seeing some really cool flux loras
But i've been seeing some really cool flux loras
Anonymous 01/19/25(Sun)01:38:40 No.103952214
>>103952032
fucking based
fucking based
Anonymous 01/19/25(Sun)01:40:00 No.103952224
>>103952032
this man is good
this man is good
Anonymous 01/19/25(Sun)01:43:03 No.103952245
So basically Hyvid img2vid MUST come out within the next 3 days or it won't be January. Come next weekend it is the start of the Chinese new year long break which means the weights cannot be released too close to the start of that weekend or they can't put out fires.
It's either this Monday, Tuesday or Wednesday or not this month.
It's either this Monday, Tuesday or Wednesday or not this month.
Anonymous 01/19/25(Sun)01:43:05 No.103952246
Anonymous 01/19/25(Sun)01:45:41 No.103952257
>>103952245
that won't happen, they're still training the model
https://github.com/Tencent/HunyuanVideo/issues/180#issuecomment-2592674405
>We are working hard on developing a text-to-video model and believe it will be completed soon. However, the open-sourcing process involves some steps that are beyond our control, so we cannot estimate the exact timeline. We believe that it will be available to everyone soon.
that won't happen, they're still training the model
https://github.com/Tencent/HunyuanV
>We are working hard on developing a text-to-video model and believe it will be completed soon. However, the open-sourcing process involves some steps that are beyond our control, so we cannot estimate the exact timeline. We believe that it will be available to everyone soon.
Anonymous 01/19/25(Sun)01:45:50 No.103952258
Anonymous 01/19/25(Sun)01:47:20 No.103952272
This very image will change the WORLD. Yeah, I'm more important than thou, 1girl slob.
Anonymous 01/19/25(Sun)01:49:11 No.103952285
>>103952272
Wow. Look at me changing the world.
Wow. Look at me changing the world.
Anonymous 01/19/25(Sun)01:50:23 No.103952292
>>103950212
Has anyone got this to work for them? Tried running the workflow to use a character Lora to replace someone in a video and it definitely did not look as good as the Keanu one, didn’t even look like the character Lora at all actually
Has anyone got this to work for them? Tried running the workflow to use a character Lora to replace someone in a video and it definitely did not look as good as the Keanu one, didn’t even look like the character Lora at all actually
Anonymous 01/19/25(Sun)01:50:29 No.103952294
>my 1girl is da best
woo
woo
Anonymous 01/19/25(Sun)01:52:37 No.103952309
I thought this was the avatar-free thread
Anonymous 01/19/25(Sun)01:53:04 No.103952315
>>103952294
deformed right hand, and you hid the feet (does she have foot diseases that would lower her mate value?
deformed right hand, and you hid the feet (does she have foot diseases that would lower her mate value?
Anonymous 01/19/25(Sun)01:53:19 No.103952318
>>103952272
change the diaper more like cause it's shit. i spit on you
change the diaper more like cause it's shit. i spit on you
Anonymous 01/19/25(Sun)01:53:46 No.103952322
>>103952309
for the avatar-free experience you use filters
for the avatar-free experience you use filters
Anonymous 01/19/25(Sun)01:55:01 No.103952331
Jimmy Carter's ghost has nominated me for a Nobel Peace Prize for this gen.
Anonymous 01/19/25(Sun)01:55:38 No.103952339
Anonymous 01/19/25(Sun)01:56:02 No.103952343
Anonymous 01/19/25(Sun)01:56:32 No.103952349
jimmy farter more like cause it's shit, fuck you
Anonymous 01/19/25(Sun)01:57:27 No.103952356
Anonymous 01/19/25(Sun)01:59:36 No.103952367
>>103952331
my mom said this one was neat
my mom said this one was neat
Anonymous 01/19/25(Sun)02:00:53 No.103952379
>>103952367
bit worried part of her foot fell off.
bit worried part of her foot fell off.
