FLUX Klein 9B Pixel Space - ComfyUI Nodes

https://redd.it/1tdqv7y
@rStableDiffusion

The issue of repetitive compositions in ANIMA.

Is anyone else having this issue? Every time I enter a prompt, the composition ends up being almost identical. It lacks the randomness you get in illustrious or NAI.

Anyone know a good way to improve this?

https://preview.redd.it/t790dskfna1h1.png?width=590&format=png&auto=webp&s=1de07356f73d4615f3cdfd00a3a8072840378209

https://preview.redd.it/bf8oyjxzma1h1.png?width=603&format=png&auto=webp&s=3b16a80daa72d4705c6b7e42cca5c928267aa57e

https://redd.it/1tdu9s9
@rStableDiffusion

It appears that Microsoft uploaded an image model on HuggingFace and then deleted it.
https://redd.it/1tdxf4t
@rStableDiffusion

Beyond Belief Fact or Fiction?

https://redd.it/1tdxnb8
@rStableDiffusion

LTX-2.3 LipDub test: Dwight reads the changelog

https://redd.it/1te0sas
@rStableDiffusion

I guess this happened a Week after Riker Rick Rolled the ship. With a Special Ending. lol.

https://redd.it/1te17ky
@rStableDiffusion

Anima is in process of being added to diffusers

https://github.com/huggingface/diffusers/pull/13732

Hopefully support on major trainers like OneTrainer is coming after this.

With all the respect to diffusion-pipe its bucketing is a headscratcher and I don't really trust all standalone trainers based on kohya-SS after issues reported and do not want a stack of those.

https://redd.it/1te2d2i
@rStableDiffusion

Microsoft lens is less than 4B params. The tendency is less params...

Ok, they have retired it. It was 3.8B IIRC. In any case, it seems there´s this tendency to do smaller and smaller models but they manage to get better and better anyhow.

My 12GB card loves it. Lets keep the good work

https://redd.it/1te4ieu
@rStableDiffusion

2K ANIMA image
https://redd.it/1te4tj7
@rStableDiffusion

Pixal3D: Generate high-fidelity 3D assets from a single image. (TencentARC, locally runnable model)

https://huggingface.co/TencentARC/Pixal3D

"Pixal3D generates high-fidelity 3D assets from a single image. Unlike previous methods that loosely inject image features via attention, Pixal3D explicitly lifts pixel features into 3D through back-projection, establishing direct pixel-to-3D correspondences. This enables near-reconstruction-level fidelity with detailed geometry and PBR textures."

Looks like no one mentioned this in the sub, so here's everyone's notification.

Some fast points:

* It's a locally runnable model

* I got it working on an RTX 5090 by yelling "Fix it!" at Claude over and over like Philip J. Fry. (This works on most models by the way, I suggest you try it if you have Claude and want to try local models before Comfy's team gets around to it)

* To my eyes, this looks like a step up from Trellis.2 raw, but don't take my word on that. It has some online demo, give it a go.

Please note that it did take a good amount of time getting creative with the yelling-at-claude part, with me having to make some judgment calls and give it advice about how to proceed. But tenacity paid off for me, and I figure it will pay off for anyone else who cares to put in the effort, at least until someone makes a more broadly available guide.

https://redd.it/1te93yi
@rStableDiffusion

stable-diffusion-webui-codex v0.3.0-beta is live (now with link 😅)
https://redd.it/1te4zvv
@rStableDiffusion

Playing with Anima Base 1.0 + Flux.2 Klein 9b + Wan 2.2 (No Audio)

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

TOOL: "AI Master Studio" // Organizer for AI prompts

https://redd.it/1teds9e
@rStableDiffusion