Tried training an ACEStep1.5 LoRA for my favorite anime. I didn't expect it to be this good!

https://redd.it/1qwrhp4
@rStableDiffusion

The real "trick" to simple image merging on Klein: just use a prompt that actually has a sufficient level of detail to make it clear what you want
https://redd.it/1qwrqph
@rStableDiffusion

Anima is the new illustrious!!? 2.0!

i've been using illustrous/noobai for a long time and arguably its the best for anime so far. like qwen is great for image change but it doesnt recognize famous characters. So after pony disastrous v7 launch, the only options where noobai. which is good especially if you know danbooru tags, but my god its hell trying to make a multiple character complex image (even with krita).
Until yesterday, i tried this thing called anima (this is not a advertisement of the model, you are free to tell me your opinions on it or would love to know if im wrong). so anima is a mixture of danbooru and natural language. FINALLY FIXING THE BIGGEST PROBLEM OF SDXL MODELS. no doubt its not magic, for now its just preview model which im guessing is the base one. its not compatible with any pony/illustrous/noobai loras cause its structure is different. but with my testing so far, it is better than artist style like noobai. but noobai still wins cause of its character accuracy due to its sheer loras amount.

https://redd.it/1qwukjs
@rStableDiffusion

Most are propably using the wrong AceStep model for their use case

https://redd.it/1qwvuiu
@rStableDiffusion

I obtained these images by training DORA on Flux 1 Dev. The advantage is that it made each person's face look different. Perhaps it would be a good idea for people to try training DORA on the newer models.

https://redd.it/1qx1rr0
@rStableDiffusion

circlestone-labs/Anima · Hugging Face
https://huggingface.co/circlestone-labs/Anima

https://redd.it/1qxawbk
@rStableDiffusion

Is CivitAI slop now?

Now I could just be looking in the wrong places sometimes the real best models and loras are obscure, but it seems to me 99% of CivitAI is complete slop now, just poor quality loras to add more boobs with plasticy skin textures that look lowkey worse than old sdxl finetunes I mean I was so amazed when like I found juggertnautXL, RealvisXL, or something, or even PixelWave to mention a slightly more modern one that was the first full fine tune of FLUX.1 [dev\] and it was pretty great, but nobody seems to really make big impressive fine-tunes anymore that actually change the model significantly

Am I misinformed? I would love it if I was and there are actually really good ones for models that aren't SDXL or Flux

https://redd.it/1qx8y38
@rStableDiffusion

Seedance 2.0 (teaser) better than Sora 2! True multimodal video creation (text + images + video + audio) and seriously controllable outputs.

https://redd.it/1qxdtjb
@rStableDiffusion

Ace-Step 1.5 + LTX2 + ZIB - The Spanish is good?

https://redd.it/1qxdwo7
@rStableDiffusion

What happens if you overwrite an image model with its own output?

https://redd.it/1qxe8b1
@rStableDiffusion

LTX-2 I was going to trim it but this made me Lmfao, Anyone found a rock solid way to reduce blur in faster motion?

https://redd.it/1qxez7u
@rStableDiffusion

Is LTX2 good? is it bad? what if its both!? LTX2 meme

https://redd.it/1qxfc9h
@rStableDiffusion

AceStep1.5 Local Training and Inference Tool Released.

https://redd.it/1qxeqyh
@rStableDiffusion

most effective ways to earn money using ComfyUI right now?

What are the most effective ways to earn money using ComfyUI right now? I’m interested in how people are actually monetizing it—client work, content creation, selling workflows, automation, or something else. If you’ve had real results, I’d love to hear what’s working for you.

https://redd.it/1qxmhf6
@rStableDiffusion

SwarmUI 0.9.8 Release

https://preview.redd.it/rfmgtb22jwhg1.png?width=2016&format=png&auto=webp&s=f8aac5ffb981c15f9d21d092c2d976f4cb16f075


In following of my promise in the SwarmUI 0.9.7 Release notes, the schedule continues to follow the fibonnaci sequence, and it has been 6 months since that release that I'm now posting the next one. I feel it is worth noting that these release versions are arbitrary and not actually meaningful to when updates come out, updates come out instantly, I just like summing up periods of development in big posts every once in a while.

# If You're New Here

If you're not familiar with Swarm - it's an image/video generation UI. It's a thing you install that lets you run flux klein or ltx-2 or wan or whatever ai generator you want.

https://preview.redd.it/0ggaa84cfwhg1.png?width=1080&format=png&auto=webp&s=ad4c999c0f9d043d9b0963ed8c9bb5087c06205e

It's free, local, open source, smart, and a bunch of other nice adjectives. You can check it out on GitHub https://github.com/mcmonkeyprojects/SwarmUI or the nice lil webpage https://swarmui.net/

Swarm is a carefully crafted user-friendly yet still powerful frontend, that uses ComfyUI's full power as its backend (including letting you customize workflows when you want, you literally get an entire unrestricted comfy install as part of your swarm install).

Basically, if you're generating AI images or video on your computer, and you're not using Swarm yet, you should give Swarm a try, I can just about guarantee you'll like it.