Anonymous 01/19/25(Sun)02:02:45 No.103952389
loli footjobs
Anonymous 01/19/25(Sun)02:02:55 No.103952394
I'm a super sayan image generator now. I'm unstoppable in all the UNIVERSE.
Anonymous 01/19/25(Sun)02:03:19 No.103952396
Anonymous 01/19/25(Sun)02:10:46 No.103952446
>>103952292
https://www.reddit.com/r/StableDiffusion/comments/1i4fkuk/comment/m7viif4/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button
>Flowedit yup.
>It’s a crap shoot honestly. You need to just knob turn and prompt tune until it works for literally every new thing you try.
>My biggest tip is prompting… it’s so sensitive to that (also the Lora is doing a ton of heavy lifting here)
>Basically what you want to do is describe the original video perfectly, then resuse the exact same prompt but change the key things (in this case the man’s description to John wick)
https://www.reddit.com/r/StableDiff
>Flowedit yup.
>It’s a crap shoot honestly. You need to just knob turn and prompt tune until it works for literally every new thing you try.
>My biggest tip is prompting… it’s so sensitive to that (also the Lora is doing a ton of heavy lifting here)
>Basically what you want to do is describe the original video perfectly, then resuse the exact same prompt but change the key things (in this case the man’s description to John wick)
Anonymous 01/19/25(Sun)02:13:19 No.103952460
no longer will i 1girl
i will 2girl
i will 2girl
Anonymous 01/19/25(Sun)02:16:53 No.103952482
Anonymous 01/19/25(Sun)02:19:18 No.103952499
>>103952460
woah, calm down. thats the gateway drug to 2girls, 1boy
woah, calm down. thats the gateway drug to 2girls, 1boy
Anonymous 01/19/25(Sun)02:19:29 No.103952502
dpm_pp_2m_sde is blurry with Flux.
Anonymous 01/19/25(Sun)02:20:29 No.103952507
>>103952499
Even worse, he probably actually has Taylor Swift hearted in Spotify.
Even worse, he probably actually has Taylor Swift hearted in Spotify.
Anonymous 01/19/25(Sun)02:20:34 No.103952508
>>103952502
go for res_multistep, that one makes the images crisp, it's way less blurry than euler that's for sure
go for res_multistep, that one makes the images crisp, it's way less blurry than euler that's for sure
Anonymous 01/19/25(Sun)02:21:19 No.103952514
wait till you losers see the 2girls 1cup gen
Anonymous 01/19/25(Sun)02:30:29 No.103952563
>>103952499
what about 2boys?
what about 2boys?
Anonymous 01/19/25(Sun)02:39:39 No.103952600
>>103952396
that's pretty good
that's pretty good
Anonymous 01/19/25(Sun)02:40:18 No.103952601
Anonymous 01/19/25(Sun)02:41:06 No.103952604
Anonymous 01/19/25(Sun)02:46:58 No.103952631
>>103952601
this is very disturbing
this is very disturbing
Anonymous 01/19/25(Sun)02:49:29 No.103952648
>>103952601
pee is stored in the balls
pee is stored in the balls
Anonymous 01/19/25(Sun)02:58:19 No.103952691
We'll never get the MLLM text encoder right?
https://github.com/Tencent/HunyuanVideo/issues/93
https://github.com/Tencent/HunyuanV
Anonymous 01/19/25(Sun)03:08:21 No.103952738
>>103952482
you've made some disgusting, degenerate lewds of her haven't you
you've made some disgusting, degenerate lewds of her haven't you
Anonymous 01/19/25(Sun)03:09:46 No.103952751
Anonymous 01/19/25(Sun)03:10:37 No.103952758
>>103952751
those are super tame by ai standards
those are super tame by ai standards
Anonymous 01/19/25(Sun)03:12:58 No.103952776
>>103952758
didn't she lick the asshole of some dude in one of her videos though?
didn't she lick the asshole of some dude in one of her videos though?
Anonymous 01/19/25(Sun)03:25:29 No.103952835
>>103952776
are you in highschool or something lol
are you in highschool or something lol
Anonymous 01/19/25(Sun)03:26:51 No.103952841
>>103952835
this is true though, she did lick a man's asshole, you really find that "super tame"? what a weirdo
this is true though, she did lick a man's asshole, you really find that "super tame"? what a weirdo
Anonymous 01/19/25(Sun)03:29:01 No.103952848
Anonymous 01/19/25(Sun)03:30:46 No.103952859
>>103952841
nta but i've genned an anime girl drinking cum using a straw out of the gaping anus of a pikachu, you can do many many things with ai
nta but i've genned an anime girl drinking cum using a straw out of the gaping anus of a pikachu, you can do many many things with ai
Anonymous 01/19/25(Sun)03:32:37 No.103952873
>>103952859
that's quite gay not gonna lie
that's quite gay not gonna lie
Anonymous 01/19/25(Sun)03:32:57 No.103952875
>>103952776
Not that anon and I don't approve of analingus, but I'm pretty sure she did stuff involving extreme choking, slapping, vomitting, etc
Not that anon and I don't approve of analingus, but I'm pretty sure she did stuff involving extreme choking, slapping, vomitting, etc
Anonymous 01/19/25(Sun)03:33:46 No.103952881
>>103952875
yeah, that's exactly it, her videos wasn't tame at all, it was quite hardcore she got really humiliated on those videos
yeah, that's exactly it, her videos wasn't tame at all, it was quite hardcore she got really humiliated on those videos
Anonymous 01/19/25(Sun)03:34:48 No.103952891
>>103952873
yes but it's quite cool what you can do with a bit of inpainting elbow grease
yes but it's quite cool what you can do with a bit of inpainting elbow grease
Anonymous 01/19/25(Sun)03:35:42 No.103952897
>>103952600
Thanks
Thanks
Anonymous 01/19/25(Sun)03:36:49 No.103952902
>>103952875
>>103952881
I'm sorry to tell you guys but that kind of stuff is tame nowadays. I guess you should be glad that you haven't seen some of the shit that gets posted out there.
>>103952881
I'm sorry to tell you guys but that kind of stuff is tame nowadays. I guess you should be glad that you haven't seen some of the shit that gets posted out there.
Anonymous 01/19/25(Sun)03:39:35 No.103952916
https://huggingface.co/ostris/Flex.1-alpha
>a finetune of FLUX.1-schnell
>8 billion parameter
lol, schnell already sucks ass and he decided to make the model even smaller
>a finetune of FLUX.1-schnell
>8 billion parameter
lol, schnell already sucks ass and he decided to make the model even smaller
Anonymous 01/19/25(Sun)03:43:01 No.103952928
Anonymous 01/19/25(Sun)03:49:04 No.103952972
>>103952916
what do you think of the image examples on the model card saar?
what do you think of the image examples on the model card saar?
Anonymous 01/19/25(Sun)04:07:32 No.103953081
Anonymous 01/19/25(Sun)04:08:43 No.103953090
>>103953081
Is that arm his stand or something?
Is that arm his stand or something?
Anonymous 01/19/25(Sun)04:16:01 No.103953135
>>103953090
Wow you don't know what that is?
Wow you don't know what that is?
Anonymous 01/19/25(Sun)04:17:31 No.103953147
>>103952272
Really like this one, great job
Really like this one, great job
Anonymous 01/19/25(Sun)04:20:35 No.103953170
>>103953135
Do you?
Do you?
Anonymous 01/19/25(Sun)04:21:16 No.103953173
>>103953170
oh heheheh yeah, of course.
oh heheheh yeah, of course.
Anonymous 01/19/25(Sun)04:25:05 No.103953199
>>103952916
I wouldn't write this off so easily. Flux Dev de-destill finetunes much better than normal dev, but is still a bit finicky (e.g. negative prompts don't completely work and fry the image a bit). If this model is de-distilled even better, that might more than make up for the lower parameter count and being based on schnell.