# Model Support

https://preview.redd.it/usr6sqf2kwhg1.png?width=2018&format=png&auto=webp&s=21b5e01a634b5e6b23c7fef5d0b3926595c41c16

New models get released all the time. SwarmUI proudly adds day-1 support whenever comfy does. It's been 6 months since the last big update post, so, uh, a lot of those have came out! Here's some models Swarm supported immediately on release:
\- Flux.2 Dev, the giant boi (both image gen and very easy to use image editing)
\- Flux.2 Klein 4B and 9B, the reasonably sized but still pretty cool bois (same as above)
\- Z-Image, Turbo and then also Base
\- The different variants of Qwen Edit plus and 2511/2512/whatever
\- Hunyuan Image 2.1 (remember that?)
\- Hunyuan Video 1.5 (not every release gets a lot of community love, but Swarm still adds them)
\- LTX-2 (audio/video generation fully supported)
\- Anima
\- Probably other ones honestly listen it's been a long time, whatever came out we added support when it did, yknow?

# Beyond Just Image

https://preview.redd.it/8om7crv5iwhg1.png?width=1428&format=png&auto=webp&s=c84eb77c7b6ca3d4be659fb98c111761f7cad1ef

Prior versions of SwarmUI were very focused on image generation. Video generation was supported too (all the way back since when SVD, Stable Video Diffusion, came out. Ancient history, wild right?) but always felt a bit hacked on. A few months ago, Video became a full first-class citizen of SwarmUI. Audio is decently supported too, still some work to do - by the time of the next release, audio-only models (ace step, TTS, etc.) will be well supported (currently ace step impl works but it's a little janky tbh).

I would like to expand a moment on why and how Swarm is such a nice user-friendly frontend, using the screenshot of a video in the UI as an example.

Most software you'll find and use out there in the AI space, is gonna be slapped together from common components. You'll get a basic HTML video object, or maybe a gradio version of one, or maybe a real sparklesparkle fancy option with use react.

Swarm is built from the ground up with care in every step. That video player UI? Yeah, that's custom. Why is it custom? Well to be honest because the vanilla html video UI is janky af in most browsers and also different between browsers and just kinda a pain to work with. BUT also, look at how the colored slidebars use the theme color (in my case I have a purple-emphasis theme

selected), the fonts and formats fit in with the overall UI, etc. The audio slider remembers what you selected previously when you open new videos to keep your volume consistent, and there's a setting in the user tab to configure audio handling behavior. This is just a small piece, not very important, but I put time and care into making sure it feels and looks very smooth.

# User Accounts

In prior release posts, this was a basic and semi-stable system. Now, user accounts are pretty detailed and capable! I'm aware of several publicly hosted SwarmUI instances that have users accessing from different accounts. The system even supports OAuth and user self-registration and etc.

If you're a bigbig user, there's also a dedicated new "Auto Scaling Backend", so if you've got a big cluster of servers you can run swarm across that cluster without annoying your coworkers by idling backends that aren't in use all the time. It spins up and down across your cluster. If you're not THAT big, you can also probably get it to work with that runpod cluster thing too.

# Split Workspaces

If you're not someone looking to share your swarm instance with others, user accounts are actually still super useful to enable - each user account instead becomes a separate workspace for yourself, with separated gen history and presets and etc. Simply use the "impersonate user" button from your local admin account to quickly swap to a different account.

You can for example have a "Spicy" user and a "Safe" user, where "Safe" has a ModelBlacklist set on your "ChilliPeppers/" model folder. Or whatever you're trying to separate, I don't judge.

# AMD Cares About Consumers?!

AMD has spent a while now pushing hard on ROCm drivers for Windows, and those are finally available to the public in initial form! This means if you have a recent AMD card, and up to date drivers, Swarm can now just autoinstall and work flawlessly. Previously we did some jank with DirectML and said if you can't handle the jank try wsl or dualboot to Linux... now life is a bit less painful. Their drivers are still in early preview status though, and don't support all AMD cards yet, so give it some time.

# Extensions

Extension system upgrades have been a hot topic, making them a lot more powerful. The details are technical, but basically extensions are now managed a lot more properly by the system, and also they are capable of doing a heckuva lot more than they could before.

There's been some fun extensions recently too, The SeedVR Extension has been super popular. The inventor of php wrote it (what?! lmao) and basically you click to enable the param and a really powerful upscaler model (seedvr) upscales your image or video as well as or even better than all the clever upscale/refine workflows could, without any thought. Also people have been doing crazy things wild MagicPrompt (the LLM reprompting extension) in the Swarm discord.

# What Do You Mean 6 Months Since Last Release Build

Oh yeah also like a trillion other new things added because in fact I have been actively developing Swarm the entire time, and we've gotten more PRs from more community contributors than ever. This post is just the highlights. There's a slightly more detailed list on the github release notes linked below. There have been almost 600 github commits between then and now, so good luck if you want the very detailed version, heh.

\-----

View the full GitHub release notes here https://github.com/mcmonkeyprojects/SwarmUI/releases/tag/0.9.8-Beta also feel free to chat with me and other swarm users on the Discord https://discord.gg/q2y38cqjNw ps swarm is and will be free forever but you can donate if you want to support https://www.patreon.com/swarmui the patreon is new

https://redd.it/1qxmzo0
@rStableDiffusion

Deni Avdija in Space Jam with LTX-2 I2V + iCloRA. Flow included

https://redd.it/1qxojjm
@rStableDiffusion

Improved Wan 2.2 SVI Pro with LoRa v.2.1

https://redd.it/1qxndw2
@rStableDiffusion

LTX-2 - pushed to the limit on my machine

https://redd.it/1qxrm4r
@rStableDiffusion