I wouldn't write this off so easily. Flux Dev de-destill finetunes much better than normal dev, but is still a bit finicky (e.g. negative prompts don't completely work and fry the image a bit). If this model is de-distilled even better, that might more than make up for the lower parameter count and being based on schnell.
Anonymous 01/19/25(Sun)04:26:56 No.103953211
>he doesn't know
Anonymous 01/19/25(Sun)04:28:14 No.103953222
Anonymous 01/19/25(Sun)04:30:56 No.103953232
Anonymous 01/19/25(Sun)04:32:01 No.103953237
>>103953232
The idea was that flux would handle the hands and fine tuning could put everything else in on top of that.
The idea was that flux would handle the hands and fine tuning could put everything else in on top of that.
Anonymous 01/19/25(Sun)04:33:56 No.103953248
Anonymous 01/19/25(Sun)04:36:50 No.103953269
>>103953237
At 50 steps usually Flux fixes hands.
limbs and hands have one problem together, which is composition. Sometimes there are too many. This is a steps management issue, but we aren't able to direct the steps (we should be able to store and fork the whole wf stepwise, whatever interface provides this is the future).
It's basically a mid-timing step issue.
The other problem is the refiner, for hands. Late stage refining. The problem with hands is they are highly sophisticated. Hands alone are basically subjects.
At 50 steps usually Flux fixes hands.
limbs and hands have one problem together, which is composition. Sometimes there are too many. This is a steps management issue, but we aren't able to direct the steps (we should be able to store and fork the whole wf stepwise, whatever interface provides this is the future).
It's basically a mid-timing step issue.
The other problem is the refiner, for hands. Late stage refining. The problem with hands is they are highly sophisticated. Hands alone are basically subjects.
Anonymous 01/19/25(Sun)04:51:24 No.103953357
>>103952631
the future is human organ slop
the future is human organ slop
Anonymous 01/19/25(Sun)05:01:16 No.103953410
Anonymous 01/19/25(Sun)05:02:49 No.103953421
>>103953410
A Belgian queen.
A Belgian queen.
Anonymous 01/19/25(Sun)05:39:39 No.103953637
Anonymous 01/19/25(Sun)05:58:26 No.103953731
Sell me on flux. Every time i'm trying to use it, the outputs are rigid as fuck, like some sort of corpo art/photography, no matter how hard i try tard wrangling it.
Anonymous 01/19/25(Sun)06:00:06 No.103953750
>>103953731
use IPAdapter with it, it's removing that style rigidity
use IPAdapter with it, it's removing that style rigidity
Anonymous 01/19/25(Sun)06:13:09 No.103953829
>>103953637
nice pose
nice pose
Anonymous 01/19/25(Sun)06:17:27 No.103953857
Comfy OS leak when?
Anonymous 01/19/25(Sun)06:26:23 No.103953933
Anonymous 01/19/25(Sun)06:31:51 No.103953975
Anonymous 01/19/25(Sun)06:33:34 No.103953997
Anonymous 01/19/25(Sun)06:33:59 No.103954000
>>103953829
Likewise. Pretty impressed by v-pred model after all.
Likewise. Pretty impressed by v-pred model after all.
Anonymous 01/19/25(Sun)06:43:59 No.103954065
>>103949863
Did you also make the hastur in the OP? because damn I really like these gens.
Did you also make the hastur in the OP? because damn I really like these gens.
Anonymous 01/19/25(Sun)06:45:19 No.103954081
Anonymous 01/19/25(Sun)06:47:06 No.103954098
>>103953857
Why are they even looking to make an operating system to run python? This sounds so stupid.
Why are they even looking to make an operating system to run python? This sounds so stupid.
Anonymous 01/19/25(Sun)06:49:16 No.103954111
>>103954098
nta and I don't have the full context, but that idea sounds so astoundingly retarded, only a hardcore psycho schizo would actually do it.
nta and I don't have the full context, but that idea sounds so astoundingly retarded, only a hardcore psycho schizo would actually do it.
Anonymous 01/19/25(Sun)07:14:15 No.103954248
Anonymous 01/19/25(Sun)07:20:30 No.103954277
>>103953857
>>103954098
>>103954111
I don't think they are making an actual OS, it was just a poorly chosen metaphor used in a blogpost.
>>103954098
>>103954111
I don't think they are making an actual OS, it was just a poorly chosen metaphor used in a blogpost.
Anonymous 01/19/25(Sun)07:26:43 No.103954314
Anonymous 01/19/25(Sun)07:34:36 No.103954369
https://www.reddit.com/r/StableDiffusion/comments/1i4xa5k/clipinterrogator_get_a_prompt_for_an_image_fully/
wtf is he doing?
wtf is he doing?
Anonymous 01/19/25(Sun)07:41:39 No.103954406
>>103951259
Holy shit there's a less emphasis command what a godsend.
Holy shit there's a less emphasis command what a godsend.
Anonymous 01/19/25(Sun)07:44:26 No.103954428
>>103952916
Seems very interesting to me, if it works.
Seems very interesting to me, if it works.
Anonymous 01/19/25(Sun)07:50:29 No.103954457
>>103954406
I mean, if I want to put less emphasis I go for 0.X like (1girl:0.5)
I mean, if I want to put less emphasis I go for 0.X like (1girl:0.5)
Anonymous 01/19/25(Sun)08:03:27 No.103954531
What a feelin', being's believin'
I can have it all, now I'm dancing for my life
Take your passion and make it happen
Pictures come alive, you can dance right through your life
I can have it all, now I'm dancing for my life
Take your passion and make it happen
Pictures come alive, you can dance right through your life
Anonymous 01/19/25(Sun)08:41:48 No.103954759
So, you guys can't make these little tik tok videos anymore. What are you going to do?
Anonymous 01/19/25(Sun)08:42:49 No.103954768
>>103954759
trump will bring it back, trust the plan ;-;
trump will bring it back, trust the plan ;-;
Anonymous 01/19/25(Sun)08:49:25 No.103954830
Noob here, why does A1111 take literally 5x as long to do anything as Comfy? I know it must be using more resources, but how is it that much slower? A 30 second job on comfy is like 2 minutes+. I have a shitty setup (3070 laptop gpu with 8GB) I know, but damn.
Anonymous 01/19/25(Sun)08:53:26 No.103954861
>>103954830
>why does A1111 take literally 5x as long to do anything as Comfy?
because A1111 is a deprecated software
>why does A1111 take literally 5x as long to do anything as Comfy?
because A1111 is a deprecated software
Anonymous 01/19/25(Sun)08:54:38 No.103954866
>>103954830
different settings/different things happening behind the scenes OR you hit VRAM limits and the nvidia driver is swapping out from VRAM to system RAM which is massively slower
different settings/different things happening behind the scenes OR you hit VRAM limits and the nvidia driver is swapping out from VRAM to system RAM which is massively slower
Anonymous 01/19/25(Sun)08:57:48 No.103954892
>>103954759
we can still make the videos? not sure if anyone here published anything on tiktok, the other boards here are fine
we can still make the videos? not sure if anyone here published anything on tiktok, the other boards here are fine
Anonymous 01/19/25(Sun)08:59:21 No.103954898
None of my gens has ever made me jump... Until I laid my eyes on picrel
Anonymous 01/19/25(Sun)09:06:03 No.103954948
Anonymous 01/19/25(Sun)09:06:53 No.103954956
Anonymous 01/19/25(Sun)09:16:41 No.103955025
>>103954768
so far trump's plan has been... releasing a memecoin??
lmao and he is the best option, we are FUCKED
so far trump's plan has been... releasing a memecoin??
lmao and he is the best option, we are FUCKED
Anonymous 01/19/25(Sun)09:19:35 No.103955044
>>103955025
Is 4chan banned in China?
Is 4chan banned in China?
Anonymous 01/19/25(Sun)09:29:33 No.103955123
Any ideas on when any UIs will support Sana finetunes?
Anonymous 01/19/25(Sun)10:05:33 No.103955426
>>103955044
What do you think?
What do you think?
Anonymous 01/19/25(Sun)10:13:38 No.103955493
While the explosive growth of this new technology has enabled rapid deployment of powerful models in many industries, the environmental consequences of this generative AI “gold rush” remain difficult to pin down, let alone mitigate.
Anonymous 01/19/25(Sun)10:15:08 No.103955507
>>103955493
Than your for your input, ChatGPT.
Than your for your input, ChatGPT.
Anonymous 01/19/25(Sun)10:20:07 No.103955562
It smells tranny in here
Anonymous 01/19/25(Sun)10:21:09 No.103955575
>>103955562
Take a bath.
Take a bath.
Anonymous 01/19/25(Sun)10:32:44 No.103955714
>>103953750
absolute cope lol
absolute cope lol
Anonymous 01/19/25(Sun)10:34:01 No.103955727
Style rigidity is ruining my day.
Anonymous 01/19/25(Sun)10:46:42 No.103955841
Bakermen
Anonymous 01/19/25(Sun)10:54:09 No.103955892
for a gpu poor fellow(12gb) what is the best second thing to flux for realistic waifus?
how is juggxl?
how is juggxl?
Anonymous 01/19/25(Sun)11:01:18 No.103955940
baking hol up yo
Anonymous 01/19/25(Sun)11:18:27 No.103956090
Anonymous 01/19/25(Sun)11:26:35 No.103956161
>>103955892
the realistic sdxl/pony/illustrious and sd finetunes should be quite fine
or you just wait minutes for your realistic flux RAM saving technique waifus
the realistic sdxl/pony/illustrious and sd finetunes should be quite fine
or you just wait minutes for your realistic flux RAM saving technique waifus
Anonymous 01/19/25(Sun)11:34:27 No.103956238
>>103956161
i would if i had more ram
i have 16 so while i MIGHT just be able to run it, i'd expend a lot of time not able to do shit on my computer...
please comfychad solve the multigpu conundrum...
i can buy like 3-4 3060s for the price of a 3090
i would if i had more ram
i have 16 so while i MIGHT just be able to run it, i'd expend a lot of time not able to do shit on my computer...
please comfychad solve the multigpu conundrum...
i can buy like 3-4 3060s for the price of a 3090
Anonymous 01/19/25(Sun)11:35:58 No.103956257
>>103956238
Get a night shift job at UPS sorting center. Save $$$$. Buy a 5090. Simple
Get a night shift job at UPS sorting center. Save $$$$. Buy a 5090. Simple
Anonymous 01/19/25(Sun)11:39:49 No.103956295
Anonymous 01/19/25(Sun)11:42:59 No.103956324
>>103956257
not for non americans
not for non americans
Anonymous 01/19/25(Sun)11:45:18 No.103956346
>>103956295
Can I put in my big Mac?
Can I put in my big Mac?
Anonymous 01/19/25(Sun)11:46:20 No.103956354
>>103956324
Learn Spanish. Fake it.
Learn Spanish. Fake it.
Anonymous 01/19/25(Sun)11:48:08 No.103956370
Are there any new hand bbox models? The ones I have are for yolo v9 and we're up to yolo v11 now.
Anonymous 01/19/25(Sun)11:51:00 No.103956396
how's noobai anatomy compared with pony?
Anonymous 01/19/25(Sun)11:59:26 No.103956457
>no mentions of LTX in the thread
What are people using for local Img2Vid nowdays?
What are people using for local Img2Vid nowdays?
Anonymous 01/19/25(Sun)11:59:37 No.103956459
How come none of these models can gen a statue? I prompt a statue and I get a character with the properties of the statue next to a random, unrelated statue, if there's a statue at all.
Anonymous 01/19/25(Sun)11:59:52 No.103956462
>>103954948
noobai vpred1.0, no lora, murata range in prompt.
noobai vpred1.0, no lora, murata range in prompt.
Anonymous 01/19/25(Sun)12:03:16 No.103956492
>>103956396
Better imho
Better imho
Anonymous 01/19/25(Sun)12:05:16 No.103956511
>>103956492
woah seriously?
anatomy was all pony really had going for it, i guess it's old and busted now uh?
woah seriously?
anatomy was all pony really had going for it, i guess it's old and busted now uh?
Anonymous 01/19/25(Sun)12:06:40 No.103956521
Divamedia from World Flipper, trained on a single picture (noobai).
Anonymous 01/19/25(Sun)12:07:22 No.103956530
Anonymous 01/19/25(Sun)12:07:41 No.103956534
>>103956396
Terrible, keeps drawing dicks in pussy when unprompted. It can't tell apart girl's crotch from something being inserted in it.
Terrible, keeps drawing dicks in pussy when unprompted. It can't tell apart girl's crotch from something being inserted in it.
Anonymous 01/19/25(Sun)12:07:48 No.103956535
Anonymous 01/19/25(Sun)12:08:58 No.103956544
>>103956521
I mean, it was augmented into 16 tho...
I should try to get rid of the flat style with regclasses and maybe train on vpred while genning at the regular noobai, it gives the best details in my experience.
I mean, it was augmented into 16 tho...
I should try to get rid of the flat style with regclasses and maybe train on vpred while genning at the regular noobai, it gives the best details in my experience.
Anonymous 01/19/25(Sun)12:09:54 No.103956558
>>103956535
I should've been more specific. I meant the anime models. Pony and Illustrious. I haven't tried yet in Noob.
I should've been more specific. I meant the anime models. Pony and Illustrious. I haven't tried yet in Noob.
Anonymous 01/19/25(Sun)12:11:11 No.103956568
>>103956558
figurines? I ain't never seen a statue of a japanese cartoon character
figurines? I ain't never seen a statue of a japanese cartoon character
Anonymous 01/19/25(Sun)12:11:19 No.103956570
>>103956459
I know that feel, I've been trying to get a fountain in the shape of a pissing girl since SD1.
I know that feel, I've been trying to get a fountain in the shape of a pissing girl since SD1.
Anonymous 01/19/25(Sun)12:14:09 No.103956601
Anonymous 01/19/25(Sun)12:14:55 No.103956605
>>103956090
Mask of the Betrayer's Wall of the Faithless?
Mask of the Betrayer's Wall of the Faithless?
Anonymous 01/19/25(Sun)12:42:30 No.103956914
>>103956544
Uhh... parameters? I haven't tried to train a lora yet but I would like to know more for the future.
Uhh... parameters? I haven't tried to train a lora yet but I would like to know more for the future.
Anonymous 01/19/25(Sun)12:45:53 No.103956963
Who's baking?
Anonymous 01/19/25(Sun)12:53:00 No.103957063
>>103956914
Params aren't that important, the augmentation is done by hand, but I certainly enabled the flip and color augs.
Other than that it was Prodigy with d_coef 1.5, 3 repeats, grad accumulation 2, 10 epochs with scale weight norms at 1.5.
Params aren't that important, the augmentation is done by hand, but I certainly enabled the flip and color augs.
Other than that it was Prodigy with d_coef 1.5, 3 repeats, grad accumulation 2, 10 epochs with scale weight norms at 1.5.
Anonymous 01/19/25(Sun)12:54:01 No.103957073
>>103956963
me, I'll bake
me, I'll bake
Anonymous 01/19/25(Sun)12:56:28 No.103957106
>>103957063
Thanks, saving this.
Thanks, saving this.